In the news
To Infinity and Beyond: ThunderKittens Now on NVIDIA Vera Rubin NVL72!
Together AI · Published · 3 min read
In 30 seconds
- What happened
- Together AI optimized ThunderKittens GEMM kernels for NVIDIA Vera Rubin NVL72, achieving over 22 PFLOPS on NVFP4 operations.
- Why it matters
- Matters for engineers building high-performance matrix multiplication kernels on next-generation NVIDIA GPUs requiring low-level ISA optimization.
- Watch out
- Optimizations are highly specific to Vera Rubin architecture; gains from individual features vary by workload size and may not transfer to other platforms.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.