311: The Fab Floor

OK, apparently I need to use this cold open to defend my honor. Oh, no, you're listening to the other podcast, aren't you? From the vicious slander that apparently has been aired on a competing Linux podcast. Wait, what did Adam say? I think it was more you. What? What did I do?

The refined is not a bootloader conversation came up. Oh yeah. It doesn't matter. And it does matter. Nope. Doesn't matter. Here we are to talk about why it matters. You brought this on yourself. Here we go. Technically correct. No, it's not. It is. Listen. Okay.

All right. This is this is diluted in the diaries material potentially. So, OK, I will keep it short. I will just say, though, if you are getting to the point where you are manually configuring your boot process, which you are clearly flirting with now. Oh, no, you haven't listened to this week's episode yet.

Oh, OK. Did you abandon the? No, I did it. It's done. OK, well, then that's the only point I would make is that it's important to know what each piece is doing and what is happening at each stage of the boot process. If you are going to manually configure things, you don't know what.

The reason I say Refind is not a bootloader is because you still need a bootloader. If you have Refind, you still need a bootloader downstream of Refind. Yeah, it's a bootloader loader. Yeah. Okay. He calls it a boot manager. I read I spent a lot of time on Rod's books. I learned about it all. Dude, you've been on Rod's books. Now we can speak a common language. I you're going to you're probably going to yell at your radio again this week because I explained to Adam how a UEFI.

boots into the os to explain a problem he's been having and i know and i prefaced it by saying this is the gist not an expert uh-huh I know you're going to be yelling at your radio. That's funny. I spent a lot of time on Rod's books and I have a fully functional, uh, great refine.com with the manually configured, uh, bootload. Okay. Now we can, now we can speak the same language. I don't know about that, but.

I still say bootloader loader. That's fine. I think as long as you're differentiating. But I mean, again, though, the real point there is if you're actually going to manually write a refine dot conf as you are now doing. Yeah. You still have to point refined at a bootloader. That's true.

Refine can't run an operating system kernel. It has to boot a bootloader that then launches an operating system kernel. Unless, you may have noticed, I don't know if you've gotten this far yet, because this is how I'm doing my ButterFS sub-volume distro setup.

Refind can manually start the Linux kernel itself. That's what I'm doing. And you might be asking, well, where is the bootloader in that situation? Is it refined? It's the Linux kernel. Oh, right. Because it can boot itself if it speaks to the, yeah. I think that's fucking crazy and rad. I think they call it the EFI stub loader, and it has to be compiled into the kernel. It's possible that you have a kernel that does not support that if they didn't compile it that way, but yes, the kernel...

The kernel can be its own bootloader and that ties into this whole move toward, I think they refer to it as unified kernel images. Yeah. Now, which is another thing that's coming about. I feel like alongside like immutable distros as a. Way of saying like, hey, here is a whole and complete Linux OS that can't be fucked up that like can't not ever boot. Like if we just start using the kernel as its own bootloader, then it probably fixes some.

reliability problems with the boot process it solves some problems now the the real irony of this whole thing is that when i was setting up linux the first time i chose refined because it was billed as the one that would just auto configure itself it would look at all the bfi boot partitions and make you a list of them with a graphical interface that looked halfway decent and scaled right

on laptops with high resolution screens sure and so i picked that one because it didn't require manual configuration and then the thing that i ended up having to create the manual configuration for it for creating snapper snapshots an interface for booting into an older snapper snapshot uh turns out if i just used grub that would have been supported automatically because it's the default so

You know, I kind of was hoisted by my own Picard here. Yeah. Yep. Yep. I know the feeling. I know the feeling.

Welcome to Brad and Will made a tech pod. I'm Will. I'm Brad. Yeah, the cards. We have a banger of an episode for you here this week, folks. And Linux is not really mentioned even once as far as I remember. So you're in luck. You mentioned it when we talked about thread directing. Oh, well, barely. Yeah. But.

A couple of weeks ago, I got to go down to Chandler, Arizona to Intel's Fab 53, where the new 18A process is being made and learn all about the first chip to roll out with that, which is called Panther Lake. And I went there with my friend and colleague from PC World, Adam Patrick Murray, to talk about, to learn about this stuff. So Adam, we managed to get Adam in here for a return visit on the show. I think this is.

I don't know, nth visit to the tech pod. Welcome to the show, Adam Patrick Murray. Let's talk about Intel's Panther Lake. I'm back. Doing weird voices today, I guess. Glad to get you guys on a podcast together finally. Yeah, Will, it's great meeting you, Brad. Long time no see. I heard you needed a Linux expert, so I'm glad you brought me on. Glad you workshopped that one a few minutes ago. Compiled any good journals lately?

Oh, yeah. I'm doing it left and right. That's just what I do. It's my thing. Look, Adam's favorite thing in the world is installing new Linux distros, it turns out. He just can't stop. I literally did it twice today already. Great. What do you got on deck?

I need to know. Ubuntu. Yeah. All right. Also, we have an upcoming video on PC World's YouTube channel about the B-Box, and I was reminded that Haiku... os is out there and i was like you know what maybe i should try that one as well yeah dude forget forget the linux podcasts when are we going to get the haiku podcast that's season two man season two of the dual boot diary is haiku and linux you know trust me trust me i had strong feelings about

your conversation about, Hey, which OS will we go to from here? Yeah. Look, what was the new contender? Somebody sent in the one that's all virtualized. Do you remember? Oh, E X VM. I think it was. That's a shell. You're talking about ESXi? ESXi. That's the VMware thing? No. Man, I can't remember what it was called. It was the thing that somebody sent into the letters episode last week.

And when I looked at it, it was like everything's running in a virtualized container. Oh, yeah. The EXWM is the Emax where it's like it's all just Emax. Yeah, the whole.

desktop managers emacs system running emacs that yeah okay that checks out anyway but anyway yeah we're here we're here today uh to talk about intel uh adam and i went uh intel intel brought us to arizona a couple weeks ago to see the their new fab and to find out what's coming up with their next gen mobile and data center processors and like spoiler alert we skipped the data center processor bit and and went to the fab and learned about laptops and then

you know did other stuff the data center day because we're fine there yeah cancer lake like i guess that's probably appropriate for your audiences i mean is that xeon still is that is that Or that's like Clearwater Lake or whatever, right? Clearwater Forest. It's all it's all code name stuff right now. Yeah. Clearwater Forest is data center and it's like for.

the kind of CPUs that it seemed like the kind of CPUs that Nvidia did the deal to get access to in their inference machines that they're using to train new AI models with.

But but like I said, I went to the first like 30 minutes of that session. I was like, we're never going to talk about this in a professional context. So I did something useful. That's not consumer stuff. But yeah, you did see consumer stuff. We did. We talk about that. Can we talk about the junket real fast before we get to the CPUs?

Yeah, they didn't call it a junket. As somebody who has been on endless numbers of video game junkets over the last 20 years, although not so much these days. Yeah. Like, what is a CPU press trip like?

It probably smells the same as the game. I'm just kidding. No, everyone smelled great. Well, this one, I would say specifically a lot of the... junkets around like technology deep dives like this are are around uh trade shows right so ces computex things like that so the best case scenario is something like this where you can

Have time dedicated. You don't want too much time. I feel like this one actually went on a little longer than it needed to, but for the most part, you want time to just sit there and absorb and get to chat with people. not only from the company that you're visiting, but also other press and analysts around. So this one, like when we're at trade shows, there's a whole mixture of analysts, press influencers.

the whole the whole gamut uh at this one there was a lot of analysts a lot of press but more traditional press i will say and then there were only a couple that i would consider influencers uh so it was it was a lot more um yeah i was there man classic yeah oh yeah i'm sorry yeah will will the influencer i was the biggest influencer there

pretty pretty serious okay did did they at any point try to put you in a helicopter or perhaps a dune buggy no they put they let's go inside the fab okay all right then then this is different than a lot of video game press trips i've been on The other big difference between most of the video game stuff I was exposed to at Maximum PC and later on is that they also brought the actual people that do stuff here.

So like at most of the video game stuff, you'd have the people who were trained to talk about the video game and that was maybe it. And there'd be people. You'd be talking to somebody and you'd be like, oh yeah, I designed all the combat for this multiplayer mode when they were doing the demos, but they didn't make those people readily available for you to talk to. And at this, we sat down with Tom Peters and Tapp.

I've talked about GPUs. He was just walking around and like hanging out. We could sit down with him at dinner. Steven Robinson, who's in charge of x86, was there and we could talk to him about x86 architecture stuff like the smart people. Yeah. Tap. like essentially spearheaded g-sync the whole g-sync initiative at nvidia this the the x86 architect guy i mean he was a nice guy he he just looked like a normal guy you know he's like oh okay cool

We're just hanging out. It's like, what hot new instructions are coming down the pike? He's like, we're not talking about forward looking products right here. But we talked to the person who the both Stephen. And Tap and Rajshree Chaboskar, who is the, they're all Intel fellows, which is like, I think when Tap was on the Full Nerd a few weeks ago, he described it as the person who's able to kind of cut through the.

the kind of the back and forth arguing that engineers like to do by saying, you know, I think that this is probably the right way to go. So they have a little bit more weight, but like Rajshree.

is in charge of thread director right the thing that determines what goes on p cores what goes on e cores and how they like how that communicates with the os's and and so we were able to have much deeper conversations about how things actually work with the people who really know instead of being filtered through the PR and marketing team, which is like, those people are great, but they don't have the same level of knowledge as, hey, I'm in charge of x86 right now.

Yeah. I mean, especially with such an intensely technical subject. Yeah. Like I was, I was over my head talking to Steven just to be like, I was like, I'm really excited to talk to you. I have no idea what to ask. Cause like you're, you, this is a level of. how this works that goes beyond what I've ever really cared about. I didn't say it that way, but that's the reality. Well, also, I would say

You know, you're looking for like extracurricular activities. I think this one, because of the tours through the fab, like they were going literally for three days straight, almost four days. Oh, yeah. And like 10 hours a day or 12 hours. I can't remember exactly. So like, I think they were so busy.

coordinating all that stuff that they just didn't worry too much about the the other stuff that's fine i mean i'm mostly joking to be clear we usually skipped the extracurricular activities because we had to work but Yeah, well, I'll say the Ryzen 7000 launch, which was like the first junket that Gordon and I went to after Gordon Mong, the late Gordon Mong. We went to it like kind of post pandemic. It was like the first one of those and it was at a resort in.

Outside of Austin, Texas, and they did have like an axe throwing time that you can go throw some axes, you know, things like that. That sounds like the classic chunk. Yeah. So those exist still. And like the meals and stuff, it was all just catered by the hotel and was just like it was nice hotel food, but it wasn't anything super like it was hot, hot table buffet style everywhere we went. And like.

cold sandwich boxes and stuff like it was the kind of stuff you get it like it was better than when you get a convention center in like the meeting room or something like that but not like not fancy Sounds reasonable. They did have. Hold on. They did. This is really important. They did have miniature horses delivering beer one night. Yes. Oh, yeah. This was during the evening dinner time. Yeah. Yeah, so the miniature horses were Taco and Corona were their names.

And they had like coolers on either side of their back that were kind of balanced against each other with like a little napkin holder and an umbrella and limes. And then they had tiny sombreros into servitude. I mean, look. Who am I to say that we shouldn't take control of the natural world, Brad? That is more Bacchanalian than anything I ever saw on a video game trip.

I petted Corona and I gave him a snack and he was we're best friends now. So I don't hear any smirching of taco or Corona. That same evening, they also had a couple people set up doing leather work so you can make your own leather. suitcase badge okay all right things like that so all right yeah All right. Well, we're here to talk about CPUs, so. Yeah. CPUs, central processing units. Well, actually, it's not just the CPU. There's APUs, or I'm sorry, GPUs, NPUs, all of the PUs were talked about.

Well, the big thing is that like, this is where this is now like the, depending on how you counted, like the second full generation of Intel's disaggregated design. lunar lake and arrow lake and meteor lake which are kind of different paths down the same trail um all have chiplets in the same way that the risins have had for a long time

And they were talking about what they learned. The big thing about Panther Lake is it's the first thing on Intel's 18A process, which is their next generation cutting edge fab process. Like intended to be a TSMC competitor, right? I mean, I'm sure that's what they would like to say, but does that, does that seem realistic? Well, yeah, I mean, they, they were.

They were very clear the whole time being like, we believe this is the most advanced process in the world and it's produced right here in the great U.S. of A. Interesting. They were hammering that home. So and not to mention the previous CEO was kind of the spearhead of this this whole effort. So he doesn't get to see the fruits of. of this labor. I mean, who knows how long AT&A was actually in development. Maybe it was way beyond, but yeah.

So, okay. So the big stuff on Panther Lake, it's like I said, it's a, it's a heterogeneous there's multiple processes. They're almost all Intel processes, except for a couple of the GPU core options are TSMC.

um all of the gpu uh oh really i thought the four one was an intel one oh okay um but yeah so there's basically the cores are the tiles on this core and they're calling these socs still but it doesn't stand for system on a chip anymore uh that what it was it's system of chips now is what they're called yeah yeah sorry what yeah there was a groan in the audience when that one went out yeah yeah um

But but the main. OK, so the big stuff that's changed, smaller transistor size, the nanometer scale of the transistor seems to be getting kind of wobbly in a way that you report the number because everybody's. Like, it seems like maybe that's not going down anymore. Now we're doing new stuff to make the machines run more better at lower temperatures and voltages. The big jump here is it seems like they're moving.

Beyond FinFET gates, which was the thing that rolled out about 10 years ago. And that was. So if you think about the way a transistor gate works, there's a conductor that goes through something and then there's a gate around it that prevents that signal from it. It either stops the signal from going through or lets the signal go through. And FinFET is called a FinFET because.

the conductor that goes through was a vertical like vertical poking out of the semiconductor base right so it generally is pretty good But if you put too much voltage through, there'd be leakage around the edges of the gate, around the edges of the conductor that don't go through the gate, right? What they're doing now is they're lifting that conductor out of the stratum and making the...

the gate, a three-dimensional loop around the conductor. And at that point, that's why they call it ribbon-fed because the conductor is completely encircled by gate now. So what they're saying is there's a lot less leakage.

or or maybe no leakage that part was kind of unclear they didn't give they were there as a general rule uh we should we should start this out they wouldn't talk about yields at all no matter how we tried to trick them into asking answering those questions well it's still early right I mean, they announced full production. Yeah, they announced full production. Oh, they are. OK, it's not early. Well, the rubber has met the road then the day that we were there.

maybe four days before we were there, they announced that the fab was fully operational on October 9th. So it's like, we were not, it is still early in production. And as they said, they're still ramping production. When we were on the Fab Tour, we'll talk about this later, there were places where there were holes where it was very clear that there were going to be some really expensive machines that just haven't been delivered or installed yet. I see.

I don't want to digress too much here. This is like the most basic like transistor one-on-one stuff, but do you know what the mechanism of action is that causes the gate to interrupt signal through the transistor? I mean, it's not...

Obviously, it's not mechanical at a nanometer scale. Do you know what is the operation there that is interrupting the electrical current going through or not? Yeah, when basically there's... signals that go into the transistor base that apply into the gate, into the base of the gate.

that make it more or less electrically conductive based on how much electricity goes into it. Got it. So if there's current at the base that it's open and if there's not, then it's closed. Okay. And that's the extent of my knowledge on how this works. Okay. Good enough.

please don't ask follow-up questions next question yeah uh so uh the next thing is the power vias So the way that the power comes into the wafer, the things that turn those gates open and closed, and those are kind of buses is my understanding, kind of like a... in the in the sense of like a system bus like that like the power comes in on lanes and then they can open or close based on state of the chip what's being used and not is that those have come in on the front side for all eternity

And now they're making those poke through the chip and coming in from the backside now, which will improve cooling and reduce power loss reduction and basically allow them to scale more power into the chip. which gives them better performance um and they specifically like they wouldn't talk about performance they wouldn't talk about yields at all really other than to give really general numbers like

It's 15% performance per watt improvement between 18A and Intel 3, which means nothing, right? In what? Yeah, what? Performance per watt, Adam. Performance per watt. And then like many of the talks, this this whole talk about what fundamentally has changed at the low level with this process was.

This is the most advanced semiconductor facility in the world. Like they said that three or four times throughout the throughout these presentations. I mean, do you think, OK, this is a very big question that they certainly would not have addressed? I mean, are we going to see iPhones?

You know, are we going to see like the A20 or whatever the next, I mean, maybe not the next one, but like three iPhones from now, or is Apple going to be fabbing SOCs in this place? Adam's making a face. Yeah. I mean, I'm sure Intel would love.

to expand their partnership to work with Apple. So that's going to happen. You know, I mean, we will see. They showed an arm chip that they built on 18 a. My understanding is that the big thing keeping Intel foundry fabs accessible only to Intel as a customer is that... the rest of the world uses a certain kind of tool chain and process to design chips and intel uses their own thing this is um i talked to ian dr ian cutris

who is Tech Tech Potato on YouTube about this a few weeks ago, a few months ago now, I guess, or last year even. And his answer is that Intel does something, does this one way, everybody else does it the other way. Part of the reason that Intel has been building chips on TSMC beyond the fact that they didn't have competitive processes for those chips is that they needed to build that internal knowledge.

so that these US fabs can be competitive with the TSMC fabs in Taiwan. And whether they're successful at that, we're not going to know for a while. So, yeah, I don't know.

The other part of it, though, that's interesting is because of the way the packaging works on these chips with the Foveros and all that stuff. Like, there's nothing to say that Nvidia can't make a chiplet at TSMC in Taiwan, ship it to Chandler to be assembled. and put that on with intel 18a uh cpu dyes and other stuff right

I mean, yeah, after after that announcement of the partnership with Intel and Nvidia working together in some capacity, which it's still too early, they keep telling us this is just the announcement, the intent to work together. We probably won't see anything for years. Yeah, I think it's the packaging part that NVIDIA is very keen on. This Foveros technology is probably where Intel is going to be able to hang their hat on.

Should we talk about that? Like what I've heard the term Foveros a million times and I still don't quite know what it is. So Foveros is the, is the kind of glue that holds all these chips together. Okay. It's, it's, it's a, it's a. general interconnect material that has varying bump pictures for the connection points for the chiplets um and they talked about like

For example, the data center parts have a different density than the consumer parts just because they want way more chip to chip, chiplet to chiplet interconnection and communication than you need on the. on the on the uh mobile parts where there's only basically one compute tile um so foveros s i think is the 36 micron bump pitch That's on the current chipping parts. That's going to go to 10 microns now. So 772 to 1600 bumps per millimeter squared.

to 10K bumps per millimeter squared. And then Clearwater Forest is going to be nine microns on that pitch. And what that basically means is how many connections each chip we can have. So that's the gate.

of that's the gate for chiplet to chiplet communication uh they they also did a bunch of other stuff that we should talk about like we should talk about um the the soc chip the one that has all the io and all the other stuff on it that makes the cpu work that isn't an 18a chip so there's the 18a chip

There's the SoC chip, which is like all the chip housekeeping memory controllers and that kind of stuff, as well as sometimes some E cores. And then there's compute die. Before we move on, like interconnect is actually like.

one of the main things I wanted to talk to you guys about because like that's been one of the big weaknesses on the Ryzen for so long since, since they went to chiplets, right? Like it's the relatively slow infinity fabric between chiplets is kind of the big problem when you get.

It's the whole reason we are still living with core parking, right? Because when you get a process running on one die, trying to hit the cache on the other die or whatever it is, like talking between those two chips, chiplets is. relatively slow enough to like cause hitching and games and other issues like that. Like, did you get any sense if, if like the Foveros tech will alleviate that issue or won't.

face that issue to the same degree? I mean, were they talking in such real world terms about things like that? I asked about that when I was talking to Stephen. Did you ask at all? No, I didn't talk too much about that. Yeah. So. The answer, they wouldn't give a clear answer on that because they're not doing consumer multi chip, multi compute chip products at this point. None of those were on the announce, like everything that's being announced.

like there were there like i said there was an eight core a 16 core and a 16 core with a big gpu model all talked about okay all of those are one Compute tile. Right. Like Arrow Lake is all single compute tile. I think the rumors, I mean, this is, this is getting further down the road. Like the rumors swirling about Nova Lake are that they are going to start shipping to compute tile designs. But that's, I mean, that's.

It's actually not fully accurate because the main tile that has the I.O. and stuff like that has low power E cores. where the you can expand the the bigger 16 core option actually adds a whole nother that's section with with other uh processor course e-course on it so

I'm getting out over my skis here now. I've read that those LPE cores are not on the ring bus, but I barely know what the ring bus actually is, so I don't know what the effects of that are. I don't remember the actual layout. I'd have to bring up the slides. I do. Okay. So the layout is that the four P cores share an L2. The four P cores are all connected via L3 cache. Everything on the computer is connected via L3 cache.

The E cores are connected via shared L2 in clumps of four. So if you have four E cores and four P cores on that compute tile, then the E cores. have l2 to l2 uh can talk across the l2 cache that's shared between them the p cores can talk across the l3 cache shared between them and if you need to dump something out to those lp those low power e cores

that are on the main IO die, then that's a really expensive process. And we talked about, I talked about this with Raj Sri a lot, because it actually comes up in benchmarking, like the thread director performance. And the way the way the system and the OS know whether it's worth computationally speaking, moving a thread from a P chord or an E chord or an E chord or a P chord is is very dependent on like how long it takes to.

transfer the data. I like how many clock cycles it takes to transfer the data. And she, she was talking specifically about like 20,000 clock cycles as an expensive, as an expensive switch over, which is like one, 200,000th of a second, just to be clear. We're talking about infinitesimal slices of time.

and and i'm not going to get into that whole thread director thing you should like the conversation i had with her i think that video is on the full nerd channel right adam on the the full nerd network youtube channel yeah is is was fast like it was i learned so much about how that stuff works and why it's hard i knew it was one of the big computer science problems going into that conversation and i came out

with a much better understanding of why it's difficult and a better understanding of kind of where we are with that stuff. Because it really plagues all of the modern architectures, Apple, Intel, AMD, everybody is having the same problems. uh well and a amd uh is is working on this as well like you reference the the chiplet design that has made them kind of uh made their mark on the industry and changed a lot of things uh with stricts halo

I was watching a video by a gentleman we actually met while we were there. His YouTube is called High Yield. He essentially does a real deep dive architecture stuff. I've learned a lot from him. I'm no expert. It was great to chat with him. He actually did a really good deep dive of the of like the I guess topology might be the right word of Strix Halo and how AMD seems to be using that. to experiment with a new type of interconnect that fuses the huge GPU.

that they have on on that processor with the uh the rest of the the processing and like at least according to him this is like super advanced stuff so that's why i think um Strix Halo is important. And then this this Fovero stuff is definitely important on the Intel side. So one of the things they talked about, like Tap in his talk, he's talking about GPU stuff, talked about using.

especially for like games that are built for consoles, using the eight E cores probably makes more sense than using the four P cores because the eight E cores have that shared L2 cache. and also they're they're comparable in terms of like instruction sets and functionality to the era of current generation cpu cores so you don't get any like it's not like there's avx2 instructions or something like that in the in these games that are going to take that the p cores are going to take advantage of

Right. Cause the current, current consoles have Zen two cores in them. So it's not like you need a lot of modern desktop cores to equal that. That's exactly it. Actually. Yeah. When I talked to, I, I had a dinner. Set up between AMD and Sony about Project Amethyst. Mark Cerny was there. That was the first time I got to chat with him. And he was saying at least, you know, I mean, they weren't talking about forward looking features, but he was talking about how even on the.

PlayStation 5 Pro, the Zen 2-based CPU cores in there, he didn't feel like that was a limiting factor, that it was more other things that they're working on. uh and yeah the same and the other thing to think about is that these are these are very compact uh apus like all the all the wattage going into that needs to be shared all the heat dissipation is is makes a huge difference compared to desktop right these these are packages that are going to be put in laptops

thin and light devices, handheld gaming PCs. So the more the more perform or the the the more juice you put into high performance cores might take up some of that.

uh, overhead when you'd rather be sending all that juice to the GPU. So, and, and the big, like the big thing that's happening with this debt, uh, this generation. like last generation we saw meteor lake and um lunar lake and lunar lake was kind of the mainstream part and meteor lake was um was in the like desktop replacement kind of workstation-y mobile parts right with this panther lake is going to take over all of that and and both the mainstream

Well, not necessarily because they in all their slides, which once again, a lot of them are vague for a lot of reasons. They never mentioned Air Lake HX competitors. So with discrete graphics. Or like the highest end, like real thick, chunky eight pound gaming laptops. Usually it's, you know, you got a low power CPU, kind of a mid range and then a really high power one. They seem to indicate that the HX.

This would, like the top end configuration of the core count would replace that, but they never set it on any of the slides. All they were doing was comparing it versus Lunar Lake and Arrow Lake H. So I don't know if that means anything, but there might be another spot in the stack. I still believe that that Arrow Lake.

uh, on the desktop is going to get a refresh. And I think maybe the highest end laptops will also get an air lake refresh, but we don't know. Those are, you know, there's solid benchmarks for the. i7 270k just plus just popped up or whatever they're calling it but so like panther lake is explicitly mobile only right there is no desktop component to this next generation that we're talking about

I mean, unless you fuse it to a mini ITX motherboard or something. I mean, we've seen some of this stuff, but yeah, this is, this is meant for, for mobile computing of various sources. Okay. Got it. So that's kind of the high points. They talked about the GPU stuff a little bit. It's still like it's Intel. They're calling it XC three, but it's basically the same core as the as the B series desktop GPUs.

Wait, are you saying that's what'll be in this? The next CPU is like in Panther Lake. That's what's in Panther Lake. That's the same technology. It's not. Yeah. Same architecture. Smaller. smaller total grunt or whatever. But, and so that's what's in like the current arc cards is what you're saying. that's in the arc b series cards yeah yeah so that battle mage yeah like b580 okay well that seems respectable first integrated gpu yeah they've done stuff like added multi-frame gen

They're not doing the kind of last second mouse warp, mouse movement warping that NVIDIA is doing with their five series cards. But like the performance, it seems fine. Like we got to I got to play the new painkiller game. on one of their dev boards one of their dev machines but it was a pretty rough demo just just because I was playing with a mouse and keyboard on a 40-inch TV about eight inches from my face. So it was non-optimal. What I will say is that the multi-frame gen on that.

on that demo, like even doing something simple, like trying to land the mouse on the corner of a pillar or something like that was really, really challenging. So it was a little floatier than I would want for a mouse and keyboard, but it seemed fine on the gamepad. wait to be clear you were you were playing on an integrated gpu on a panther lake machine not a discreet okay i mean it was i don't they didn't have the machine out they had a lot of panther lake dev units which were like

sorry, NUC, bigger than NUC size open boxes with small coolers, like 15 or 40 watt coolers on them. Are these, are these like prototype, like not for production? Like, yeah. Kind of similar to when we would see consoles before they came out and it was not the console. It was like some giant cobbled together. They had a mixture. Yeah, like they did have like.

They had one of the same things that I actually saw at Computex, which was kind of like a huge dev box board, like not pretty at all. Yes, I love the utilitarian hardware like that. They did show off the first nooks.

style uh development kit which is a little more polished but still meant you know for developer usage it was not i didn't see one that was powered on the only stuff that i saw running that was powered on were uh reference laptops which are usually taken from an oem and just kind of configured with a different platform so i i i don't know do you remember seeing any of the the

So the painkiller one wasn't out on the shelf. It was running on something that was in a cabinet. So I don't know what it was. The other ones were almost all, I think that they had one of those dev kit boxes running something in the middle. but I don't remember for sure. Anyway, it doesn't matter. Like the demos were like, they had some premier demos. They had a bunch of AI crap that was kind of iffy. I just, I also think like.

It is I mean, it's hard to set up an event like this, right? It takes it takes a lot. You know, my hat's off to people who have to organize this kind of stuff because I'm sure it's a lot. But unfortunately, and you kind of pointed out a little bit, sometimes like the environment that you try something out. in just like kind of like sets the tone, if that makes sense. Like I listen to Digital Foundry a lot and very often they'll say they'll go to something like recently they went to.

what was before you put gamescom and they took some demos at gamescom but like the tv had like motion smoothing on and it was just not the hdr wasn't calibrated right and you know like they're like guys i can't I can't really pay attention to what you're showing me here when the environment that I'm supposed to see it on is not doing it justice.

So I think part of it was that for Will. But I'm sure you deal with that all the time with games, right? When you go see something. I think that was like the infamous Games Beat Cuphead footage thing, right? Wasn't he? Wasn't he like kind of, Oh, Dean Takahashi. I think, didn't he say he was like, like talking to a PR person and kind of half paying attention to the game when they were recording that? Yeah. I mean,

The error there was then going and posting that footage. I mean, is it? That's where you maybe practice some editorial discretion and say, oh, this footage is maybe not usable, but it absolutely happens all the time. Hey, this, this environment that I'm checking this out in is compromised in one, two, three different ways. And I've got to kind of integrate that into my impressions. That's why it's hard. Cause it's just like, like.

And I'm not saying that your experience was not valid, Will. It's just more like... It's not representative. Yeah, not representative. So when it's like, oh yeah, we got to try something, but it was... for all intents and purposes not worth anything uh because it was you know just a bad experience it was pre-release hardware pre-release software on a game that just barely got announced all that kind of stuff and it's just like oh

When you talk about organizing these events and how Herculean that task must be, like I think about they probably had to commit to the footprint of that demo station way before they knew if that demo would actually be ready to show or not. Right. Like they had to.

They had to build out a list of here's what we're going to show with this thing probably weeks in advance of saying, hey, this demo is actually functional. I'm 100% sure. Like the actual demos, I'm sure, are all coming in as hot as possible. And that's why I like I'm always so like I like going to these things and learning about the technology and, you know, growing my my knowledge set because I'm still pretty pretty fairly early in my journey here. But like.

Really, the fun part is like, OK, when when can I get it and try it out for myself, have my own experiences in a in in. you know a controlled environment you know we always say wait for independent testing because it isn't it is important in a lot of ways but that that's what i'm always excited like okay cool thank you for showing me all this stuff but man when can i get my hands on it because

I just, I want to, I want to see it. I want to feel it. I want to see how, how it, uh, how it works. Um, so, okay. So I did, I do have a update on something that I'd forgotten. We asked during the architecture round table when we were, when Steven was talking, uh, and.

we asked what the bandwidth between the CPU and GPU tile was like, because they're both connected via that Foveros backhaul. And he specifically said it can be as high as they want. Like there's, it depends on how many of the... the connection connections you choose to use um up to like like you can do you can bump up the frequency and you can do up to like 150 gigabits per second but

the thing you said is once you do that then you start shifting bottlenecks around and there's definitely sweet spots at different places that require um that are difficult to predict because of the way you're moving the bottlenecks The takeaway I had from this is that when they're working inside these SOCs, especially when they're talking about doing bigger GPUs in some places than others, is that you can get kind of the same benefits that we see by using APUs on.

consoles where they share different parts of the of the of the they share some high speed connections and then some low speed connections out to memory and stuff like that so you can end up in these weird hybrid architectures that are similar to desktop, but not exactly analogous and have some real strengths over the, the, the change where the bottlenecks are on these, on these new processors. And I thought, I thought that was kind of interesting, but also.

They're not at a point where like the GPU can dig into L3 cache on the compute die, although it can do like some he said snooping, which I don't know if that's a technical term or not. But anyway, speaking of gaming. I was surprised by how much they talked about handheld gaming. That's obviously one of the new areas. And I know, Brad, you have not really messed around with it too much, but then also talked a lot about AI.

you know, a multi-frame gen, all that kind of stuff. When Tap was on The Full Nerd, we had a very long discussion about where he thought AI was going. But then also ray tracing, like the ray tracing performance in this GPU is one of the places where they wanted to really boost performance because it is such a big thing. Like, man, there's been so much talk.

recently about the the next gen consoles i know you guys have talked about it over on next lander stuff like that it it's it's always interesting to hear You guys not or maybe I don't know how to phrase this. You guys talk way less about ray tracing over a next lander and like that kind of stuff than what a lot of these junkets will talk about. And like.

Yeah, I've always wanted to like write in a question to NextLander and be like, you know, are you do you guys notice the ray tracing stuff? How do you guys feel about the ray tracing stuff? Obviously, they're building that into the hardware. Mark Cerny is coming out saying, you know, that it's it's an important part. Yeah, absolutely. I mean, I personally, I remain incredibly bullish about the role of ray tracing and development and.

graphics in in the long term but i think we're still like on the in the middle of the you know whatever we're defining as long term right like i felt like when real-time ray tracing first became a thing it was like okay this is going to take like 10 to 15 years Before this is absolutely integral and everywhere. And I would add that I think that is also kind of not a popular topic among the mainstream right now. Yeah. Because we're not to that point yet. Like.

A lot of people still just see it as a feature that offers diminishing returns for a huge performance hit. And also, I feel like a lot of people don't know that they're seeing it. Right. I mean, especially like on PC, there's usually a toggle.

some sort of you know hey do you want to turn ray tracing on do you want path tracing things like that on console you're usually picking between different uh quality modes and sometimes the quality mode will say oh yeah this involves ray tracing but sometimes you have to wait for the digital foundry video where they're like oh by the way this part of this pipeline is using ray tracing uh you know it's one of those things where it's like

Man, how do you how do you really sell that? How do you get excited for next consoles when it's like, man, check out this ray tracing performance? I don't think they're going to have to sell it so much as it's just going to become a fact of life, because when the next consoles come out, they are going to be so capable of ray tracing that everybody's just going.

to adopt it into their pipeline and I think it'll just be there. I love that optimism. You're seeing it now. You're seeing it like Indiana Jones and Doom were the big games this year that you can't run on non-ray tracing hardware. That's true.

I think there are going to be tremendously more games. In fact, I think just about every game. But again, that's that's 10 years from now is what I'm saying. It's like it's not going to be until the next consoles have a pretty significant install base that then developers, because that's how it's always gone since. Since the 3D, 2D transition. Since the PS2 generation, if not before, it's always been the consoles. What the consoles can do dictates what trickles down to PCs.

And so once every console on the market, like even the Switch 2 has capable ray tracing, right? I mean, honestly, once every console on the market can do significant ray tracing, then developers are going to be like, okay, well, now that we are doing this by default on the... set-top boxes it's also going to come to every pc game

Yeah, and that it's it's it's just interesting, right? It's like you always hear a bunch of buzzwords and the companies are like, to a certain degree, we have to put the cart before the horse and we need we need NPUs in millions of. of cpus before uh or processors before you know you can see anything that actually uses it day to day right uh it's interesting getting capable ray tracing into

you know, mainstream laptop hardware, the kind of computers that most people buy is part of that. Right. And I think that's, that's, that's a step in a, in that direction, whether it's, it's obviously not going to perform as well as a discreet.

standalone GPU or console even, but, but it's a step in the direction. Yeah. I mean, it's, it's, it's, it's the kind of thing where like, you know, iPhones also have retracing hardware in them now, like at some point in the not too distant future, every current shipping product.

It's going to have some amount of ray tracing and some amount of machine learning capability, and it's just going to become a mundane fact of software development like everything else. And then the other part of that is that.

as the comes more capable than we'll see software that actually takes advantage of it like that's what i mean there were a handful of puzzles in indiana jones that used the the light rendering to be as part of the puzzle but it was kind of a it was it wasn't super common in that game right there weren't a bazillion of those so um and also

If you didn't know how it worked, then it was functionally no different than like a light reflection puzzle in Breath of the Wild or Tears of the Kingdom 10 years ago. Yeah. And like, I thought it was cool because I recognized it and I saw it.

But everybody else is just like, oh, okay, whatever. I mean, that's always been part of the hard sell with ray tracing, right? Yeah. The fakery got so good that telling the difference between fake lighting and quote unquote real or realistic, physically realistic lighting.

Look, it's the same for a lot of people. It's the same industry, though, that counts the reflections on sweat on people's faces on console drums. So like, hey, look, look at your nose shadowing the side of your face should be an easy talking point after Puddlegate. To me, to me, the ray tracing thing has always been a dark alleyway phenomenon, meaning that like every game that has fake lighting gets a lot of nice artistic attention in the critical path. But like.

Dark little alleyways off to the side, they don't bother doing all the hand-tuned lighting to make everything look super great, right? Like, the thing with ray tracing is, like, hey, everything gets the same lighting now. Like, they don't have to go in and hand-place lights and art direct. lighting to make things look the way they want like like now algorithmically just everything gets the same kind of realistic lighting globally the the early ray tracing demos right that were touted out

whether it's Battlefield or even Control, where they're like, oh, look at these reflections. Wow, these reflections. And even I was like, this is cool stuff, but eh. And it wasn't until like the global illumination stuff, like Metro, was just like, oh, wow. that like that is impressive for me because you could actually like

like real light bouncing around. Yes, that's exactly it. I noticed that kind of stuff. That's exactly it. I got, I got kind of bummed out when ray tracing became the, even for people who were into it, it kind of became the like better reflections tech quote unquote for a long time because like. Bounce lighting is the actual super interesting, like brain tickling one for me is like, you know, light shines on this apple and then it picks up red in its, in its bounce as it, you know, now, now.

Now you've got like a red glow on the thing it's bouncing on too. Like that's the kind of stuff you didn't really see before. Well, and there's secondary and tertiary effects as well. So. Not to make this whole thing a treatise on why ray tracing rules and the Internet's wrong, but like because the cost of moving lights and changing things that affect the light rendering in a game goes down.

For developers, when they're iterating, it means they're able to iterate more and they're able to make more changes and they spend less time on that and more time on other things. And then there's the whole tertiary effect of, oh, by the way, if we have.

hardware that's really good at racing tracing rays we can make more realistic sound models and things like that and we because we can use that same hardware to do to do sound renders yes i have spoken to at least one game developer who has talked about how excited they are to use the

Use the properties of ray tracing hardware for things other than lighting and sound. Sound was one of those like AI, not not machine learning AI, but enemy AI was another one like they basically were very excited about like the you know, it's. because it's just hardware for modeling interactions in a physical space. And you can think of all the other ways that that would be useful beyond lighting.

Well, it's funny. I talked to somebody the other day who's building a pathing system for their game that uses the ray tracing hardware. Right. Because like.

the expensive and hard thing about generating what's pathable and what's not is that you run this thing that takes hours and hours to run but if you have ray tracing hardware in the machine you know you're going to have that you can change that dynamically based on changes in the environment in a way that's unscripted and like like anyway there's there's a bunch of cool knock-on effects that come with ray tracing hardware that aren't reflections or shadows or

Or reflected light even. So anyway. Yeah. My feeling is it's going to be like next consoles plus three years or so before we really start seeing like kind of universal effects. I think that's showing up in games, maybe four years. I don't know. I mean, the good thing is, I think, I think, I think it's going to be based on the sales rate and how long the, like the inner, the kind of transitory part of the generation is like, if we look at last go round.

There was a long period from launch of PS5 and Xbox Series X to the games not being released on PS4 and Xbox One. Yes. And I think like that. being so long kind of held back that first part of the generation yeah that's totally right and that's going to be the gate on this because like the thing is the hardware has been there for a while ue5 is good at ray tracing both in software and in hardware

And will like the technologies there, the people who've been working on this will be able to do it. The question is, does it make a market sense? Cause they want to be able to sell games on PS fives. I don't think they care about the series X. So that's half of that problem solved. Anyway, should we talk about the fab? Yeah.

Because I think this was this was the cool like this is why I went to this trip. I could have looked at the slides when I got back and gotten most. I mean, actually, that's not true. Talking to the Intel fellows was it was. incredibly informative and super worth making the trip to arizona for the fab tour was kind of icing on the cake for me did you guys put on bunny suits we did oh yeah head head to toe we we had a

We had a beard nets as well as hair nets. Well, back when I had a beard. Yeah. Back in those days. Yeah. No. Yeah. I felt like. Because we had small groups. I think it was like 12 of us that was going in one at a time. Yeah, that's right. 10 or 12. I'm sorry, not one at a time, like the group at a time. And yeah, I felt like I was a piece of meat getting shoveled through because they would give you your suit.

Your whole your whole setup and then they would have somebody like help you do it step by step. And it was intense. They kind of warned you it was going to be warm inside the suit, but also in some ways, like then later.

when we were in it, like we could kind of stick out our nose and they were like, oh yeah, it's fine. You can have your nose exposed. We want you to be comfortable. It's like, wait, I thought you just had us go through all this stuff to minimize dust particles in the air, skin particles.

Does that officially qualify as a clean room, did they say? Oh, yeah. So there's ISO standards for clean rooms that are based on how much particulate there is per square foot or square meter of air or cubic meter rather of air. And this one is not. So this fab, the 18A fab that we were in, which was fab, what, 50, 53? Oh, I don't remember the numbers. I don't remember. Anyway.

We went through basically one fab to get to the new fab, and both of those use these things called foops that are enclosed wafer holders. We've seen them at the flea market before. They're like big plastic containers that are the ones there were kind of orangey. I seem to keep certain wavelengths of light light out and they carry the wafers from machine to machine in these caddies that ride on rails above the floor of the fab.

like 30 feet up. And so they have a lower standard than some other fabs that use exposed wafers where they need much lower particulate. They wear fully enclosed heads or rebreathers or masks or whatever.

um but yeah we so we walked into the the like the clean room to get into the clean room area we had to be badged in through a gate and once we were on the other side of the gate we stepped in and put uh uh like foot coverings on shoe coverings on i was waiting to hear about the booties i knew that they had to be in there yeah so we put booties on to keep dirt from our shoes off of the floor in the clean room gowning area

And then we put on beard nets and hair nets to keep the particulate from our face. Cause like the humans, the grossest thing in the whole fab, right? We're just constantly shedding anywhere. Really? Not just them. And so then we, we got, we went in, they sized us up for the bunny suits. The people there immediately knew what we were doing and fit us perfectly. And we got the bunny suits and then outer outer.

Oh, and we put on gloves before we got into the clean room area to like cotton gloves. Yeah. So we put on the gloves we put on. We got boots. We got a hood. We got a bunny suit. And we put the bunny suit on, then put the boots on on top of those, put the hood, which got tucked into the bunny suit, had our hairnets and beardnets on already and put on like.

Latex gloves on top of the cotton gloves. Is that right? Or was it the other way around? Latex first and then cotton. Yeah, yeah. So, sorry, nitrile. We used nitrile as a latex allergy. People didn't die. But we didn't have to wear a mask. Like the hood kind of covered your mouth. Kind of covered your nose if you wanted to. And we put goggles on on top of the glasses we were wearing. Or if you wore eyeglasses. And then once that was all done, we...

like put our little badges on. So they knew we were part of a tour and not to be trifled with and kind of got led out into the fab floor, which was almost like there was the thing that struck me immediately was how much movement there was. There was a lot of people, a lot of machines. Scurrying. The caddies that carried the foops around were constantly going and they were moving the what?

The foops, the things that carry the fiber, the wafers. Yeah, they look like little plastic containers that just have like a stack of wafers in it. And like that, that's sealed as well. So like, I think that's why like they have multiple layers of ceiling. Like I always thought of a clean room is like a.

like a, like a kind of like a medical space. Right. And you got very limited people in there and they're doing stuff real carefully. This was like, I mean, this is a huge, huge building, uh, you know, with, with dozens and dozens of people all doing all this different kind of stuff. And they were talking about how the advanced the filtration system was in the building, like the air the the air was moving from top to bottom and the bottom grading was like vented and it was just like, yeah.

So even if you shed a particle, it was going to be sucked into the floor within a matter of seconds. And like you could feel it just to be clear, you could feel the breeze even inside the bunny suit. And so. So you walk into this room and there's like a bunch of hallways that I think were like drop points so they could drop those caddies. I sent you a link to a Wafer World article.

That explains what foops are and doesn't exactly show them, but it stands for kind of kind of looks like a record holder. Yeah, it's almost like a big CD changer, honestly, a CD changer cartridge. But it stands for Front Opening Unified Pod or Front Opening Universal Pod. And the idea is that they're sealed containers that carry the wafers from machine to machine.

So like these caddies are zooping along at like 20 or 30 miles an hour, 30 feet in the air. And there's a bunch of tubes and stuff coming up out of the floor. And the whole thing is set up. When I talked to the guy, I was like. Do your guys play Factorio when they go home? Is this like the shit that they love? Because the whole thing is set up so that the really expensive machines are never sitting idle, right? Like as soon as one of the foops gets lifted off, the next one gets put in place.

So that it's fooping constantly like like it's it's foop to foop man. Never stop fooping. Yeah. So. So, yeah. So you see these things. I'm cracked up, Adam. I'm sorry. You see these things flying around and then they get to the machine that they're above. And it lowers down the caddy on like two wires from this thing 30 feet in the air and drops it directly on the mounting port. And then the machine docks with it and sucks the wafers in.

does its thing and then spits them back out probably in the same same place. Right. And and all of the machines are lined up. So those foop docks are even with each other and directly below the tracks that that seems to be the main. defining characteristic. And then, uh, the guy who gave us the tour bill, uh, something or other was saying that they're, uh, they're, uh, uh, like that whole structure is something they have, they have buffers.

for the slower machines so that the expensive machines have wafers ready to go that are like X number of minutes deep in case one of the machines goes down so that this thing that costs half a billion dollars is never sitting idle, basically. Sure. Yeah, that makes sense. Yeah, it was kind of crazy. It's intense.

yeah i mean like and that that was the weird part is that yeah they were i mean what like 40 feet 30 feet over i think 30 feet probably yeah yeah like really high and and when they would go down into the machine they're coming down just by like a couple of wires and they're not

They're not like swaying in the wind or something like that. It just looked like they came straight down, came back up. Three miles an hour, like kind of seems pretty fast for delicate stuff like that to be whizzing around overhead. But it's all I mean, it's all sealed and it's like all that. There's no it's not like rattling around in there. Yeah. I want to ask real quick in your notes here. Yeah. You have you have listed that this land was bought in the 70s. Yes. And they.

Sorry, and they broke ground in the 90s? Uh-huh. Is this a brand new facility? Or have there been parts of this facility that were already in operation prior to this? This fab is new. The whole thing? Yeah, because it's huge. It's bigger than any of the ones I've done before. So my actual ultimate question here, have they been working on this facility since the 90s? No, no, no. There's other fabs at this facility. Oh, that's what I was asking. Okay. So there's a lot of other.

processes going on with other, you know, process technologies. Okay. I think that they had, I think they said they have a million or 2 million square feet of fab space at this, at this campus. The clean room area, you can walk from one fab to the other all in clean room. You can move foops from clean room to clean room. But I mean, it seemed like they don't do a whole ton of that.

Does anybody ask why they set up a water intensive industrial process in the desert? Yeah, I did ask that. That was one of the first things I said. 52, by the way, not not. So the reason they did in the desert is because being tectonically stable and having reliable power is like the water problem is tractable and they can reclaim.

a fairly enormous amount of water i think they said they go through 9 million gallons of water a day and they treat some enormous percentage of it i don't think they gave an exact number um But the power and tectonic stability is much more important. As the person who was giving us the tour said, if they had known that the Oregon fabs are on a huge fault at the time that they built them, they probably wouldn't have built them there.

Oh, OK. And it's interesting because, of course, Taiwan has the other big fabs like this, and it's also on like pure ring of fire out there. So I asked what happens if there's an earthquake because they do occasionally have earthquakes that. like southern california earthquakes sometimes make it through to arizona and basically all of the seismically sensitive machines have the equivalent of like old hard drive head parking technology where they just

immediately freeze and shut down if there's any kind of ground trembler. And also the building is somewhat seismically isolated, apparently. The other thing that Adam and we didn't talk about at all is that there's some floor beneath the area where this is all built. We're like.

A lot of the machines you see the tops poking out, but they're like icebergs and the rest of them are like that. 75% of the machine is in the subfloor underneath where like water and chemicals and whatever, you know, ions and stuff come through like. I've been in a couple of places that are mechanically incredibly complicated before, like car factories and stuff like that. And I've never seen anything that had this level of density when it comes to technology and kind of what you see.

it's pretty wild yeah no i mean the the the actual fab tour was definitely once of a lifetime kind of thing there's been other opportunities in the past and and i've never been able to attend so I appreciate going and checking it out and just like it is it is crazy that.

Not that I take any of this PC hardware for granted, but at the end of the day, when you're playing a game or working on Excel or something like that, you're not thinking about all the crazy technologies that it took just to get that.

that CPU into your hands. Yeah. I mean, the thing, the thing about the thing about, you know, semiconductor fabrication to me is that it's so elemental, you know, it's just metal. It's just giant disks of metal that they are turning into something that can do all the things that computer.

can do like it's just hard to wrap your head around that process sometimes well in these um so this this fab in particular takes wafers in and spits wafers out there's no like they're not cutting the chips out here or anything like that they're only doing those compute tiles here right now they're not doing the yeah they're stamping the wafers yeah or etching whatever the technical term is in smartening um yeah

And then the other thing is that it operates 24-7, 365, and there's always around 4,000 people on campus, which I didn't, like a lot of the other fab situations that I've been familiar with are a little more automated. and a little less human intensive but like there were a plot load of people around here all the time kind of kind of crazy to think about punching the clock on the graveyard shift at the at the 18a fab i mean you wouldn't you wouldn't be able to tell there's no windows there's

Nothing like the lighting was had to be weird anyway, for whatever reason of, you know, light waves. But yeah, it's got to be intense for sure. I mean, so the other couple other things that I thought were kind of weird and interesting. Apparently, if you work with copper in the fab because it's a contaminant at a molecular level.

You only can work on like, yeah, they make you wear a different suit, a different color suit, and you use different tools. You come in through a different door. And every time they walk by the non-copper people, they give you a noogie. Like, yeah, man, it is really, really classist over there. Different. Oh, go ahead.

Sorry, sorry. You mean contaminant for the fabrication process, I assume, not for human beings? Yeah, contaminant for the fabrication process. Got it. They, like, they really, like, the machines either had... copper okay or no copper written on the side of all of them and it was because like a handful of atoms of copper can contaminate a whole wafer or a whole machine apparently it seemed real real serious they were very serious about that

I guess that makes sense. It would, it would alter the conductivity properties, right. To have a different metal mix. Presumably. Yeah. They, they didn't, we weren't allowed to wear a hair product or like makeup, cologne, makeup. Nothing with volatile esters in it, basically. They did tell us we could wear deodorant, which I was grateful for. I skipped it just for fun. Yeah, I mean, I thought about it, but I decided for your benefit, I would deodorant up.

And yeah, like it was, I think we were inside for maybe 45 minutes and walked like what a half a mile or something. We couldn't take any electronics in both from an EM interference perspective because a lot of the machines are EM sensitive. But also, obviously, they didn't want us taking pictures and they like it was funny because we saw like machines running old software, like old versions of Windows and stuff hooked up to different things.

And when we asked what those machines were, they were like, well, we're not talking about that machine today. We're not, you know, blah, blah, blah. Sneak some spy photos onto a rage bait YouTube channel. It's like Intel's running 18A on Windows 7. absolutely we're running windows seven um we did see the euv machine which is the half billion dollar extreme uh lithography machine that's like the size of three school buses and we didn't get to touch it they didn't let us touch it

Yeah, it's big. Yeah, it's big. It's nondescript. You could walk by it and just think it was nothing. But yeah, it was it was intense. And there was space for at least two more of those next to it. So, you know, are those the ones that I keep seeing? Stuff about the supply chain for those is like a single point of failure where it's like one company in Germany makes the mirrors like one company in Denmark makes this and that. I don't know. It seems like they're all mostly bespoke.

But they also wouldn't be specialized. Yeah. Like they did. I don't think that machine even had logos on it for whoever made it. It's like it's like. It's like if you have to ask where you get an EUV machine, you're probably out of luck, I guess. You're probably dealing with some nation state level approvals or something. Yeah. To be clear, they do this. This machine operates in vacuum.

And you can't use glass for mirrors because the UV light, the extreme ultraviolet light is absorbed by both air and glass. So they did. They had to do all sorts of like. crazy material science stuff to make these things even possible um anyway uh that that was kind of the hype we stopped at the ev machine and they made us turn around and go back we didn't get see what's on the other side of that

And it's unclear to me if that machine, it was really long. I don't know if like the wafers went in one end and they got blasted by the thing at the end of it. Or if it was like going on a little line down it and stuff was happening as it went. So who knows? Is it possible to say? I don't even think we could tell like where the foops entered. I didn't see a foop hole on those. No.

Yeah. And there wasn't like people manning it either. Like nobody was like standing around it. It was just just sitting there doing his thing. But like talking about how they kept keep that thing fed. Was giving me real real factorial and satisfactory flashbacks to like, oh, OK, so you need a hopper that stores foops that are ready for this nearby.

And and the guy, Bill, who's taking us on the tour, it was like, yeah, all that stuff kind of happens. Like the systems that drive that just know that the the food that they're carrying has this wafer on it at this stage and it needs to go to this area next. and it knows that there's slack space at these five places and it just puts it the one that's closest and he's like it's not it's it's it's almost like a packetized delivery network like a network like an ip network but for wafers

Which which like when he explained it that way, it made sense. So that's wild. Yeah, that's all wild. It was it was like being in Willy Wonka's chocolate factory, but less Hugh Grant and more foops. snozzberries taste like snozzberries yeah adam but they didn't let us taste the snozzberries adam didn't pick any usb keys off the floor either so i was looking yeah not that i wouldn't have yeah probably for the best yeah yeah

Is this a good place to ask some wrap up questions? Sure, let's do it. So. As somebody who's ready for a new desktop CPU, there's nothing here for me yet. Panther Lake is all mobile. Nova Lake, they're not even saying the words. There's nothing for me to grasp onto yet.

and they have confirmed no panther like on desktop right so right yeah okay they haven't said what's next but like i said i i believe that it's going to be an error like refresh at ces yeah but but as far as as far as like a fully next generation desktop part on 18a and stuff that's all that sounds like that's all a good ways off i mean yeah or maybe they skip 18a for desktop i don't even know i have seen i have seen the term 14a starting to float around

Yeah, so I actually don't know. Do you guys have any sense of like, how did Intel get on this leapfrog? Hadens where like Meteor Lake and Lunar Lake were mobile only and then like Arrow Lake got a desktop part and now Panther is mobile only again. Is it is it purely a function of laptops or what sell now and they just are not emphasizing desktop parts as much or what? No, they've done that in the past before, like Ice Lake was mobile only. OK. And that was around like.

10th gen yeah yeah so like they do that every once in a while and and i think that's the thing is that like where amd takes pride on sale scaling the same core from mobile to all the way up to to server. Intel usually has different platforms for different market segments. I think I think it's more like, oh, hey. Lunar Lake, we really wanted something that was power efficient, that had the NPU that could hit Copilot Plus.

We can't or there's no point in scaling this technology to a desktop part. Let's have something else do that. Like Intel, Intel kind of plays mix and match, which is why they're really leaning into Foveros kind of stuff. And so. Something that they've always said before, whether it's around NPU or why they got rid of hyper threading or why they've used TSMC sometimes.

Their own process sometimes is like, oh, we use the right part for the right time or, you know, something along those lines. So like they they they pretty much mix and match. So this isn't unprecedented to have architectures that.

that are on one area and not the other okay that makes sense the kind of feeling i wanted i got was that they knew that lunar lake was going to be really good early on like the performance was going to be there for mainstream parts it was capped at 32 gigs because the memory is on the on the chip um 32 gigs of ram but um like

So that bought them headroom to do more with this. Although this also still feels like a fairly concerned, like this feels like the small bump rather than the big bump. If we're looking at the kind of old TikTok metaphor that. Every time you mentioned while you were there, he's like, well, we really don't like that metaphor anymore, but this one's definitely a tick, not a talk. And we were like, okay, thanks. That clears this up a great deal.

I always thought TikTok was backwards anyway because isn't talk supposed to be the... Oh, no, let's not get down. Yeah, we got into that with the Intel people. People try to argue that the talk was the next generation and the tick was the evolved version. We're like, what are you talking about? That is backwards.

We had the stupidest conversation with a really smart person about this at one point. And I don't, I like at the end of it, everybody was like, I have regrets. Let's just never mention this again. Yeah. Let's talk about something else. Yeah. I think the thing is, is that for. for many years. I mean, like the way I step back and look at it when Ryzen when Ryzen hit, we had like we had like core wars of like, hey, how many cores can you get? Right. And I think we're we've.

I mean, there's maybe talk about whatever the next version might be able to scale up to more cores. But for the most part, we've settled at. There's a lot of cores. We have a lot of cores. We should probably figure out efficiency. We should figure out scheduling, things like that. So I feel like a lot of that has swung back in that direction. Lunar Lake was a big part of it.

Obviously, Qualcomm, you know, with their Snapdragon Elite stuff was pushing for it. AMD, you know, definitely continues to refine that stuff.

Yeah, I think that's what we're looking at for a good while. That's exciting. Yeah, refining efficiency, which is a big deal. And honestly, yes, there was a lot of stuff that... scratched heads around error lake on the desktop but if there's one thing you can say is that hey it was more efficient uh so yes maybe it wasn't the top of the gaming charts but it used way less power

and used less heat or produced less heat and like like yeah that doesn't look sexy on the back of a box but i mean that's i'm okay with that that's that's a big deal for me so i will absolutely i'll absolutely take like 90 of the performance

or like 60, 70% of the power consumption of the, of the high end, like the market leader, like a hundred percent all the time now. Yeah. Uh, like that's, that's, that's exciting to hear because those have been the pain points of the last few years, right? Like scheduling being a mess on.

on ryzens and intel it's everywhere through the roof like cpus have kind of not been in a great place this decade in a lot of ways to me and it sounds like these are the growing pains that needed to happen for these new paradigms to really take hold and mature Yeah, the big, like I asked Rajshree about the thread scheduler. One of the things I hadn't really fully understood is that that's a kind of conversation between the CPU and the OS, not just the CPU saying this because no one has complete.

a complete picture like neither side is a complete picture um and and her take was that they were pretty long way along the way of making that process like they've gotten most of the pain points out of that from their perspective also at the end of the day though and they they kept hammering home like they're they're just providing hints to the os yeah hints that's not a conversation

That's a hint. That's true. That feels very, yeah, like, oh, you know what? This is what we think you should do. But they did say at the end of the day, the scheduling is on the operating system. And because Will and I are Linux experts now, we actually did get to ask some questions of like, hey, how does the scheduling work on Linux versus Windows and things like that? Her answer was...

It really depends on which version of the of the kernel you're using, because the newer ones are pretty good and the older ones are kind of bad. So yeah, I can I can tell you as somebody who followed the kernel mailing list for several months watching Intel engineers. Send patches for for big little core scheduling. That's yeah. Yeah. So it was it was really fascinating. I have a better understanding of how this stuff works. I came out with.

The big takeaway for me is that they're still sticking with the big little cores. The little cores are getting more competent all the time. And the big cores are, you know. the like i used to think the big cores were the interesting thing now i'm kind of on team little core yep for the way that i work most of the time and like i almost given the opportunity i would probably buy an oops all e cores cpu yep

Before these weird hybrid architectures to simplify things. Yeah, like I'm very excited about the prospects of Wildcat Lake, which I know that's all rumors. You guys can't speak to that because obviously Intel isn't going to confirm anything. They didn't talk about any of that. Yeah, but the rumors about that, I'm looking at that for like a router or whatever and thinking like, could you just disable the P cores on that and just use the LPE cores? Because those are plenty fast now.

Well, and remember when Alder Lake came out and it was the first time they had the performance cores and efficiency cores. I can't remember who had brought it up, but somebody said that they equated the E cores to the performance of a 9900K P core. And I don't remember the generation of. It was like Skylake. That was the thing I always heard. The Alder Lake E cores were roughly the performance of the cores of a 6700K or a 6500K, which is not that old of a CPU.

I mean, that's what's in my stream machine, right? It's a 9900K. Yeah. And I did try to ask, and I don't remember the answer now. I'd have to go back and look at my notes. I don't have them right here. But I did try to ask, like, hey, what is the... what is the the processing grunt of an ecore right now equate to like how many generations back do we need to go to that's a great question of a p core knife i think he said alder lake p core no um he said

He said the IPC is Raptor Lake. Oh, no, no. I didn't write down what the answer was. I have your question. And I did not write down the answer because I'm a dope. What you're saying is basically which generation's P core is equal to the current generation E core. Exactly. Yeah. If I remember correctly, it was somewhere around like 12th gen. Wow. Dude, that's crazy. That is crazy. Yeah.

and on a smaller process yeah yeah no you know way more like the idea of a way more power efficient say like four core little box that is as fast as an alder lake would have been but with way way less power consumption like that we're really in like a We're a golden age for like home networking tinkerer types to buy a bunch of cheap, little powerful, low power boxes and do fun stuff with them. Well, and really like, and that's.

like imagine imagine chips that are all oops all low power e-course yes right like you actually have three sizes of of course that's that's that's the rumor of what wildcat let's if you look at all the wildcat lake leaks it's like two p cores four lpe cores and that's why i'm thinking can you just disable those p cores and just roll with a quad core that's like close to an alder lake apparently

So those LPE cores are all on the IO die, which is where the memory controller and all that stuff, the NPU and all that crap is. And they're clocked lower. Yeah. Next two, three years is going to be very exciting for new CPUs from the sound of things. I think so. Yeah. Well, except for at the end of the day, and this is what we're talking about, this thread scheduling thing. It all depends on the OS, because if Windows still has.

issues with scheduling. You know, there's only so much Intel and AMD and Qualcomm can do. I've heard there are other operating systems that might be able to help. Yeah, run cashing 2025. I'm talking about free BSD now. Year of Linux, baby. Actually, the year of the 3DSD desktop is coming. Yeah, I don't know. Well, actually, this is a real quick side before we get out of here. We had. A friend of the show, Wendell, from Level 1 Techs.

on Toolbook Diaries. Trust me, I listened to that episode. I love Wendell. I love Wendell. I have to speak to Wendell now because half the shit he was talking about is like extremely stuff I want to pick his brain about. You have to talk to Wendell. And one of the things he was talking about was that because, you know, he does a lot of consulting and stuff for corporations. That's kind of his main stuff. The YouTube, the level one stuff is just the fun side stuff.

Which is crazy, but he was saying that actually because of all this stuff that's going on with Windows and Microsoft kind of, you know, hitting the rake really hard that. Stepping on the rake or what is the race like stepping on the rug? Yeah. And then it just, uh, yeah, sorry. Uh, that, that, that actually, you know, it's not so much the year of Linux, which I mean, we are seeing.

some some play on that obviously but that he's seeing a lot of people move over to mac os that secretly this is actually this could be the year of mac os because that that's a better or that's a a safer place for people to learn in because they probably already have an iPhone. So they're like, oh, I hate this Windows stuff. I'm just going to go over to Mac. So what I'm saying is that we need to make a Mac podcast. Apparently. Oh, yeah. I use I use and quite enjoy a MacBook Pro.

I don't want to go back. It's for professionals. I don't want to go back. Mac OS is an excellent offer. My Mac OS is more Unix than Linux is. You're more low. Okay. Well, now I'm in. Yeah. Have you ever tried to load Linux on a Mac? Yeah, I've got Fedora running on my old. I have a very old 2011 MacBook Air that's running Fedora just fine that my girlfriend uses for light tasks that are pretty good.

We should have her on the Dolby Diaries. My Mac is an M1, which compatibility with the ARM Macs is a little bit more limited right now. There's a Sahi Linux, which... is a good chunk of the way there, but there's still quite a bit of hardware that doesn't work with our Macs and Linux yet. But again, I mean, whatever, we're getting way off topic here and you have to go, but like Mac OS has enough of the common POSIX Unix like stuff in it.

but I don't need Linux on a MacBook. Like Mac OS is good for that. It does all the same shit. Double Diary season three, baby. That's right. That's right. Let's talk about the Mac. Let's talk about my Mac OS package manager. Haiku year two. mac os year three um bsd year four year four yeah and then compile your own kernel yeah year five like getting to roll your own linux year five yeah perfect um great

Thanks for coming by, Adam. Always a pleasure to see you. It's nice to finally get to do a podcast with you. I've been a big fan for a long time and really appreciate your time over there. Easy world in the full nerd for folks who don't know. Cause I'm bad at telling them where can they find you and your work? Yeah. So, uh, I, I am the head of video at PC world and the full nerd. So go over to PC world's YouTube channel. Go over.

to The Full Nerd Network, which is a series of podcasts. We have The Full Nerd Podcast, where we talk about DIY desktop computing and all that fun stuff. I have a podcast with Will talking about Linux called the Dual Boot Diaries. I have a third podcast coming up talking about... tiny computing platforms so handhelds mini pcs laptops retro devices things like that so look forward to that uh is the retro stuff that we did with um i don't want to say who but

Is that going on Full Nerd or PC World? That's going on PC World. Retro Roadshow. Retro Roadshow brought a TI-99 4A and a B-Box, the PowerPC B-Box, BOS boxed by... So we could get hands on with those. And there's videos of both of those coming shortly. I just watched the newest cut.

yeah today is it good yeah they're turning out real good i think people are really going to like these historical deep dives i really got to nerd out on stuff i'd never well one of those machines i spent a lot of time with one of them i spent no time at all with and was really excited about so

But it started your career. Started my career. Yeah, exactly. Look forward to that. Anyway, thanks. Thanks, Adam. Thanks for coming by. And we'll see you. See you next week on The Full Nerd. Bye. Next time I'll have a beard. Hopefully. I hope so. There's no video here, so they don't know if you have a beard or not. Yeah.

Thanks, dude. Thanks for coming by Adam. Please check out the full nerd and double diaries. They're both, you know, I think they're good podcasts. I'm on both of them, but, but definitely go check out. the stuff on the PC world, YouTube channel. If you like computer stuff, I listened to DVD weekly at this point. It seems it's, I feel like for so many years, the rule was, Oh, you don't have to listen to your friend's podcasts. And then I never do.

Like I almost never do. Like I'm, I'm trying to think like I listened to the and or run of a more civilized age earlier this year. Yep. And now I'm listening to this and that might be it. Apologies to all of my other friends who do podcasts. but I think those are the only friend casts I have ever listened to. I listened to more friend casts when I'm in the car a lot, when I have a lot of commuting to do or when I'm like driving to LA or something, but, but yeah, for the most part,

It's it's AMCA and then a couple of tech podcasts. But if you if you like the show and you want to support the show, we are listener supported. We don't take money for ads. We don't take money other places. We only take money from you, the listener, because that's who we want to work for. And you can do that. You can give us money by going to patreon.com slash tech pod.

where five bucks will buy you access to the discord, which is full of nerds that talk about stuff like we talk about on the tech pod all day, as well as the monthly patron exclusive episode where this month, I feel like we recorded that yesterday. And and we just kind of talked about like Halloween decorations and other like we were kind of all over the place, but it was a fun rambly conversation. Fixing inflatable decorations. Yeah.

Probably some Linux in there. Homeschooling. Yeah. Yeah. Working on your house. Yeah. House projects. Yeah. That's probably up by now. Oh, yeah. Yeah. You can get that now. If you sub, if you sub, you get all of them. There's like 70 of them. If you want more tech pod. Yeah, so that'll do it for us. As always, thanks to everybody who subscribes at patreon.com slash checkpod. But a very special thank you to our executive producer, dear patrons, including Jason Lee and Felicitas Rips.

Andrew Slosky, Jordan Lippitt, celebrating 40 years of bunny crimes. Congratulations, bunny crimes. Twinkle Twinkie, David Allen, James Kamek, and Pantheon, makers of the HS3 high-speed 3D printer. Thank you all so much. Thank you. Uh, and that'll do it for us this week. We'll be back next week with another edition of the tech pod until then, please consider the environment before printing this podcast, which we also talked about last week.