@themeghamachine

AI infra + energy efficient computing 🌱⚡️🤖 eng @crusoeAI

 San Francisco, CA
meghajain.me
Joined March 2016
  • Tweets1.2K
  • Following1.2K
  • Followers1.8K
65Media
there’s many ways to optimize intelligence per watt — even outside of GPU architecture or improving data center power efficiency (PUE). we ran a fun experiment to see how @crusoeAI’s inference stack itself can save energy a layer above the silicon, and the results were pretty cool! crusoe.ai/resources/blog/how…

Tokens per watt: tuning the inference stack for efficiency

Crusoe measured tokens per watt on NVIDIA HGX B200 and found that tuning the inference stack alone cut energy per output token 26.9% with no hardware changes.

crusoe.ai
11
4
1
34
10,271
me when I’m a kubernetes pod perpetually stuck in ImagePullBackOff
kat
@katsunheOct 24
1 thing about me i am always waiting for a package 📦
2
478
huge moat for humans right now is thought organization. this has always been true: the people who excel at problem decomposition, finding patterns where no one else sees them, and building the right taxonomy for a messy domain. AI exacerbates it though
1
2
253
more replies
even defining evals being so hot (from a recruitment perspective) is just a symptom of how hard it is to find people who can synthesize and structure their findings in a solid way
1
75
same basic principles as context engineering, but applied to yourself. this is the crux of what people mean when they say they’re looking for “clarity of thought”
33
crusoe 🤝 ollama ⚡️ fast, energy efficient inference
Crusoe
@CrusoeAIOct 21
New open models ship fast. Ollama's developers expect them on day one. @ollama runs that workload on NVIDIA Blackwell GPUs through Crusoe Managed Inference, in Crusoe Spark™ modular data centers in Nevada. Signed deal to production: two weeks 👇 crusoe.ai/resources/customer…

Ollama scales open models in 2 weeks on Crusoe Managed Inference

Watch how Ollama scaled open models on Crusoe Managed Inference and solar-powered Crusoe Spark data centers, with day-zero model access and U.S. data residency.

crusoe.ai
2
5
514
if you dig deep enough, you’ll realize it’s all just one big array
5
1,528
feeling a little prideful bc I know exactly which street (and step) these were taken from ✨ very lucky I get to see this every day
oomf magazine
@oomfmagazineOct 15
San Francisco could be the greatest city on earth if it wanted to be
12
1,685
Neoclouds can't just lease GPUs. They need to own the full agent inference platform: open-weight models (SGLang), fine-tuned models (Miles, River), and secure sandboxes (SPACE). We look forward to working with Coreweave, Akamai, Crusoe, and more partners.
Reed Albergotti
@ReedAlbergottiOct 15
New from me: Perplexity is open sourcing its sandbox, called SPACE, and has partners like Akamai, Coreweave and Crusoe in the big push for AI security
54
40
7
628
58,992
listening to feelings are forever and thinking of how @yanated is criminally underrated. not only an insane vocalist but also a great songwriter
1
846
also wow the banjos and guitars and just music production in general. so so good
1
588
Megha Jain retweeted
tender
@tenderizzationOct 13
running into your ex in public is annoying but you ever have slurm repeatedly allocate you the one node you don’t like
12
28
1
493
15,049
engineers def have an equivalent of this (RFCs, architecture reviews) and get daily feedback on PRs — I think it’s more that you’re drilled to not see it as “personal” whereas design may feel more inherently personal. hence better translation of design crit -> personal feedback
christine
@christineistOct 12
the fact that my non designer friends have such a hard time with feedback makes me think that everyone should go through regular design crit in all fields
6
999
I wanna see more Carl Sagan energy in energy
maja 🔭🍒
@majamediacoOct 11
I wanna see more Carl Sagan energy in tech
7
687
recently learned about B and D (the programming languages) and realized I had the opportunity to do something really funny…
1
4
646
spoiler alert: on identical B200 hardware, tuning Crusoe's stack cut energy per output token by 26.9%, and MemoryAlloy's KV cache cut another 23% in isolation
1
3
297
the earth’s favorite cloud for a reason 🌱
1
215
📸 a well spent weekend in the redwoods @hardlaunching @BainCapVC @Accel
3
3
1
32
3,303
it’s about damn time - cool to see the conversations about a national compute reserve from last year becoming a reality. even cooler that we get to partner with them :-)
Anjney Midha
@AnjneyMidhaOct 7
hundreds of MI355x and B300 nodes now live on nationalcompute.com subsidized for all .edu and .govs it’s time to scale the frontier ecosystem for the public benefit
19
2,952
Load more