社区测试称Qwen八路并发接近线性扩展
AI 总结
Reddit 用户称其在 vast.ai 租用的实例上花费约 40 美元，测试采用大内存服务器的 Qwen 3.8 Flash Next Unsloth Q8 推理表现。两颗 AMD EPYC 7642、配备 251GB 或 504GB DDR4 内存的系统在 124k 上下文下，测试到八路并发时每路性能接近线性扩展。原始数据发布在 Reddit 评论中，测试未覆盖八路以上并发。
深度见解
这项社区测试为评估大内存 AMD 服务器上的大模型并发推理提供了实用数据，但单一测试者及八路上限使结论的普适性仍不确定。
Raw data in the comments, if you go for that sort of thing.
The TL;DR
This build generates a nearly linear performance per stream up to at least 8 streams using a context window of 124k. I didn't test beyond that.
Abstract
Until today, I considered Strata a toy. I don't anymore.
Rented a few instances on vast.ai to do some benching. Blew about $40 in all on the benchmarking.
#Strata#Unsloth#Qwen 3.8 Flash Next#Maya#GLM 5.3F#DeepSeek V4.1F#AMD EPYC 7642#vast.ai