Outperforming cuBLAS on NVFP4 @GPUMODE

Outperforming cuBLAS on NVFP4 @GPUMODE

Uploaded September 2026 | Updated September 2026, 3 hours ago

GPU MODE talk by Pranjal: Outperforming cuBLAS on NVFP4.

cudaforfun.substack.com/p/outperforming-cublas-on-nvfp4

Outperforming cuBLAS on NVFP4

Lecture 99: Distributed ML on consumer devices

Lecture 79 Mirage (MPK): Compiling LLMs into Mega Kernels

Lecture 107: PithTrain

Lecture 58: Disaggregated LLM Inference

Lecture 59: FastVideo

Lecture 93: Cornserve Easy, Fast and Scalable Multimodal AI

Lecture 84: Numerics and AI

Live - Disaggregated LLM Inference: Past, Present and Future

CuTe

Lecture 108: One Layer Deeper competition

Livestream: Distributed GEMM

Outperforming cuBLAS on NVFP4