Reader Mail for "October" Loose Ends
Added 2023-11-16 00:29:07 +0000 UTCIn the next 24 hours Tom will be doing the monthly Livestream! Within this episode you should expect more details about Nvidia SUPER, Blackwell, RDNA 4...and whatever you all ask him!
You have ~16 hours to write in below with your questions and thoughts, but be sure to be as concise and grammatically sound as possible to have your submissions considered. For completionism's sake, here is a list of the major stories from the past 4 weeks:
- Nvidia RTX 4090 Price Increases
- Nvidia 40 SUPER full Specs, Pricing, and Performance Leaked
- AMD's APU Roadmap for 2024-2025 Leaked
- Tons of Intel Meteor Lake Benchmarks Leaked
- Updates on Arrow Lake, Lunar Lake, and Panther Lake Release Dates
- AMD RDNA 4 Performance and Pricing Targets Leaked
Comments
Hi Tom, A hypothetical. Could Intel take a leaf out of Arm's book and operate something like a license model for x86, reasoning being their own design team seem to always be behind so why not outsource effectively?
coltmarshmallow
2023-11-16 23:17:21 +0000 UTCWhere do you see non-AI development in the coming months or years? It appears many things are being sacrificed at the altar of AI. AMD is dropping CoWoS supply for high-end RDNA4, in favor of solely using it for AI. AMD dropped Vega driver support for products like the 5700G released late 2021, and still being sold new today.(1) Nvidia did not have day one drivers for Starfield. Nvidia is finally adjusting prices (and price to performance) of GPUs as part of their SUPER branding, even though the 4080 has been the worst selling high-end GPU for a year.(2) And the AI development doesn't appear to be that great for AMD. Only a couple high-end consumer GPUs support ROCm (3). MI350X was cancelled (4). Projected Data Center AI revenue is promised to average above 500 Million a quarter in 2024 (5). That looks like a rounding error compared to Nvidia quarterly numbers. Are AMD and Nvidia going to moderate their singular focus on AI, or will more be sacrificed? If so, what? --- supplemental references and quotes: (1) https://www.tomshardware.com/pc-components/gpus/its-curtains-for-polaris-and-vega-as-amd-reduces-driver-support https://en.wikipedia.org/wiki/Radeon_RX_Vega_series https://www.amazon.com/AMD-Ryzen-5700G-16-Thread-Processor/dp/B091J3NYVF?th=1 (2) https://www.techpowerup.com/gpu-specs/geforce-rtx-4080.c3888 Release Date Sep 20th, 2022 Availability Nov 16th, 2022 (3) https://community.amd.com/t5/ai/amd-extends-support-for-pytorch-machine-learning-development-on/ba-p/637756 from 10-16-2023 11:00 AM “We are excited to offer the AI community new support for machine learning development using PyTorch built on the AMD Radeon RX 7900 XTX and Radeon PRO W7900 GPUs and the ROCm open software platform." https://rocm.docs.amd.com/en/latest/release/gpu_os_support.html (4) https://www.semianalysis.com/p/nvidias-plans-to-crush-competition from Oct 10, 2023 : "Due to competitive pressures from Google, Amazon, Microsoft, AMD, and Intel, we believe Nvidia accelerated their plans for B100 and “X100”. In response to this accelerated time schedules for Nvidia, we hear AMD completely cancelled their MI350X plans. Going back to the technical specifications we exclusively detailed for the configurations of MI300. The modular XCD building block is 40CUs and on TSMC’s 5nm process technology. AMD used to have the MI350X which had the same AIDs, but different XCDs, that were on TSMC’s 3nm. That part was cancelled for a variety of reasons, including that it would be completely uncompetitive with B100, when you compare the two’s on paper specs." (5) https://seekingalpha.com/article/4645677-advanced-micro-devices-inc-amd-q3-2023-earnings-call-transcript Lisa Su: "Based on the rapid progress we are making with our AI road map execution and purchase commitments from cloud customers, we now expect Data Center GPU revenue to be approximately $400 million in the fourth quarter and exceed $2 billion in 2024 as revenue ramps throughout the year. This growth would make MI300 the fastest product to ramp to $1 billion in sales in AMD history. I look forward to sharing more details on our progress at our December AI event." "December AI event" is a reference to Dec 6th event: https://www.amd.com/en/corporate/events/advancing-ai.html
Nicholas Buckner
2023-11-16 20:21:17 +0000 UTCHi Tom, with Meteor Lake seemingly being a much more complex design compared to Phoenix/Hawk Point, do you expect Laptops equipped with Meteor Lake SoCs to be noticeably more expensive than AMD-equipped counterparts? Do you think Intel will have to accept smaller margins than AMD again to sell well?, given that it doesn't look like it will outperform the competition by large amounts, if at all.
KingKoro
2023-11-16 19:47:39 +0000 UTCHey Tom, after the reveal of the new Snapdragon X Elite SoC for Windows, do you think we'll soon see a shift in PC hardware? Maybe an SoS with soldered on Memory and only expendable Storage and a GPU slot? I mean even the X86 giants are seemingly more and more interested in ARM and RISC-V.
Dave Scholze
2023-11-16 17:39:27 +0000 UTCHi Tom, do you anticipate that the 5090 will have 36GB of GDDR7, or is the 90-class destined to max out at 24GB of VRAM with higher memory amounts reserved for Pro cards?
GrandDemand
2023-11-16 11:14:00 +0000 UTCI tried to look for it but I can't remember exactly. But it was around the time that RDNA2 drivers weren't being released and people were getting nervous. You mentioned they invested most of their resources to fixing RDNA3 drivers and even pulled some people out of other departments. You surely remember RDNA2 driver situation. In this video you mentioned some of the problems. https://youtu.be/c5ufWyH3XbE?si=-Wftj5XPDVKJYkIt
QuickJumper
2023-11-16 08:51:58 +0000 UTCCan you link to that? No offense, but it's becoming a little too common place for people to vaguely state something I've said a while ago, and expect me to know what they are talking about. And I am not saying this is necessarily the case now - but often times half of the details are forgotten by the people telling me to remember something... :)
Moore's Law Is Dead
2023-11-16 07:53:53 +0000 UTCHello Tom you mentioned before RDNA3 launch they needed to get help from other departments in the company just to be able to get RDNA4 out of the door. What's the current situation with the staff? Will they enough people to launch RDNA4 without problems? Do you maybe think that they eate of innovation is also limited by their staff?
QuickJumper
2023-11-16 07:39:29 +0000 UTCHow do you think they will name Strix Halo and Fire Range given how similar they are. Assuming top Zen 5 will most likely be 9955HX for 2025
Benjamin Cannon
2023-11-16 03:55:25 +0000 UTCIs Intel really just gonna be in a perpetual state of being leap frogged until 2026/2027? How do they even fix this? Like you've mentioned before by the time they have their answer to Zen 6/Zen 7, we might even be in a completely post Zen era, not to mention Qualcomm are getting into the PC market and I bet Nvidia will be as well.
CompressedAIBlocks
2023-11-16 02:45:12 +0000 UTCTom, what is AMD's ultimate goal with RDNA 4? You've mentioned they might be using this to make sure they get their software on point otherwise there's no reason to release high end/ultra high end/omega high end to consumers. Is this ALL they're using this generation for? Is this generation meant to be aggressive with market share or will it be another sort of thing like "Yeah, we're here. We exist."? I'm not disappointed at the performance or pricing you've leaked from RDNA 4 but to the masses (which are the majority of the market) it just feeds into the "Radeon just isn't as good GeForce" narrative and I dislike that for them.
CompressedAIBlocks
2023-11-16 02:42:39 +0000 UTCHello Tom, do you think AMD needs a complete rebrand of their GPU division? I see this discussion being brought up a lot and would like to know your thoughts. You've mentioned that Intel would do well to rebrand Arc and I do not disagree with that but I'm also wondering if this would help AMD at all, or if you think them executing an entire stack without software and hardware fails 3+ generations in a row would be enough. Every time they do well one generation, they seem to fumble a bunch of that good will they earned the next. As someone who mostly uses Radeon GPUs (as in 80% of my GPUs have been Radeon over the past 2-2.5 decades) it's a bit tiring seeing this again and again.
CompressedAIBlocks
2023-11-16 02:40:03 +0000 UTCWhen the 4080 Super comes out at $999, will the standard 4080 still be in the lineup?? If so, its MSRP would need to be reduced by quite a lot. Any AIB’s with leftover 4080 stock would surely be pretty upset, since they would be forced to drop prices to… below their cost?? ( A 4080 currently at $1099 now must drop to $899? ).
JR
2023-11-16 02:00:07 +0000 UTCHi Tom. How likely is it that Nvidia will screw AMD's upcoming laptop APU releases up by paying companies to add Nvidia GPUs to AMD laptops, destroying the entire point of having a powerful laptop APU in the first place (lower price + better efficiency)? If likely, will AMD try to fight it? What can we do about it? I ask this because I am quite concerned by how difficult it is to find a 7940HS laptop with no discrete GPU. What's more ridiculous is that they're paired with 4050s and 4060s.
David
2023-11-16 00:59:39 +0000 UTCHey Tom, With Black Friday around the corner, are there any parts that your sources say are in short supply AND are worth getting if there's a good deal?
Cleansweep
2023-11-16 00:38:48 +0000 UTCWith the upcoming Super series, part of me wonders how much sooner it would have released if RDNA 3 lived up to the 100% raster increase hopes. Would it be possible that AMD could invest heavily in RDNA3 driver development paired with a marketing campaign to combat the Super series release? Or are the RDNA3 margins the only thing they will try to leverage with a price war? Maybe the hopes for strong fine wine this gen is just copium.
Holden Mobley
2023-11-16 00:35:36 +0000 UTCHi Tom, MSI laptops with meteor lake have already leaked with leaked pricing. You had mentioned oems are buying up phoenix due to the lateness of the release. Since strix is a drop-in upgrade, can we expect broader OEM support coming soon or is it wait for 2025?
B. Fish
2023-11-16 00:32:18 +0000 UTC
