The Next 3x in Inference Won't Come from Faster Kernels | Hacker News - But Better

The Next 3x in Inference Won't Come from Faster Kernels twitter.com

The Next 3× in Inference Won't Come From Faster Kernels

2 points•floathub•1 day ago•1 comment•

#faster #inference #kernels #faster kernels

1 comment

floathub1 day ago

One unique idea for solving the chip shortage: pack more models into each existing GPU by having them share the resources. :-)

Read the full thread on Hacker News →

Related stories

Launch HN: Magnitude (YC S25) – Self-optimizing inference engine for agentsgithub.com
Hacker News · 134 points · about 11 hours ago
Writing Rust code that's faster than state-of-the-art libraries by asking agents to make the code fasterminimaxir.com
Lobsters · 5 points · 8 days ago
Show HN: Droid ASC – An On-Demand Android Decompiler, 41–269x Faster Than JADXgithub.com
Hacker News · 2 points · 8 days ago
Rust code faster than SOTA libraries by asking agents to make the code fasterminimaxir.com
Hacker News · 2 points · 9 days ago
Writing Rust code faster than SotA by asking agents to make the code fasterminimaxir.com
Hacker News · 1 points · 9 days ago
The Basics of Transformer Inferencejax-ml.github.io
Hacker News · 2 points · 9 days ago