fcalva

@fcalva@cyberplace.social

This is a remote user, information on this page may be incomplete. View at Source ↗

Lisp enjoyer, occasional code golfer.
I self host what i can, because i can.
As for the rest, i invite you to read my timeline.

⁨Comment⁩ on ⁨China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs⁩ ⁨⁨13⁩ ⁨hours⁩ ago⁩:
@carrylex @kalkulat If you zoom in, you will see that it is in fact a real photo. It's just someone who thought putting a CPU upside down was a good idea.
⁨Comment⁩ on ⁨China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs⁩ ⁨⁨13⁩ ⁨hours⁩ ago⁩:
@Treczoks @flemtone Thing is, the final LLM inference is usually done at reduced precision. 8-16 bits usually, but even 4bits or lower with different layers of varying precision.
⁨Comment⁩ on ⁨China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs⁩ ⁨⁨13⁩ ⁨hours⁩ ago⁩:
@TheBlackLounge @kalkulat LLM inference is definitely theoretically possible on analog chips. They just may not scale :v