Unverified50% confidenceFactExact time
On the B300, GC-MoE allocates only 20 of the 148 SMs to dispatch and leaves all the rest for computation.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Related Entities
Related Claims
UnverifiedQwen3-235B-A22B adopts a MoE architecture, activating only 22B out of 235B total parameters at a time.66% similarUnverifiedConstrained core SKUs (such as M32-16) provide the memory and throughput of M32 but only 16 cores are usable, and licensing is billed for only 16 cores.62% similarUnverifiedFor a dense 27B model, --max-num-seqs 128 is a suitable batch limit; its inflection point comes earlier than that of the A3B MoE architecture, for which 160 is a better setting.62% similarUnverifiedDeploying Qwen3 27B with 16 GB of VRAM limits the choice to the Q3_K_XL GGUF quantization version, which is the highest precision that 16 GB of VRAM can accommodate.60% similarUnverifiedThe A100 has 108 SMs, the H100 has 132 SMs, and the B200 has 192 SMs.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1133725API
curl https://kongchang.com/api/v1/knowledge/claims/1133725MCP
get_claim(id=1133725)