Skip to content
View hdcdn's full-sized avatar
  • Joined Oct 24, 2026

Block or report hdcdn

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse

Popular repositories Loading

  1. heterogeneous-pd-lab-data heterogeneous-pd-lab-data Public

    社区 heterogeneous-gpu-pd-lab 的 14 个实验、9 份 CSV 的中文整理:小卡做 Prefill 的真实收益、数据口径陷阱、与四机 GB10 实测的对照

    2

  2. rtx40-ada-unlock-notes rtx40-ada-unlock-notes Public

    gpu-unlock-ada 的逐卡实测整理:4090 FP16 165.6→331.6 TF/s(2×),4070 Ti SUPER 95→189.9,4060 Ti 仅验证机制,4080/4080S/4070/4060-8G 无验证。附三处辟谣。

    1 2

  3. dgx-spark-lab dgx-spark-lab Public

    四台 DGX Spark(GB10) 本地推理集群实测:DeepSeek-V4.1-Flash MXFP4 / MiniMax-M3.1 双方案 / RTX40系解锁 / 显卡性价比

  4. deepseek-v41-flash-4x-gb10 deepseek-v41-flash-4x-gb10 Public

    4× DGX Spark GB10 纯 TP4 部署 DeepSeek-V4.1-Flash-1017 MXFP4 的完整实测:解码 C1 106.7 / C4 222.8 / C8 317.0 / C16 439.9 tok/s,干净预填 ≈5,000 tok/s,1M 上下文实测边界

  5. minimax-m31-flash-gb10 minimax-m31-flash-gb10 Public

    MiniMax-M3.1-Flash 在 GB10 上的两条路线实测:单机 EXL3 2.15bpw 跑通(19.7 tok/s / 8K ctx),四机 vLLM TP4 卡在 sm_121 的 FlashInfer MXFP8 内核

  6. gpu-price-perf-ranking gpu-price-perf-ranking Public

    9 张显卡(3080 魔改 / 4080S / 4090 / 5080 / 5090 / PRO 6000D / PRO 5000 / PRO 6000)的三维性价比测算:算力、显存、三年总成本