#cuda

• [0007]2026-07-02Writing A Megakernel For LLM Decode - A Worklog#cuda#megakernel#inference#blackwell#sm120#rtx#5080
• [0006]2026-05-19Unleashing Blackwell's 4-bit: a surgical look at MXFP4 and NVFP4#cuda#quantization#fp4#nvfp4#mxfp4#blackwell#sm120#rtx#5080
• [0005]2026-05-13From 429 GB/s to the DRAM wall: writing an FP8 quantizer on an RTX 5080#cuda#quantization#fp8#blackwell#sm120#rtx#5080

all posts →