GitHub · Cuda

Megla123/Scream-P100

😱 Scream P100: big MoE LLMs on two retired, screaming Tesla P100s. Qwen3.8-Flash-Next & Edge0-145B, 512k context, ~67 tok/s writing, 110+ tok/s edits, ~1,000 tok/s reading, exact by default. A Pascal fork of Strata's split engine. Test build for 2× Tesla M40 24 GB on branch m40-2x.

Stars
12
Forks
3
Stars gained in 24 hours
+5
Forks gained in 24 hours
+2

Topics: cuda, gguf, homelab, llm-inference, local-llm, maxwell, mixture-of-experts, nvidia, openai-compatible, pascal

Latest data point: Nov 22, 2026 · 12 stars · 3 forks

Open on GitHub