Unverified50% confidenceFactExact time
The decode unit of DeepSeek V3 uses 144-card EP, amortizing the weight load overhead across large batches and freeing up GPU memory to increase batch size.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/23/2026
First Seen
Valid until: 12/22/2026
Sources
Related Entities
Related Claims
UnverifiedIn the DeepSeek V1 era, each incremental token required nearly 400 KB of KV cache, which dropped to 48 KB in V3.2.78% similarUnverifiedDeepSeek V3's activated parameters are compressed to 3 to 5 billion, with a total parameter count of 671B.68% similarUnverifiedThe official deployment requirements for the DeepSeek V4 vision version demand 4 NVIDIA GB300 graphics cards to support full inference services.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/942625API
curl https://kongchang.com/api/v1/knowledge/claims/942625MCP
get_claim(id=942625)