Unverified50% confidenceBenchmarkExact time
ThunderKittens' NVFP4 GEMM performance has increased from previously only 42% of the roofline theoretical peak to over 22 PFLOPS.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
UnverifiedNVFP4量化方案相比BF16基线实现了1.5倍的推理加速，同时在top-1准确率上保留了92%到97%的水平68% similarUnverifiedIn an FP16-precision 4K×4K×4K transposed matrix multiplication task, the M5 Ultra with Neural Accelerator enabled surpassed 107 TFLOPS, which is 3.51 times its own non-accelerated performance and 4.79 times that of the M3 Ultra.62% similarUnverifiedOn the Frontier Code programming benchmark, the old model scored 34.4%, while Gemini 3.7 Flash improved to 43.6%.62% similarUnverifiedOn a three-sample math problem, DeepSeek V4.1 Flash scored 74 points, on par with Opus 5.0 and surpassing GLM's 67 points and V4 Pro's 62 points.60% similarUnverifiedRTX 3060的INT4 Tensor Core理论峰值算力约为101 TOPS，而FP16仅约12.7 TFLOPS，差距达8倍以上60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/971647API
curl https://kongchang.com/api/v1/knowledge/claims/971647MCP
get_claim(id=971647)