YOOOOO guys, Snackwell has kernels now :), Last time I said I'd improve shared memory, build an assembler and write vector and matrix programs.Those now run in RTL simulation, along with single head attention. Here's what changed and what I did. 🧵👇
over the last 6 months I spent my time building https://t.co/PxQeUIK68k !
our reinvention of Groq's LPU that supports full forward pass on our custom transformer MicroGPT model
take a look at how we did it 👇
Hey guys! I’ve been working on Snackwell again, and it’s come a long way from the 4-lane SIMD core I started with.
It’s now a working SM with warp scheduling, scoreboarding, shared memory, load/store, branch divergence, reconvergence, and a top-level wrapper in SystemVerilog.
Here’s how it all came together 😎👇
"claude fable just proved the jacobian conjecture is false"
"incredible this changes everything"
"do you know what the jacobian conjecture is?"
"no, do you?"
"no"
@evanliin I think as someone still trying to figure out for sure what field of engineering I want to go into, AIs ability factors in cause it feels like no fun if AI is better than most people at the job