1
What happens when you run a CUDA kernel?
Tracing one vector-add kernel from nvcc all the way down to the warps that execute it.
Fergus Finn
A blog about LLM inference
- Popular posts
- 3
- Total points
- 569
- HN comments
- 105
- Rank
- #4,419 personal
Visit fergusfinn.com → RSS feedOn Hacker News Jun 2026 – Jun 2026
Hardware & Electronics 2AI & Machine Learning 2Systems & Performance 1Databases & Data 1
2
Speculative KV coding: losslessly compressing KV cache by up to ~4×
Lossless compression of a target model's KV cache by up to 4×, using a cheaper predictor model to drive an arithmetic coder.
3
Bringing Up DeepSeek-V4-Flash on AMD MI300X
A story of sharp edges, segfaults, and standards