8 ms·
Nvidia Speeds Key Chipmaking Computation by 40x
- bottlepalm 4y agoThat's pretty cool they're using GPUs to design GPUs.
- LoganDark 4y agoWell, they're using GPUs to perform the light calculations required to create a photomask that will get the designs they've already produced onto real silicon. But sure, they're using GPUs to design GPUs. I'm sure this technology will create more opportunities to use inverse lithography and therefore have huge implications for the actual designs, which may no longer have to hold back quite as much due to compute concerns.
- hackernewds 4y agoWhat about using GPUs to design GPUs that will design GPUs
- imhoguy 4y agoAdd some AI and we can sit back and see it building Dyson sphere soon /s
- tysam_and 4y agoIntroducing the RTX for T 80!: the worlds' first nearly fully recursive GPU.
- amelius 4y agoThis is not fundamentally new. We've been using CPUs to design CPUs for ages.
- carlmr 4y agoAnd compilers to compile the next version of themselves.
- ChuckNorris89 4y agoIt's just sad how absent AMD is from such innovative uses of GPUs. It wouldn't bother me too much if at least their GPUs were much more affordable than Nvidia's, but they're not. They're at best playing catch-up months or years later with slight price discounts (the Steam HW survey still reflects this market discrepancy) And I'm not talking about gimmicks like RTX, I'm talking about all these cool use cases around ML and DL like background noise cancelling, video upscaling, camera eye contact real-time deepfakes and now this. And that's if you ignore all the mind-blowing research papers put out by Nvidia which aren't featured in consumer apps yet. This is Nvidia's biggest moat and AMD isn't even in the race here and for some reason Lisa Su seems to not give enough of a shit to compete. I hate Nvidia for their price gouging and anti-consumer practices, but at least they haven't gotten complacent and are innovating on all fronts to keep pushing the envelope. Massive respect for the tech leadership at Nvidia.
- 0dayz 4y agoSure, but I think you can't really blame AMD for it. Nvidia has dominated the market for a long time, and unlike Intel they up the prices and wisely spent enough of it on R&D, Nvidia is just reaping the rewards for it.
- blihp 4y agoSure you can. The difference between the CPU and GPU markets is that AMD is a competent competitor in one of them. Pre-Ryzen AMD knew they weren't competitive and AMD CPU pricing reflected that. The reason nVidia is getting away with their pricing is that AMD has been producing mediocre hardware (and drivers) and pricing it as if it were premium and competitive for the money. AMD's continually declining GPU market share indicates that customers aren't buying their argument.
- iforgotpassword 4y agoBut it applies to AMD in every way. I could understand that pre-ryzen when they were really cash restricted and had nothing to compete anyways. So you don't throw money at software devs. but the software side apart from (Windows) gaming is still a shit show with AMD today. We wanted to use MI50s at work because it was promised they can do SRIOV, but we never got any further with AMD support than "it should work". They took ages to respond, and could not tell what was wrong from the extensive logs and hwinfo we provided them with. Also the PCI reset bug that plagued multiple generations. There's a guy maintaining a kernel module that works around that issue in a whacky way. According to his research and reverse engineering, AMD could fix that with a firmware update to those cards. Even got in contact with AMD engineers briefly and outlined what the problem was. Then radio silence, and a couple months later AMD added a very similar workaround in their kernel module, the amdgpu driver. It's just that a fix in there doesn't make any sense whatsoever, because you need that fix when you do PCI passthrough, in which case you explicitly do not load the amdgpu module, as you don't use the GPU on the host machine but, well, pass it through to the VM.
- iandanforth 4y agoAside: Nvidia named this cuLitho, which I learned from my Spanish speaking mother-in-law basically looks like they named it 'butt'. So if you see a bunch of rear-end related memes about this software you know why!
- wslh 4y agoI am a native Spanish speaker, you can check what it means in google images (NSFW): https://www.google.com.ar/search?client=safari&hl=en-ar&sxsrf=APwXEdcWnzjcvjujU3xYoAqtAANea_xyHA:1679830813864&q=culito&tbm=isch&sa=X&ved=2ahUKEwj2s8W3wfn9AhWHnpUCHRmQCIoQ0pQJegQIChAB&biw=430&bih=737&dpr=3 https://www.google.com.ar/search?client=safari&hl=en-ar&sxsr...
- majewsky 4y agoOr you can just look at a dictionary: culito, noun. Diminuitive of culo. culo, noun. Slang for arse. https://en.wiktionary.org/wiki/culito https://en.wiktionary.org/wiki/culito https://en.wiktionary.org/wiki/culo https://en.wiktionary.org/wiki/culo
- fnordpiglet 4y ago2nm arse to be precise
- amelius 4y agoHow much does that mean in practice, considering: - the computation output depends only on local features (my guess) - most transistors look the same, so you can cache these results heavily - the same holds for the interconnect layers
- tomxor 4y agoAccording to the article, rule based patterns are already used, but inverse lithography is required in more unique or problematic cases. Ideally they would use it in all cases, maybe we end up with slightly wobbly but tolerable features for the more generic or less esoteric features that use the heuristics based approach. > Even a change to the thickness of a material can lead to the need for a new set of photomasks You can imagine the light spreading out through the other side of the mask, diffracting and interfering with neighbouring features through some radius. That it's not trivial to compute suggests the number of combinations within this radius is large enough to not be highly cacheable.
- pjc50 4y ago> - the computation output depends only on local features (my guess) This is not correct: the features on the mask are larger than the features required on the target, which is the whole problem, so you need a diffraction pattern over a large area to produce a target feature. But then you need to overlap with the diffraction pattern of the next feature, and so on. I suspect that every pixel on the output gets determined by almost every pixel on the input, which is why it takes so much computation in the first place.
- jmartrican 4y agoI noticed TSMC and ASML were mentioned but not Intel. I wonder if this will set Intel further behind TSMC.
- ftxbro 4y agoThe chipmaking computation is inverse lithography, and the Nvidia system that is speeding it up is cuLitho on DGX H100. I like how inverse lithography and neural network backpropagation were both techniques introduced in the 1980s and now we are finally seeing them both come to life, so to speak, with our sufficiently advanced GPUs.