Unverified50% confidenceFactExact time
FP32的atomicAdd从Volta架构开始支持硬件原生浮点原子操作，FP16的atomicAdd从Pascal架构开始支持
1
Sources
50%
Confidence
Long-term
Relevance
8/12/2026
First Seen
Sources
固定随机种子GPU结果仍不可复现？深度解析原因与解决方案
Related Claims
UnverifiedThe 16-series Turing architecture introduced first-generation Tensor Cores supporting FP16 mixed-precision training, while the 30-series Ampere architecture upgraded to third-generation Tensor Cores and added hardware acceleration support for BF16 and TF32.65% similarUnverifiedV100所属的Volta架构不具备原生FP8硬件加速支持，FP8原生支持于Hopper架构的H10064% similarUnverifiedAs a data center flagship, the P100 is equipped with dedicated FP16 SIMD units, theoretically capable of 21.2 TFLOPS of FP16 compute, making it the only model in the Pascal family with fast fp16 hardware units.63% similarUnverifiedThe A100's Ampere architecture supports the TF32 format, while the V100's Volta architecture does not.63% similarVerifiedFP8 was first introduced with hardware support by NVIDIA in the H100 architecture, with two subtypes: E4M3 and E5M2.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/735648API
curl https://kongchang.com/api/v1/knowledge/claims/735648MCP
get_claim(id=735648)