Unverified50% confidenceBenchmarkExact time
Across 8B and 70B shapes and batch sizes from 1 to 32, FuTransformer's INT4 decode is 1.3 to 2.4x faster than the faster of PyTorch's BF16 and INT4 blocks.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/19/2026
First Seen
Valid until: 11/2/2026
Sources
Related Entities
Related Claims
UnverifiedFuTransformer's INT4 decode is 31% to 51% faster than its own FP8 path.77% similarUnverifiedOn a single H100, FuTransformer's prefill path is up to 1.48x faster than compiled PyTorch FP8, reaching 1.29 PFLOP/s.70% similarVerifiedNVIDIA's H100 features a Transformer Engine specifically designed for AI workloads, delivering nearly 4,000 TFLOPS of compute at FP8 precision, approximately 3 times the improvement over the A100.69% similarUnverifiedB200采用Blackwell架构，引入第二代Transformer Engine和FP4精度支持，推理吞吐量较H100提升最高30倍65% similarUnverified低比特量化的实际加速效果取决于硬件是否具备对应的INT4计算单元，并非所有显卡都能获得明显提速64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065909API
curl https://kongchang.com/api/v1/knowledge/claims/1065909MCP
get_claim(id=1065909)