No Priors · Sep 18, 10:00 AM UTC
Diffusion LLMs match custom-chip speed on Nvidia GPUs with lower cost
Historical operation · Retained
“If you have a diffusion based LLM that it's built to be parallel, it's accelerating at the software level, then you can get the same speed on Nvidia GPUs, uh, which means much more, um, availability. I mean, GPUs are scarce, but there's more of that than than custom chips, uh, and lower cost, higher quality.”
Quoted source · 18:29–18:51 · 9 source segments
- [1] 18:29–18:32
- [2] 18:31–18:34
- [3] 18:32–18:37
- [4] 18:34–18:40
- [5] 18:37–18:42
- [6] 18:40–18:44
- [7] 18:42–18:48
- [8] 18:44–18:50
- [9] 18:48–18:51
Model assessment
The transcript describes a customer switching from custom chips to diffusion-based LLMs that achieve the same speed on Nvidia GPUs, with lower cost and higher quality. This indicates that Nvidia GPUs can serve workloads previously requiring custom hardware, potentially increasing demand for Nvidia products.
9 exact transcript citations