Unverified50% confidenceFactExact time
The deployment scale of V4.1 is prefill EP2 to EP8 and decode EP64 to EP128; DeepInfer adds intra-group voting and a second round of confirmation in the scheduler to coordinate card pacing under wide expert parallelism.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
DeepSeek Open-Sources Inference Engine DeepInfer: All the Cost-Saving Tricks Laid Bare
bilibili最佳拍档大飞10/23/2026
Related Entities
Related Claims
UnverifiedThe official recommended starting configuration for running DeepInfer efficiently is 64 cards for prefill plus 6 tokens.62% similarUnverifiedx86平台的AVX-512指令可在单条指令内并行处理64个int8元素62% similarUnverifiedEPYC 9006系列很可能是AMD高端数据中心产品线的延续与升级，但具体规格尚未明确披露62% similarUnverifiedAMD EPYC 9004系列（Genoa）单颗处理器可提供最多96个物理核心与192个线程，并支持AVX-512向量指令集处理512位宽数据62% similarVerifiedLeveraging the CPU's SIMD instruction sets (such as AVX-256 or AVX-512) enables processing of 8 or 16 float32 operations simultaneously in a single instruction.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1090447API
curl https://kongchang.com/api/v1/knowledge/claims/1090447MCP
get_claim(id=1090447)