San Francisco Compute (@sfcompute) 's Twitter Profile
San Francisco Compute

@sfcompute

we sell very-large H100 training clusters with IB & fast disk you can buy in short-bursts (ie: 96 H100s for a week) 🇺🇸🏳️‍⚧️🌁

ID: 1690471129121705984

linkhttp://sfcompute.com calendar_today12-08-2023 21:12:39

98 Tweet

6,6K Followers

5 Following

JohnPhamous (@johnphamous) 's Twitter Profile Photo

app banner - no layout shift - doesn't block render - corner radii change to 0 to reduce visual contrast - bezier curves for the notch

gerred (@devgerred) 's Twitter Profile Photo

Side note everyone - we are hiring a Financial Systems Engineer at San Francisco Compute! Come help build a market that enables liquidity in a GPU market that otherwise prefers massive up front commitments, and enables researchers and companies to get the resources they need for ML.

gerred (@devgerred) 's Twitter Profile Photo

This is fucking _awesome_ and trained on San Francisco Compute. Lots of intense work into making this product really great, and stuff like this is why I'm so excited. More accessible AI infrastructure. You can't do this slinging YAML alone.

Latent.Space (@latentspacepod) 's Twitter Profile Photo

🆕 SF Compute: Commoditizing Compute latent.space/p/sfcompute We're excited for our latest deep dive into the compute market with evan conrad of San Francisco Compute! It should not be normal for the prices of one of the world’s most important resources right now to swing from $8 to $1

gerred (@devgerred) 's Twitter Profile Photo

Alpha users on San Francisco Compute can inference DeepSeek v2 Prover 671B on 24 H100s IB at market rate since a few minutes after downloaded, and cache compiled. vLLM project, multi-node LWS ready. $0.44/gpu/hr right now per hour, but market rates apply.

Alpha users on <a href="/sfcompute/">San Francisco Compute</a> can inference <a href="/deepseek_ai/">DeepSeek</a> v2 Prover 671B on 24 H100s IB at market rate since a few minutes after downloaded, and cache compiled. <a href="/vllm_project/">vLLM</a> project, multi-node LWS ready. $0.44/gpu/hr right now per hour, but market rates apply.
evan conrad (@evanjconrad) 's Twitter Profile Photo

We've partnered with Modular to create Large Scale Inference (LSI), a new OpenAI-compatible inference service. It's up to 85% cheaper than other offerings & can handle trillion-token scale. We originally created it at the request of a major AI lab to do large scale multimodal

We've partnered with Modular to create Large Scale Inference (LSI), a new OpenAI-compatible inference service. 

It's up to 85% cheaper than other offerings &amp; can handle trillion-token scale.

We originally created it at the request of a major AI lab to do large scale multimodal
Chris Lattner (@clattner_llvm) 's Twitter Profile Photo

I'm very excited to partner with SFCompute - Evan and team are phenomenally driven and built a powerful platform for scaling GPU solutions like never before. Combined with Modular's high-performance inference solutions, they're able to deliver incredible TCO advantages! 👇

evan conrad (@evanjconrad) 's Twitter Profile Photo

we, sfcompute, are hiring someone for a short term contract that involves running around the city if you are in sf, want to work in startups, and want to have an adventure, please DM us

evan conrad (@evanjconrad) 's Twitter Profile Photo

every time we work with a customer on inference we deeply embed with them, setup an apples-to-apples comparison, and use that to quote a price that beats what they're currently doing while keeping the same accuracy we're starting to productize that whole experience now

swyx (@swyx) 's Twitter Profile Photo

buried in Sriram Krishnan's America's AI Action Plan is endorsement that the US compute market will financialize with spot and forward contracts. this podcast explains why this is so necessary, not just for speculation one of the most consistent themes with Latent.Space's GPU

buried in <a href="/sriramk/">Sriram Krishnan</a>'s America's AI Action Plan is endorsement that the US compute market will financialize with spot and forward contracts. this podcast explains why this is so necessary, not just for speculation

one of the most consistent themes with <a href="/latentspacepod/">Latent.Space</a>'s GPU