Unverified50% confidenceReleaseExact time
ROCm 6.x版本在2024年引入了对FP8的硬件级支持、改进的RCCL通信性能以及更完善的MoE模型算子优化
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
DeepSeek-V4 Flash RL Training Successfully Ported to AMD MI355X GPU: A Complete Breakdown
twitterlmsysorg7/10/2026
Related Claims
UnverifiedIt is expected that by 2026–2027, as terminal chip processes advance to 2nm and NPU architectures are optimized, terminal devices are expected to run tens-of-billions-parameter-scale models smoothly.60% similarUnverifiedGLM 5.2引入了多令牌预测（MTP）技术，该技术由Meta在2024年研究中提出，可通过投机解码提升推理速度56% similarUnverifiedBlackwell架构于2024年发布，是NVIDIA继Hopper之后的新一代架构，支持FP4/FP6原生张量计算56% similarUnverifiedMTIA第一代于2023年正式对外披露，采用台积电7nm制程，专为Meta推荐系统和排序等推理工作负载设计56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/497625API
curl https://kongchang.com/api/v1/knowledge/claims/497625MCP
get_claim(id=497625)