Evidence receipt / belief
Published · transcript-backedDylan Patel: belief
2 Oct 2024 Dwarkesh Podcast @Asianometry & Dylan Patel — How the semiconductor industry actually works
“Next year, 300-500k depending on whether it's one site or many. 300-700k I think is the upper bound of that.”
Source trail
Everything needed to verify it.
- Speaker
- Dylan Patel
- Attribution
- Verified speaker
- Claim type
- belief
- Recorded
- 2 Oct 2024
- Publisher
- Dwarkesh Podcast
Transcript context
…Rather than who it is, I just want to know how many… I don't know if it's better to denominate in H100s… It’s 100k GPUs this year for the biggest cluster. Next year, 300-500k depending on whether it's one site or many. 300-700k I think is the upper bound of that. But it's about when they tier it on, when they can connect them, when the fiber's connected together. Let's say 300-500k but those GPUs are 2-3X faster versus the 100K cluster. So on an H100 equivalent basis, you're at a million chips next year in one cluster by the end of the year. Well, one cluster is the wishy-washy definition. It’s multisite, right? Can you do multisite? What's the efficiency loss when you go multisite? Is it possible at all? I truly believe so. What's the efficiency loss is the question. Would it be like 20% loss, 50% loss?…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.