To Infinity and Beyond: ThunderKittens Now on NVIDIA Vera Rubin NVL72!
Together AI has ported the ThunderKittens library to support the NVIDIA Vera Rubin NVL72 architecture. The implementation includes a rebuilt NVFP4 GEMM kernel that achieves over 22 PFLOPS performance.
Verified State Diff
Impact & Verification Analysis
AI infrastructure engineers, GPU kernel developers, and high-performance computing researchers.
This demonstrates the ability to achieve near-native performance on next-generation NVIDIA hardware using custom kernels, reducing reliance on closed-source libraries like cuBLAS for high-throughput AI workloads.
Full Fact Overview
The integration involves a low-level optimization of the NVFP4 GEMM (General Matrix Multiply) specifically tailored for the Vera Rubin NVL72 hardware. By leveraging new ISA (Instruction Set Architecture) capabilities, the team improved efficiency from 42% of the theoretical roofline to a performance level competitive with NVIDIA's proprietary cuBLAS and CuTe DSL libraries.