Discussing Meteor Lake, Zen 5, Arrow Lake, GPUs w/ Level1Techs
Added 2023-09-29 21:14:16 +0000 UTCThe next episode of Broken Silicon will see the return of Wendell from Level1Techs! The past couple of months have seen tons of AMD and Intel CPU information drop both through leaks, but also through Hot Chips Presentations and Intel's Innovation 2023 event - let's talk about it!
Planned topics include:
- Intel's Next Gen Packaging Techniques
- Intel Meteor Lake Competitiveness (Launching December)
- Intel Adamantine & other Future Competitors to AMD V-Cache
- Intel Emerald Rapids Competitiveness (Launching December)
- Zen 5 & Zen 6 Thoughts
- AMD Turin vs Intel Granite Rapids / Sierra Forest
- Arrow Lake Expectations
- AMD RX 7800 / 7700 XT
- If Nvidia & RADEON need to drop prices on more GPUs for the Holidays
As often happens - this episode will be heavily steered by your submissions below! Please feel free to ask about anything having to do with AMD, Intel, Nvidia, Apple, PlayStation, XBOX, or anything else even remotely relating to gaming or server hardware! Wendell can handle it all!
You have ~40 hours to submit below! Be concise, respectful, and use good grammar for consideration.
Check out Level1Techs: https://www.youtube.com/c/Level1Techs
Last Time Wendell was on BS: https://youtu.be/7thf6CRILQk
MLID Zen 5 / 6 IPC & Release Date Leak: https://youtu.be/ueXUoRw5LZk
MLID MTL Benchmark Leak: https://youtu.be/H8ryAvX63Tc
Level1Techs 7800 XT Review: https://youtu.be/AdWv3GwS-kY
Wendell's recent thoughts on Intel announcements: https://youtu.be/r-lyUAxybxM
https://www.intc.com/news-events/ir-calendar/detail/20230919-intel-innovation-2023
Robert Hallock Meteor Lake Interview: https://www.youtube.com/live/gb58WV5RHRg
Comments
Hi Tom & Wendell This is not quite related to Adamantine, but I guess I could chuck it in. I recall recently (a few weeks ago) seeing an image of what supposed to be an Intel mobile cpu with LPDDR on substrate, similar to an Apple Silicon. Given most ultra book these days already has soldered memory, when would you see Intel / AMD doing such packaging for Laptop? I supposed that would greatly help with design as there won't be worries on how to run the memory trace, while potentially reduce the size of the motherboard.
Dark Side of the Force
2023-10-01 23:06:54 +0000 UTCSeems to me that when engineers don't shake the house much, they face little hassle and end up with a good product. When they do, it gets more difficult to deliver since they've changed a lot. Zen 5 is the final child of Zen 2's design. Zen 6 will renew a lot of things and be a new era. What do you expect out of each?
KingHarkinian
2023-10-01 20:20:16 +0000 UTCHello and welcome back to my favorite plumber in the PC space. I just wanted to get your thoughts on AMD's more recently aggressive push into AI with Radeon. I am wondering if you have noticed any parallels between what they have been doing with LLM's and Pytorch versus the rate at which new software features are being added to the gaming hardware and software. One more question, what are your thoughts on Smart Access Video and have you tested it yet from the newest preview driver? Thanks and a pleasure as always!
MelodicWarrior
2023-10-01 20:01:58 +0000 UTCHi Wendell, in your review of the 7800 XT you called it “Your Next 5-year GPU!” and you highlighted the VRAM capacity and longevity. Despite all the fuss made about VRAM usage on the latest games in the first half of the year, I didn’t see another review that highlighted VRAM capacity. So, thank you for doing that and please keep on doing it. https://www.youtube.com/watch?v=AdWv3GwS-kY
Chris Rijk
2023-10-01 17:33:49 +0000 UTCFor PC gaming on desktop and laptops, how do you think the market will look by the end of 2024? Will the NUC surge continue? Will Meteor Lake reshape the market? Will PC derived handheld consoles go anywhere (eg ROG Ally and Lenovo Legion Go)? Is the market ready for a processor like Strix Halo?
Chris Rijk
2023-10-01 17:01:25 +0000 UTCHi Tom and Wendell, here are some discussion points that are floating in my head for some time, maybe you can clarify some of them. Emerald Rapids: In comparison to SPR, do you think EMR will bring huge efficiency gains due to architecture and manufacturing node improvements? Do you think Intel will offer its accelerators (QAT and IAA) more broadly with EMR, or will they still be limited to selected SKUs on its server platform? Also, is there any idea when the EMR release for Workstation will be? And last but not least, what do you guys think about EMR's competitiveness against AMD Genoa and future Turin? AMD Zen stuff: Genoa is huuge, so huuuuge, I need one. What are your thoughts about the probability of AMD launching SP6 based EPYCs with normal Zen4 ccds? In connection to the other question, I saw that the new Threadripper Workstation will be mainly on SP5, thus the question are there any SP6 Threadrippers planned? Also, do we have any chance of having HEDT return for AMD? Do you have any idea why AMD waited so long for the workstation platform to be released? When do you think Threadripper 7000 will be available to buy for normal people (non OEMs)? And a question directly to Wendell, do you know if SP3 Heatsinks fit on SP6 sockets, it should be possible, no? For Zen 5, what do you think they meant on the leaked slide with FP-512 variants? Like custom ccd designs for special workloads or just better binned SKUs, like Intel does with some its XEON Scalable SKUs. AMD MI Instinct Stuff: Wendell, do you have any idea why the software (ROCM) support life span of AMD accelerators is so short compared to Nvidia? For the Instinct MI50, it is a little over 5 years (End of Maintenance is Q2 2024). In comparison, NVIDA still actively supports Maxwell era Teslas from 2015.
Josie Herzog
2023-10-01 13:15:30 +0000 UTCHey Tom and Wendell. Question for Wendell: Which do you think is the stronger long term competitor to CUDA in the AI space: ROCm or OpenVino? AMD's been making strides with ROCm, but is that enough to keep Intel from getting everyone to jump on their platform, especially if they offer better NPUs for sale to developers and end users?
Cleansweep
2023-09-30 20:21:25 +0000 UTCHello there, I was wondering what is your opinion on the future of optane?
joryuken
2023-09-30 20:13:15 +0000 UTCHi Wendell - Zen 6 EPYC is expected to use a new socket, SP7, which is rumoured to support 16 channels of DDR5 memory. When you get that big is there any point in doing dual socket systems? Would you be happy to trade off dual socket support for more I/O? Would you like to see AMD offer EPYCs processors with their Pensando networking technology integrated?
Chris Rijk
2023-09-30 17:18:17 +0000 UTCWhat do you think about the current rate of progress in computer hardware? Where do you think the most interesting battles are? Where do you think things have stagnated?
Chris Rijk
2023-09-30 16:39:52 +0000 UTCHey Wendell, Here's kind of a broad question. From your point of view since, as you say, the consumer side is the "leftovers" from server/datacenter innovation, what is the 10 year projection of what consumers will see taking into account the progression of server and data center? AI in everything? More cores or less cores but more powerful and how will memory fit into that equation? I've seen new memory designs like Ultra RAM and new LPDDR and SO DIMM concepts. Just wanted your take on all this. By the way huge fan and no Tom this isn't a hit on you, as I know you give us great foresight as well just wanted to hear his take on the matter.
GaminSinceReagan
2023-09-30 16:17:58 +0000 UTCHey Guys, with Zen 6 having a new architecture, do you think AMD might make AM5 short lived and already move to a new platform?
Dave Scholze
2023-09-30 14:43:42 +0000 UTCThere will be plenty of core count and performance talk about both Intel and AMD server platforms. My question is what do you see as the two most underrated features or over looked use cases for each of the platforms?
Mark Mitchell
2023-09-30 12:46:20 +0000 UTCHow many years behind AMD do you think Intel is in terms of packaging and MCM/chiplet designs? Do you think it's even possible to assess the state of their competition in terms of one being behind and the other being ahead on a timeline?
qhfreddy
2023-09-30 11:12:30 +0000 UTCThoughts and experiences running Radeon and Geforce GPUs? What about the professional SKUs?
qhfreddy
2023-09-30 11:12:25 +0000 UTCHave the arm shills finally lost? Where are they now, shilling AI? Are people finally coming to terms with the fact that ISA is pretty meaningless in the grand scheme of things?
qhfreddy
2023-09-30 11:10:02 +0000 UTCWendel seems to have the opposite opinion of Tom and CarbinCry regarding Software advantages for Intels and AMDs AI acceleration. That could be included as well. Also Siena...
crumbling_earthblocks
2023-09-30 08:29:30 +0000 UTCIn the recent MLID Zen 5/6 leak, Zen 5 is said to move to a 16 core complex (CCX), although this doesn't appear to be leveraged on Desktop and seems likely to only see usage in Zen 5c. What workloads are actually likely to benefit from a Zen 5c CCD now having a 16 core CCX vs the 2x8 CCX of a Zen 4c CCD?
raysin bisket
2023-09-30 00:16:01 +0000 UTCIn the recent MLID Zen 5/6 leak, Zen 5 is shown as having a "Low power core option". Is this just a renamed dense version (so Zen 5c), like is listed under Zen 4? Or does this refer to something fundamentally different?
raysin bisket
2023-09-30 00:12:29 +0000 UTC[Copied from Discord Reader Mail] Hi Tom and Wendell, Intel is moving from monolithic which enables lower latencies to a disaggregated architecture which will inevitably increase latencies. The recent MLID Zen 5/6 leak suggests that AMD with Zen 5 is addressing some fundamental architecture 'bottlenecks', and with Zen 6 is optimising their chiplet design to approach/achieve monolithic levels of performance, improving the infinity fabric and lowering die-to-die latencies. From a purely performance perspective, given how competitive AMD is to Intel currently, this transition seems like a slam dunk for AMD. What are your thoughts? Can Intel counter this? If so, how?
raysin bisket
2023-09-30 00:05:31 +0000 UTCHi Tom and Wendell, Wendell - I know you're a big fan of CXL. Is CXL a technology that can benefit consumer products (desktops/laptops), or for the foreseeable future is it a technology reserved for datacenters, HPC and the most extreme Homelabs? Consumer-space aside then, what potential uses of CXL are you most excited for?
raysin bisket
2023-09-30 00:04:52 +0000 UTCHello, It seems like a large amount of SPR's sales come from its AI performance and accelerators. Turin seems to be cooking the same kind of accelerator in the very near future, do you think that they will become commonplace in server in the future? How about in consumer CPUs for inference?
KingHarkinian
2023-09-29 23:57:58 +0000 UTCThoughts on Sapphire Rapids and Emerald Rapids workstation parts? These were basically not sampled to the press and I asked HUB Steve a while back and he said he didn't have much interest in benching them. Can you maybe give some insight into the sampling and reviewing of these parts? Do you think they have a place in a mixed workstation-gaming build?
qhfreddy
2023-09-29 22:21:39 +0000 UTCHello Tom and Wendell. Wendell you have been a strong proponent of VFIO/GPU pass through. What other server technologies do you think major companies need to focus on bringing to the consumer space?
BaconHouse Collective
2023-09-29 22:14:39 +0000 UTCHi Tom and Wendell, given the success of the PS5 dedicated IO chip to handle decompression of game assets, do you think we will see Amd and/or Nvidia add dedicated hardware to there gpus to handle decompression of assets instead of using there shader cores? If so do you think they will do it with a “static” asic chip and support a finite number of compression types, or maybe implement and fpga chip that would in theory support any compression algorithm?
Josh Audibert
2023-09-29 22:11:04 +0000 UTCCould Zen 5 vs Raptor Lake Refresh could actually be worse for Intel than Zen 3 vs Comet Lake? Thanks to Tom's leaks, I think Zen 5 could win almost all benchmarks while also being cheaper than Zen 3, with RPR-R being more expensive than Comet Lake. Let's hope that Intel can make a comeback with Arrow Lake.
Chris Rijk
2023-09-29 22:07:06 +0000 UTCHey Tom and Wendell, do you think that the current AI boom is a bubble? How does it compare to the crypto bubble? If you had to make a rough guess, what do you think the Ai market looks like in 2 years?
Xavbeat03
2023-09-29 21:43:45 +0000 UTCHi Tom and Wendell, what are your views on the overall health of the Server market? Is there a company that you think has too much power/market share? How do you view Nvidia's massive success with hopper affecting the server market in the medium term(2-3 years).
Xavbeat03
2023-09-29 21:42:12 +0000 UTCHi Tom and Wendell, what are your latest thoughts on the HEDT platform market, is it dying? Are the consuner platforms fast enough and have enough features that the segment won't survive in the long run. As actual workstation platform budgets are out of reach for hobbyists.
TommiP
2023-09-29 21:38:54 +0000 UTCWendell. What are your expectations with Emerald Rapids? Does this move the needle at all for Intel in the server space or does it just allow them to hold onto what they have?
CompressedAIBlocks
2023-09-29 21:32:31 +0000 UTCHi Tom and Wendell. I'm a proud patreon of BOTH Level1Techs and Moore's Law is Dead. What sort of AI features would you like to see in AMD's future GPU/CPU products and is there any insights on when AMD will have the full ROCm SDKs on Windows? We are missing things like MI Open which Py torch needs to work in Windows. I'm loving the ROCm powered llama text inference I get but I want more. Thanks for BOTH of your channels and the great work you guys do.
Joseph Kelly
2023-09-29 21:31:01 +0000 UTCHello Tom and Wendell, I hope that you are both doing well today. Wendell could you possibly confirm whether or not you work in hardware or software? Or is it both? Thank you guys, I appreciate your time and I can’t tell you how much joy your content brings me.
Shallon Ogden
2023-09-29 21:25:18 +0000 UTCT&W. Is Meteor Lake poised to maintain or even take more mobile market share from AMD? AMD seems to have a supply/capacity constraint with less robust OEM support, will Meteor Lake only widen this gap?
CompressedAIBlocks
2023-09-29 21:22:52 +0000 UTCHi Tom can you please specify the timeline in which zen 6 will be released compared to the arrow lake? Will Zen6 beat arrow lake in performance and if yes by how much? (What is the estimate from you and Wendell?)
QuickJumper
2023-09-29 21:22:20 +0000 UTCHow are the non-x86 server software stacks compare? Especially ARM and RISC-V?
KarbinCry
2023-09-29 21:21:34 +0000 UTCHow do you guys think Adamantine will stack up versus AMD's V-cache? Are there any inherent advantages that Intel will have? Capacity, speed or latency? Will we see see a future Intel product akin to the 5800X3D/7800X3D?
CompressedAIBlocks
2023-09-29 21:21:20 +0000 UTCTom & Wendell. How do you guys feel about the GPU market after the launch of Navi 32? Do these hit the sweet spot gamers wanted and more importantly will this anchor future mid range pricing that is accepted by PC gamers?
CompressedAIBlocks
2023-09-29 21:19:56 +0000 UTCDo you ever plan a follow up on your court video? It's old, but lots of fun!
KarbinCry
2023-09-29 21:18:29 +0000 UTCIs 2024 going to be the year of Linux in desktop?
KarbinCry
2023-09-29 21:17:50 +0000 UTCCan you explain where Sienna is useful? Why not just use lower core count SP5 SKUs?
KarbinCry
2023-09-29 21:17:09 +0000 UTC
