6 ms·
Kepler, Nvidia's Strong Start on 28 nm
- behnamoh 3y ago[flagged]
- jeeyoungk 3y agoI don't think 2.8T market cap company combining with 1.1T market cap company can be called an "acquisition", nor three letter agencies would ever approve such a deal.
- varispeed 3y agoI hope not, given Microsoft's typical strategy of: Embrace Extend Extinguish It also means less competition. Corporations like Google, Microsoft, Amazon are embodiment of what is wrong with late stage capitalism and the result of lack of regulation. If anything Microsoft and corporations of similar size should have been broken up decades ago. They have too much power, they don't answer to anybody and are a form of unelected government that rules digital world.
- amelius 3y agoMicrosoft might start making their own silicon like everybody else.
- jmisavage 3y agoThey have the Azure Maia 100 (AI accelerator) and Azure Cobalt 100 (128 core ARM Neoverse derived cpu). https://www.theverge.com/2023/11/15/23960345/microsoft-cpu-gpu-ai-chips-azure-maia-cobalt-specifications-cloud-infrastructure https://www.theverge.com/2023/11/15/23960345/microsoft-cpu-g...
- latchkey 3y agoAMD isn't asleep. MI300x announcement on Dec 6th.
- graphe 3y agoHow is AMD addressing CUDA dominance?
- depereo 3y agoInvesting into making their stuff work with the abstracted libraries. Pytorch etc. Collaboration with Microsoft to make their hardware easier to get time on and test with. Starting to support tooling on their lower tier consumer GPUs.
- latchkey 3y agoThere have been several announcements: "Azure announces new AI optimized VM series featuring AMD's flagship MI300X GPU" https://news.ycombinator.com/item?id=38280974 https://news.ycombinator.com/item?id=38280974 Doubling down on ROCm: https://www.theverge.com/23894647/amd-ceo-lisa-su-ai-chips-nvidia-supply-chain-interview-decoder https://www.theverge.com/23894647/amd-ceo-lisa-su-ai-chips-n... Putting resources into pytorch: https://pytorch.org/blog/experience-power-pytorch-2.0/ https://pytorch.org/blog/experience-power-pytorch-2.0/
- latchkey 3y agoBuild a developer flywheel by making ROCm more accessible. https://www.tomshardware.com/pc-components/gpus/amd-arms-three-of-its-gaming-gpus-with-pytorch-and-rocm-support-for-ai-development https://www.tomshardware.com/pc-components/gpus/amd-arms-thr... That said, I'm approaching this from the other end, by make their high end GPUs available to developers.
- jewel 3y ago[flagged]
- treesciencebot 3y ago> November 24, 2023 It was just published a couple hours ago. Chips and cheese usually go back in time and post deep dives into old chips, not everything has to be cutting edge to get an insightful article.
- titaniumtown 3y agoNo it's not.
- scrlk 3y ago> Nvidia’s Fermi architecture was ambitious and innovative, offering advances in GPU compute along with features like high tessellation performance. However Terascale 2’s more traditional approach delivered better power efficiency. Fermi was given the nickname "Thermi" for a good reason. AMD marketing had a field day: https://www.youtube.com/watch?v=2QkyfGJgcwQ https://www.youtube.com/watch?v=2QkyfGJgcwQ It didn't help that the heatsink of the GTX 480 resembled the surface of a grill: https://i.imgur.com/9YfUifF.jpg https://i.imgur.com/9YfUifF.jpg
- anvuong 3y agoWell AMD marketing turned out to be a joke, remember Poor Volta? I still think AMD haven't even recovered from that. There marketing for GPUs have been terrible since.
- DiabloD3 3y agoYou mean Vega. Volta is a Nvidia arch. Vega's marketing pushed Nvidia to make what is ending up to be the best product series they will ever make: series 10. That isn't much of a joke, it scared the shit out of Nvidia, and they blinked. Vega was too late in the pipeline to stop, and Raja was ultimately let go for his role in the whole thing. He refused to start making more gamer-friendly cards, and was obsessed with enterprise compute/jack of all trades cards. Immediately afterwards was a pivot towards a split arch, allowing multiple teams to pursue their intended markets. Its why AMD won against Nvidia. Nvidia still has no real answer to AMD's success, other than continuing to increase card prices and making ridiculously large chips that have poor wafer yields. Nvidia won't even have working chiplets until series 60 or 70, while AMD already has them in a shipping product.
- Toqoz_ 3y ago“Poor Volta” was the line from AMD’s marketing team.
- FirmwareBurner 3y ago>Its why AMD won against Nvidia. Won how? >Nvidia still has no real answer to AMD's success Which success? Answer to what? Are you from a paralel multiverse? In this reality, it's the other way around. Nvidia is making so much money from the AI hype than AMD is the one trying to play catch-up.
- dist-epoch 3y agoSamsung has a fab. Anyone knows why they don't want to enter the game and create an AI chip.
- treesciencebot 3y agoThis is not a good comparison. Nvidia doesn't have a fab, but they are the lead player in the AI chip space. Intel had both and look where it got them. TSMC has a good model, and you can basically take any of your designs for the same node and manufacture it in any of their plants. Same strategy can be applied to Samsung, and they already help a lot on the memory segment. The new HBM3E memory chips for H200s might be even coming from Samsung.
- systemBuilder 3y agoIntel was infected with marketing people who diseased the entire C-Suite and drained the company for 8 years without doing anything other than make up new marketing names for the 5000-11000 series of chips and their stagnant iGPUs. That level of thievery would kill any leading company. ..
- caslon 3y agoIntel actually did a lot in the last decade. Skylake alone was a massive improvement over the previous generation. AVX-512, countless open source projects—not to mention Optane, which was one of the most radical innovations in hardware in years. They did some really cool stuff with Altera IP, but the market didn't really care for it as much as it probably should have. Much of the value of Mobileye happened under Intel's ownership, too. You mention their iGPUs, but their iGPUs actually got radically better in the timespan you mention; Iris, if properly cooled and not memory-choked, was actually pretty decent for a lot of purposes. It does and historically has had a pretty terrible board, and its management hasn't been the greatest, but people who complain about them doing nothing for most of a decade generally are making a reactionary take about the lack of post-Skylake microarchitectures, while ignoring that pretty much everyone's performance gains got swallowed by vulnerability mitigations for years because they care about video games more than safety.
- frozenport 3y ago[flagged]
- wtallis 3y ago> There are is a new wave of silicon companies that use photonics, in memory compute, deterministic/fogs data flow, etc Those innovations are really cool sounding and make for great press releases, but are much less amenable to third-party benchmarking and analysis on account of those "innovations" largely still being stuck in the lab and small-scale proof of concept products, whereas Nvidia's GPUs are mass-market products that actually ship.
- frozenport 3y agoSo they ship because they aren't innovative? Google TPU for example has been shipped.
- wtallis 3y ago> So they ship because they aren't innovative? Ok, you're clearly not trying to make sense here. And there's no way you can believe that Google's TPUs have shipped as broadly as Nvidia's GPUs (or even just Nvidia's datacenter GPUs).
- m3kw9 3y agoWhy are they going back to 28nm?
- colechristensen 3y agoThis is a new article about the historical usage of 28nm.
- m3kw9 3y agoI checked the title, the 28nm and the date of the article.
- nitinreddy88 3y agoIt's more of like documentary writing on 3-4 generations old architecture. The opening statements clearly indicate why they published
- vitus 3y agoIndeed. These are cards from ~2012 (Kepler came about with the GTX 600 series).
- TaylorAlexander 3y agoI guess the only thing you’re missing then is the content of the article. ;)
- dannyw 3y agoOn a related topic: does anyone know why NVIDIA keeps reducing the bus width on their latest gen cards? A 2060 has a 192-bit bus. A 3060 has a 192-bit bus. A 4060 has a 128-bit bus! ### A 2070 has a 256-bit bus. A 3070 has a 256-bit bus. A 4070 has a 192-bit bus!
- wmf 3y agoWafer prices have increased and so has Nvidia's greed so you get less hardware for your money every generation.
- FirmwareBurner 3y agoPretty much. Also lack of real competition.
- wtallis 3y agoAt most points in the product stack, memory frequency increased by enough to compensate for the narrower bus. Dropping to a narrower bus and putting more RAM on each channel allowed for some 50% increases in memory capacity instead of having to wait for a doubling to be economical. And architecturally, the 4000 series has an order of magnitude more L2 cache than the previous two generations (went from 2–6MB to 24–72MB), so they're less sensitive to DRAM bandwidth.
- noch 3y agoWith respect: Have you actually measured performance or are you merely quoting Nvidia marketing?
- wtallis 3y agoThere's not much measurement necessary for peak DRAM bandwidth; bit rate times bus width is pretty much the whole story when comparing GPUs of similar architecture and the same type of DRAM. That's not to say that DRAM bandwidth is the only relevant performance metric for GPUs (which is why a DRAM bandwidth regression doesn't guarantee worse overall performance), but there's really no need to further justify the arithmetic that says whether a higher bit rate compensates for a narrower bus width. If you were specifically referring to the performance impact of the big L2 cache increase: I don't know how big a difference that made, but it obviously wasn't zero.
- throwit12 3y agoAlthough you wouldn't know it from the documentation, both the GK10X/GK11X silicon had serious problems with the global memory barrier instruction that had to be fixed in software after launch. All global memory barriers had to be implemented entirely as patched routines, several thousand times slower than the underlying, broken silicon. Amusingly, that same hardware defect forced the L1 cache to be turned off on the first two keplers. I suspect if you ran the same benchmark on GK110 and vs the GK210 used in the article, you'd be surprised to see no effect from the L1 cache at all.
- frognumber 3y agoI'll be honest: I find GPUs confusing. I use Hugging Face occasionally. I have no idea what GPU will work with what. How does Fermi, Kepler, Maxwell, Pascal, Turing, Ampere, and Hopper compare? How does the consumer version of each compare to the data center version? What about AMD and Intel? * Arc A770 seems to provide 16GB for <$300, which seems awesome. Will it work for [X]? * Older NVidia card go up to 48GB for about the cost of a modern 24GB card, and some can be paired. Will it work for [X] (here, LLMs and large resolution image generation require lots of RAM)? I wish there was some kind of chart of compatibility and support.