Nvidia, every mention
34 scenes (2023), the whole family · ← back to Nvidia
tap a year for its mentions
every year 2023 anyone Shawn Wang 80Dylan Patel 55Kyle Kranen 43Ali Taha 26Ethan He 24Chris Lattner 24George Hotz 23Philip Kiely 20Doug O'Laughlin 19Sarah Chieng 17
Verbatim, from the transcripts: the passages where Nvidia comes up
The State of Silicon and the GPU Poors - with Dylan Patel of SemiAnalysis
- ▶ 4:40 Dylan Patel Like 20,000 A-one hundreds. 2 times in the scene
- ▶ 7:03 Dylan Patel We have 512 H-Hundreds coming online in, in August, and it's like, oh, cool, like, but then you're like, you know, going through the supply chain, it's like, dude, you realize there's 400 to 500,000 being, 400,000 manufactured last… 3 times in the scene
- ▶ 7:46 Dylan Patel NVIDIA is going to sell well over three million, you know, total GPUs next year.
- ▶ 15:26 Dylan Patel The H-One-Hundred has 3.35 terabytes a second of memory bandwidth, and it has a thousand teraflops of FP-sixteen, B-foot-sixteen. 5 times in the scene
- ▶ 16:05 Dylan Patel Uh, H 200 will come out soon enough, which will help the ratio a little bit.
- ▶ 16:12 Dylan Patel Uh, just like the A 180 gig did versus the A 140 gig. 4 times in the scene
- ▶ 22:13 Dylan Patel They were doing good, and NVIDIA bought them, you know, in 19, I believe, or 18. 2 times in the scene
- ▶ 27:34 Dylan Patel That's, they're very, they, while they do have quite a few GPUs, they made a big announcement about having 4000 H 100, that's still relatively poor, right, when we're talking about hundreds of thousands of like the big labs, uh, like…
- ▶ 30:32 Alessio Fanelli It's like, you know, if you buy an H-one hundred, sure, the next series is gonna be better, but, like, at least the hardware is good. 3 times in the scene
- ▶ 39:16 Dylan Patel So, so NVIDIA was the only company that bet on the other side of, of more memory bandwidth, right? 16 times in the scene
- ▶ 41:08 Dylan Patel They're gonna release a better chip than the H 100, ah, within the next quarter or so, right? 6 times in the scene
- ▶ 45:13 Dylan Patel But yeah, I think, I think alternative hardware is like, it really does hit like, like sort of a peak hype cycle, kind of end of this year, early next year, because all NVIDIA has is H 100 and then H 200, which is just better, more, more…
- ▶ 47:48 Dylan Patel Um, and like, yeah, I, I think, I think in general, right, like these, these lab partnerships are going to be nice, but they're probably incentivized to, uh, you know, hey, NVIDIA, you should, 5 times in the scene
- ▶ 48:55 Dylan Patel Um, it's worse performance than the H-H-E-N-H-E-D, uh, but the cost effectiveness of it is, is better for Microsoft internally, just because they don't have to pay the Nvidia tax.
- ▶ 1:01:28 Dylan Patel Oh, NVIDIA doesn't even care about us.
Beating GPT-4 with Open Source Models - with Michael Royzen of Phind
- ▶ 4:44 Michael Royzen NVIDIA was working on this, like, back in 2019, 20, 20.
- ▶ 1:01:44 Michael Royzen Back then, the GPU shortage wasn't even nearly as bad as it is now, but, like, even then it was still challenging, um, to get, like, the quota that we needed, and he's like, okay, no problem, um, and then, like, he leaves a couple hours… 9 times in the scene
- ▶ 1:06:07 Michael Royzen Um, and so I'm very curious if the future is something like what NVIDIA is doing with their implementation of FP-Eight, um, which they're implementing in their transformer engine library, where basically, um, 2 times in the scene
- ▶ 1:06:44 Michael Royzen Um, and, like, NVIDIA claims that this strategy, um, that they're kind of demoing with the H-one-hundred has no degradation.
Why AI Agents Don't Work (yet) - with Kanjun Qiu of Imbue
- ▶ 45:27 Shawn Wang Like, like, so, for example, uh, like, the Voyager paper coming out of, uh, NVIDIA, um, played Minecraft and set, set their own benchmarks on, uh, getting the Diamond X or whatever, and, uh, and exploring as much of the territory as…
RWKV: Reinventing RNNs for the Transformer Era
- ▶ 55:31 Eugene Cheah And donated the A-One-Hundreds needed to train the basic models that RWKB had.
FlashAttention-2: Making Transformers 800% faster AND exact
- ▶ 8:55 Tri Dao I think there was a paper from NVIDIA folks back in 20, um, 18 about this, and then there was a paper from, um, Google, so, um,
- ▶ 12:28 Tri Dao If you're using, um, a-one-hundred, and you, you list the GPU memory, it's like, 40 gigs or 80 gigs, so that's, that's the, that's the HBM.
- ▶ 15:11 unnamed speaker I think the, the latest NVIDIA thing as a HBM three on this.
- ▶ 31:22 unnamed speaker Um, I read in the, in the blog post that, um, a lot of the work was, like, also related to, like, some of the NVIDIA library updates. 7 times in the scene
- ▶ 34:43 unnamed speaker It's a, there could be all these things that are much better, like our architecture that are better, but they're not better on Nvidia. 2 times in the scene
Ep 18: Petaflops to the People — with George Hotz of tinycorp
- ▶ 2:18 George Hotz Uh, what are the odds they nationalize NVIDIA? 3 times in the scene
- ▶ 4:33 George Hotz The only company, there's one other company aside from Nvidia who's succeeded at all at making training chips. 4 times in the scene
- ▶ 17:44 George Hotz Whereas if you're trying to build something that is just straight up good on NVIDIA and we have a lot of people and complexity to throw at it. 6 times in the scene
- ▶ 24:10 George Hotz So you don't need CUDA installed, just the Nvidia open source driver. 3 times in the scene
- ▶ 41:33 George Hotz We like the term compute cluster, so we can use NVIDIA GPUs. 2 times in the scene
- ▶ 41:53 George Hotz Um, so, the bandwidth is the, is roughly 10 X less than what you can get with NV-linked A-Hundreds. 2 times in the scene
- ▶ 48:36 George Hotz Uh, it's, uh, it's 400 grand. 2 times in the scene
- ▶ 1:14:38 George Hotz I mean, actually, you know, I'm kind of grateful for NVIDIA, right?