Discussing Nvidia Chiplets, RDNA 4, Meteor Lake w/ High Yield
Added 2023-08-19 04:07:11 +0000 UTCBroken Silicon's next guest will be High Yield! We plan to heavily discuss the state of the GPU & CPU markets, Intel's future, and everything having to do with chiplets and GPU Design! This is a great episode to ask whatever you want about GPU Competition and the Future of these companies!
Our conversation will be HEAVILY steered by your submissions below, so don't hold back! Just make sure you comments and questions are respectful, well written, and as concise as possible to be considered. You have ~36 hours to submit below!
https://www.youtube.com/@HighYield
Comments
For Zen 5, AMD seems set to have 3 variations on their CCD chiplet - a standard 8-core, the same with V-cache and a “cloud” type16-core. Let me throw out another possibility - a chiplet with 4 standard cores and 8 “cloud” cores, basically the cores on Strix Point as a CCD. Let’s say AMD offered to make you for free a Ryzen processor with two of any of those four CCDs. What combination would you pick and why? What combination do you think would be most popular with consumers?
Chris Rijk
2023-08-20 15:09:59 +0000 UTCMeteor Lake seems to only support one CPU tile and one GPU tile. Do you think it would have been better if it could support two CPU tiles? Overall, what do you think are the strengths and weaknesses of Meteor Lake, particularly from the chiplet side of things?
Chris Rijk
2023-08-20 15:07:16 +0000 UTCSame. That 7700x combo deal for $400 is excellent
FloridaMan
2023-08-20 13:15:29 +0000 UTCAMD with zen4c seems to have followed Apple's route of sacrificing 1t performance for efficiency, do you think that x86 will move in this direction as a whole or do you think 1t is just too important to leave by the wayside?
qhfreddy
2023-08-20 11:39:11 +0000 UTCJust wanted to say thanks for coming on Broken Silicon!
Benjamin Cannon
2023-08-20 06:12:32 +0000 UTCWhen it comes to Intel's x86 processor's across all their segments, do you think their biggest weaknesses are in their cores or the interconnects?
qhfreddy
2023-08-20 03:45:08 +0000 UTCHow do you view demand for high end (large die size or high complexity) consumer GPUs? What about more generally in terms of desktop/workstation GPUs with display outputs? Do you think there will still be a place in the market for RTX4090 and 7900XTX like products in AMD and NVidias lineups in 5 years?
qhfreddy
2023-08-20 03:41:39 +0000 UTCFor the PS6 or the successor to the Xbox series X, would you be surprised if they have a chiplet design, or would you be surprised if they don’t? What about on-package main memory?
Chris Rijk
2023-08-20 01:11:34 +0000 UTCFor x86 processors, what do you think Intel and AMD could learn from each other, particularly with regards to chiplets?
Chris Rijk
2023-08-20 01:11:13 +0000 UTCDo you think price and price/performance are taken seriously enough? When people talk about Moore’s Law they’re quick to point out the performance improvements but the cost improvements are rarely mentioned. If the price/performance of silicon only slowly improves and chip designers don’t try to mitigate against that then consumers will see a slow rate of improvement and will respond by buying less frequently or not at all. That leads the market to shrink, which isn’t good for anyone.
Chris Rijk
2023-08-20 01:11:01 +0000 UTCHey HighYield, your video explaining Meteor Lake was incredibly informative! Do you think Intel will try to leverage this new tile architecture to convince Nvidia to start manufacturing with them? We know TSMC is very secretive with it's knowledge. Maybe intel will say, "We have this tile design that is proven to work in Meteor lake and beyond. And we will make sure your chiplet designs work too." Thanks!
Jamwap
2023-08-19 18:38:33 +0000 UTCHas AMD jumped into GPU chiplets too early? They seem to hold a price/perf advantage no better than with RDNA 2, while offering problematic performance. Nvidia meanwhile seems fully content with staying monolithic for quite a few years yet.
KingHarkinian
2023-08-19 18:37:56 +0000 UTCSRAM has been a big focus of Max's videos for a good while now. Both of you care to give a rundown of how you think the SRAM problem will be circumvented in the future? Or are we doomed to be SRAMless?
KingHarkinian
2023-08-19 18:30:04 +0000 UTCHi guys, I enjoy both your channels, keep up the great work! I was wondering what is your take about the Intel X86S and if this will become a thing? Reducing the amount of legacy instructions sets can benefit us long term and as technology moves forward I think it’s necessary to cut out old unused tech and not carry over legacy stuff in order to really advance. Cheers from Israel.
Roy Ackerman
2023-08-19 16:13:28 +0000 UTCWith RDNA 3 vs Lovelace, Nvidia has downsized their dies to a scandalous degree (AD107 for 4060, 4070 Ti as 4080 12Go, etc), while AMD has pushed for everything Navi 31 could do with the XTX. Yet Nvidia is clearly holding the line. 4060 vs 7600, 4080 vs XTX. Despite putting so little effort and selling for a grossly inflated price, Nvidia seems to deal with AMD without a hitch. Is this a generational problem? Is this significant for the future? Is AMD's best failing against Nvidia's middling, cut down effort, a writing on the wall or just one pothole on the way?
KingHarkinian
2023-08-19 16:00:50 +0000 UTCHi Tom and guest. I was bit surprised by the complexity of RDNA4 which Tom leaked. I was maybe expecting AMD to hit pause button and try to iron out their current design - GCD+MCD . But maybe I am looking at it backwards. Maybe this initial design will be more complex at first but then it will be very similar to ZEN or MI300. You could just design one or two things and "glue" more complex design together. How do you look at innovations that AMD is doing with RDNA4. Do you think that Nvidia is cooking something similar or are they maybe bit behind AMD?
QuickJumper
2023-08-19 13:59:48 +0000 UTCHello Tom and guest, who do you see being the next player in GPU? For example Moore Threads a Chinese company? Do you foresee a new player being able to enter the market? What would be the biggest hurdle to overcome?
Bryan P.
2023-08-19 13:33:25 +0000 UTCHi guys, high yield, love your channel. I had a laptop question building on Tom's recent leak about AMD going after market share. What are you hearing from the OEM side? It seems to me that and has had great products in theory but design, configuration and availability fall flat. Cezanne was great, but Rembrandt and phoenix really took a step back.
B. Fish
2023-08-19 12:58:51 +0000 UTCHey guys, I have a intense discussion question for you. How much of the decision to cancel high end rdna4 was because of the incredible chiplet based battlemage MONSTER Intel is planning to dominate enthusist gaming. Are we talking 90% or 100%?
Swiggles
2023-08-19 12:44:58 +0000 UTCHello, The GPU market seems endlessly supply constrained. Jensen said it would be so for the next 10-15 years back with Turing. But right now, it's riding on AI more than anything. While they're great for HPC, they seem to be repurposed for AI rather than actually fit for purpose. Do you think that highly specialised AI chips will come to market and succeed? Do you think the AI market will disassociate from the GPU market in the future?
KingHarkinian
2023-08-19 10:43:08 +0000 UTCDo you think Intel will finally support its platforms for longer than just two or (since Raptor Lake refresh) three generations?
Dave Scholze
2023-08-19 09:00:01 +0000 UTCNvidia Blackwell is rumored to have a 512-bit bus on the top configuration but they have been on 384-bit for top configuration even on the workstation for nearly a decade. Another recent rumor was shifting the gaming lineup naming to GB20X rather than the expected GB10X, what are the implications of these changes and why is Nvidia choosing to do this now if true. Is it possible that if the AI boom continues, Nvidia will make the 384-bit chip the 5090, reserve the big one for the far more lucrative datacenter and maybe some halo Blackwell Titan monster. Intel Arrow Lake is an interesting one since it's in this weird transition where they couldn't get HT to work and rentable units isn't out yet (if I have this correct). Many people are going to be asking whether they should just upgrade to Zen 5X3D (when it's out) or waiting a little longer for Arrow Lake. With Zen 5 looking like a big performance increase and Arrow Lake even bigger one but with no HT, what do you guys think about this?
ferdock
2023-08-19 07:40:04 +0000 UTCAlso, I work in this space - happy to offer my understanding if it's helpful.
Christopher Ricks
2023-08-19 06:45:32 +0000 UTCI've been following and supporting both channels for a while. That said, High Yield focuses on specifics and gets basic units right. Tom needs to listen - the units for die area are **square millimetres**. Your house isn't offered as "feet squared" when it's on the market. High Yield also focuses on data instead of speculation - find a middle path please.
Christopher Ricks
2023-08-19 06:44:45 +0000 UTCMoin high yield, Big fan of your channel! What do you guys expect would be AMDs most likely foundry alternative, I.e. in addition to TSMC: Samsung or Intel? I can’t really see how they would get over the obvious concerns working with Intel foundry, even IF they proved to be competent and reliable. On the other hand Samsung has no interesting node with FinFETs and the first gate all around node seems to have issues with yield and performance based on how late it is compared to Samsungs road map. But say one of those were in the race: what would they most likely second source? My guess is only Monolithic chips like laptop or low end GPU due to their advanced packaging products (basically all :D) relying on TSMC proprietary tech. But that’s fine as those are both products where they might need significant volume in order to serve upwards of 30% market share.
dj5k
2023-08-19 06:33:52 +0000 UTCHi all, Assuming that Intel can deliver their products on time (and drivers are not a problem), what features does Intel in their processors in order to stay relevant? For example, strengthening iGPUs (Alchemist and Battlemage tiles), adding in more coprocessors (AI engine, decompression chip), or maybe adding higher amounts of cache (Adamantine L4 cache)?
Woody Chang
2023-08-19 05:30:07 +0000 UTCHi Tom and Guest, do you think it's possible that in the next several years we may see AI take a more key role in the rendering portion of game engines? This year at GTC NVidia demoed Neural Radience Caching, which is something like DLSS but for RT rays instead of images. And at Siggraph NVidia also demoed Neural texture compression (https://research.nvidia.com/labs/rtr/neural_texture_compression/) Are either of you aware of any other such tech coming down the pipe? Could we possibly see future GPUs have an even bigger silicon budget for AI?
Timothy Baldridge
2023-08-19 05:18:37 +0000 UTCNews of the high end of rDNA 4 being cancelled and Nvidia being Nvidia got me to thinking: microcenter has 7900xt reference editions for $750 with a free copy of starfield. That makes it effectively the launch price of the 6800xt. While yes, we might get that level of performance for less in 2 years, is it me or is that almost too good a deal to pass up? Also thanks microcenter, I've successfully talked myself into 7700x for $300 with free ram, new case, new motherboard and now new GPU because the prices and selection are just too good.
B. Fish
2023-08-19 04:43:41 +0000 UTCHi Tom and Max?, what are your opinions on AMD canceling their higher end dies for RDNA 4? Did they try to make it too complicated too soon, was it another issue? Do you think consumers will be happy with another RDNA 1-like generation? Cheers!
Xavbeat03
2023-08-19 04:30:35 +0000 UTC
