Comments for The Next Platform

Comment on How Did DeepSeek Train Its AI Model On A Lot Less – And Crippled – Hardware? by itellu3times

itellu3times — Tue, 28 Jan 2025 00:50:57 +0000

So it’s all mechanics, and really not a touch of actual theory.
But then, what is the theory behind LLMs? Doh!

Comment on How Did DeepSeek Train Its AI Model On A Lot Less – And Crippled – Hardware? by Carl Schumacher

Carl Schumacher — Tue, 28 Jan 2025 00:03:51 +0000

If this is indeed “DeepFake”, then its one of the best engineered shorts (both in an energy and IT capital sense) in decades.

Comment on How Did DeepSeek Train Its AI Model On A Lot Less – And Crippled – Hardware? by Tapa Ghosh

Tapa Ghosh — Mon, 27 Jan 2025 22:33:03 +0000

“And here is another side effect: The V3 model uses pipeline parallelism and data parallelism, but because the memory in managed so tightly, and overlaps forward and backward propagations as the model is being built, V3 does not have to use tensor parallelism at all. Weird, right?”

This is mostly because of the small # of GPUs used, they can use expert parallelism as well, to eliminate the need for TP, if you used more GPUs, you’d need TP

Comment on GenAI Boom: Datacenter Spending Forecast Raised Again by Kris

Kris — Sat, 25 Jan 2025 21:04:31 +0000

With Gartner’s report, which companies will benefit the most worth investing in growth stocks?

Comment on OpenAI Declares Its Hardware Independence (Sort Of) With Stargate Project by HuMo

HuMo — Thu, 23 Jan 2025 18:54:41 +0000

Spot on! And yet rather soap-opera-esque … if I understand well, OpenAI found itself a wealthier new “suitor” who promised it the Stargate, but it also wants to remain friends with its ex, Microsoft. And meanwhile, its ex-ex, or former ex, now of xAI, is building its own Colossus dream castle, while suing OpenAI for lying about being open during their past romance, and it is also stating that its new suitor is not even that rich anyways (appearing obsessively jealous in the process)! Keeps me at the edge of my seat, wondering who will end up with who, when, where, and for how much? Particularly seeing how, as far as openness is concerned, a new potential romantic partner migh just have hit town, looking to exotically challenge OpenAI’s o1, on the large language catwalk: https://api-docs.deepseek.com/news/news250120 8^b

Comment on HLRS Takes First Steps To Exascale by Slim Albert

Slim Albert — Thu, 23 Jan 2025 15:44:09 +0000

Nice to see Hunter embracing the MI300A, for serious matrix-vector oomph at 1/10ᵗʰ the power consumption of an all CPU machine (for example Top500’s #38 Shaheen III – CPU) — granted though that all-CPU would be the way to go for graphs (eg. Fugaku vs Frontier and Aurora in Graph500).

Comment on TSMC Can’t Be Caught Or Bought, Only Sought Or Stolen by Timothy Prickett Morgan

Timothy Prickett Morgan — Mon, 20 Jan 2025 16:11:21 +0000

In reply to Eric Olson. Correct. Happy fingers hit a zero unintentionally.

Comment on TSMC Can’t Be Caught Or Bought, Only Sought Or Stolen by Anonymouse

Anonymouse — Mon, 20 Jan 2025 15:09:10 +0000

5 times faster is 300%?

Comment on TSMC Can’t Be Caught Or Bought, Only Sought Or Stolen by Eric Olson

Eric Olson — Sat, 18 Jan 2025 01:29:40 +0000

By my calculation 6 times 5 is 30.

Comment on The Bespoke Supercomputing Architecture That Stood the Test of Time by Ray McConnell

Ray McConnell — Fri, 17 Jan 2025 19:21:24 +0000

It would be interesting to understand where you are today given the recent advances in AlphaFold and related protein design tooling coming from DeepMind. BTW are they now your competitors in drug design going forward?