Everyone's trying to make CPUs faster.
I’m trying to make them worse.
New project: CPU deoptimization.
Searching for the slowest possible machine instructions
x86 single instruction record:
198,002,498,236 cycles, 62 seconds
The assembly hall of shame: https://t.co/G4QnFQjsZx
@nirajbhusal Unix says: do one thing and do it well
Nagarik App: do 50 things, half of them crash, nobody cares, nobody uses, the app just exists to exist
I've been working on a new LLM inference algorithm.
It's called Speculative Speculative Decoding (SSD) and it's up to 2x faster than the strongest inference engines in the world.
Collab w/ @tri_dao@avnermay. Details in thread.
Visualizing the hyperbolic Voronoi diagram in 5 models:
➀ Poincaré upper plane
➁ Poincaré disk
➂ Klein disk
➃ Beltrami hemi-sphere
➄ Minkowski hyperboloid
When you fork 25 processes and make them fight for a CPU you get to see the Linux completely fair scheduler red-black tree dance in real time
Oddly satisfying