Unverified50% confidenceFactExact time
With a configuration of 16384 Tokens per card, 7168 values per Token (2 bytes each), and 6 copies, a single Dispatch requires approximately 1.4 GB of data movement per card, and FP8 can halve this amount.
1
Sources
50%
Confidence
Long-term
Relevance
10/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA dual 24GB GPU tensor-parallel configuration increases context support from 134,049 to 262,144 tokens at f16 precision.70% similarUnverifiedFP16使用16位存储数值（1位符号位、5位指数位、10位尾数位），可表示约65000个离散值70% similarUnverifiedUP主使用FP8精度部署千问3.8 27B在48GB显卡上，跑4个并发request占用约46GB69% similarUnverifiedWith an approximately 1.6M ultra-long context and 128 concurrent requests, throughput can reach approximately 3,700 tokens/s (aggregated).68% similarUnverifiedA speed of 4–5 tokens/second is too slow for practical use on an 8GB GPU; a 16GB GPU can improve this to 12–14 tokens/second, entering a practically usable range.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/970626API
curl https://kongchang.com/api/v1/knowledge/claims/970626MCP
get_claim(id=970626)