The Minutes of Dylan Patel on blackwell https://minutesof.com/dylan-patel/on/blackwell/ Everything Dylan Patel has said on blackwell: 13 verbatim quotes between January 2025 and August 2026, each with a timestamp and a link to the recording… en Sun, 30 Aug 2026 16:22:17 +0000 Patel says SemiAnalysis initially doubted Jensen's 25x Blackwell claim, predicting only 15-20x improvement. https://minutesof.com/q/17ee08e0-f212-40ab-9204-64e701a10690/ https://minutesof.com/q/17ee08e0-f212-40ab-9204-64e701a10690/ “Jensen, when he originally launched Blackwell, had claimed it would be a 25x improvement. And at the time, no one believed him. Right? It's Jensen, right?” — WisdomTree in Europe Thu, 09 Jul 2026 06:16:39 +0000 blackwell Patel's benchmarking found Blackwell is 30x faster than Hopper on DeepSeek v3, exceeding Jensen's 25x claim. https://minutesof.com/q/be57f11d-1f2b-4256-80d2-16bbf2bed240/ https://minutesof.com/q/be57f11d-1f2b-4256-80d2-16bbf2bed240/ “In DeepSeek v three, Blackwell is 30 x faster than Hopper on on somewhere on the continuum.” — WisdomTree in Europe Thu, 09 Jul 2026 06:16:39 +0000 blackwell Patel emailed Jensen documenting skepticism about Blackwell's 25x claim from industry observers. https://minutesof.com/q/a7272d50-fda2-432c-aac5-59f78a8b24e6/ https://minutesof.com/q/a7272d50-fda2-432c-aac5-59f78a8b24e6/ “I emailed him. I like, hey, Jensen. You know, back in 2024, you said or back back when you launched Blackwell, said '24 twenty five x.” — WisdomTree in Europe Thu, 09 Jul 2026 06:16:39 +0000 blackwell Patel says Hopper GPU failure rates improved to 5% while Blackwell remains at 10-15%, expecting higher rates f https://minutesof.com/q/8d098967-35ce-4277-b237-0688cd59c6c6/ https://minutesof.com/q/8d098967-35ce-4277-b237-0688cd59c6c6/ “hopper's now at 5%, but black belt's still 10 to 15%. Wow. Right? Actually started out higher than that. Sure. And when a new generation comes out, it's gonna be higher than 15%.” — TBPN Tue, 03 Feb 2026 23:08:03 +0000 blackwell Patel finds 10x performance TCO advantage for Blackwell over H100 at certain operating points. https://minutesof.com/q/7eaa498f-3259-4e8c-a196-98d19ea3ec72/ https://minutesof.com/q/7eaa498f-3259-4e8c-a196-98d19ea3ec72/ “But at other points, you can have a 10x performance TCO benefit from going with a more expensive Blackwell server than a cheaper H100 server.” — Open Compute Project Thu, 23 Oct 2025 06:01:19 +0000 blackwell Patel says Blackwell cost is up 50-70% over Hopper but performance exceeds 2x, making it an obvious choice. https://minutesof.com/q/7a1b8698-ea50-4724-a073-b8e5df30f151/ https://minutesof.com/q/7a1b8698-ea50-4724-a073-b8e5df30f151/ “while cost is only up, know, call it 50% or 60% or 70% versus Hopper, the performance is well north of 2x, right?” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel explains Blackwell introduces third memory tier within Tensor Cores requiring complex programming model https://minutesof.com/q/89ddcb92-6d43-43ed-8e99-976561050d18/ https://minutesof.com/q/89ddcb92-6d43-43ed-8e99-976561050d18/ “Now with Blackwell, there's actually even a third tier of memory, which is memory within the Tensor Core.” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel quantifies Hopper performance improved 30-40% over 2024 with Blackwell showing similar gains in months. https://minutesof.com/q/ec5cbc83-8dcb-4132-b787-4f7c4639cd91/ https://minutesof.com/q/ec5cbc83-8dcb-4132-b787-4f7c4639cd91/ “Even Hopper in January 24 to December 2024, you still had like a 30%, 40% performance improvement. And likewise for Blackwell, you've seen a similar sort of improvement, right?” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel predicts another 50-100% Blackwell performance improvement from software optimization of standard librar https://minutesof.com/q/1220feb8-33a2-4114-958d-7e559ad3806a/ https://minutesof.com/q/1220feb8-33a2-4114-958d-7e559ad3806a/ “I see another 50% to doubling in the cards, just from the fact that there is so much low hanging fruit or performance left on the table if you are using the general open libraries, right? Whether it's cuDNN, VLM, SGLANG, all these other things,” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel predicts general-purpose Blackwell users could see another performance doubling over the next year from https://minutesof.com/q/e17fa153-5a23-488d-99aa-a4fe4fc4a705/ https://minutesof.com/q/e17fa153-5a23-488d-99aa-a4fe4fc4a705/ “for the general purpose user, yeah, you're going to have another doubling of performance potentially on the tables over the next year.” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel reveals NVIDIA doubled system-level testing time for Blackwell compared to Hopper generation. https://minutesof.com/q/a16842d1-124f-447a-89d3-73aa00b75632/ https://minutesof.com/q/a16842d1-124f-447a-89d3-73aa00b75632/ “NVIDIA learning from the issues on Hopper, with Blackwell, they actually doubled the time for system level test.” — Together AI Fri, 03 Oct 2025 20:23:30 +0000 blackwell Patel estimates Blackwell delivers 10-15x cost improvement for inference despite NVIDIA claiming 30x at GTC. https://minutesof.com/q/879abc84-3fdc-4dd7-9b63-baf6f6541b7c/ https://minutesof.com/q/879abc84-3fdc-4dd7-9b63-baf6f6541b7c/ “But now, like, Blackwell, NVIDIA's pitching 10 to 15 x improvement in cost. It's like, well, you know, they're massaging the numbers marketing.” — Unsupervised Learning: With Jacob Effron Tue, 21 Jan 2025 14:00:12 +0000 blackwell