Unverified50% confidenceFactExact time
The Mega MoE fused operator makes communication warps and compute warps share the same batch of SMs, causing contention for registers, shared memory, and issue slots, with the synchronization domain covering the entire SM array, and each replay requires clearing the workspace and performing a cross-rank barrier.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSM Occupancy refers to the ratio of the number of active warps per streaming multiprocessor to the maximum number of warps supported by the hardware.69% similarUnverified占用率（occupancy）衡量实际活跃warp数与SM支持的最大warp数之比，受每个线程使用的寄存器数量、每个线程块的共享内存大小和线程块配置方式影响63% similarUnverifiedCoalesced memory access requires threads within the same warp to access contiguous memory addresses in order to maximize the efficiency of each memory transaction.60% similarUnverifiedDuring the jailbreak, by reallocating CPU priority, unlocking general-purpose GPU compute access, freeing additional memory bandwidth, and combining with a model fork (PS5LM) optimized for parallel processing and memory bandwidth, an inference speed of 100+ tokens/s is achieved.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1133728API
curl https://kongchang.com/api/v1/knowledge/claims/1133728MCP
get_claim(id=1133728)