Field report: GLM-5.3-Flash dual-rail on real 2× GB10 (GX10 + Spark) with new launcher knobs
EAccelerate_42 · x · 2026-09-26
A developer submitted a field-report PR to MiaAI-Lab/GLM-5.3-Flash-EXL3-2x-DGX-Sparks (647 stars) validating prior optimizations on real hardware:
- Hardware numbers for #251 (staged auto-loader weights + hybrid prefix caching), which had merged without a live boot, measured on a real 2× GB10 dual-rail setup (GX10 + DGX Spark).
- Three opt-in launcher knobs in start.sh for mixed and dual-rail pairs, defaults unchanged, documented in .env.example.
- Three findings from tuning the same hardware pair.
Related event: Dual GB10 cluster tuning boosts GLM-5.3-Flash 32k prefill by 20%(5 posts)→
More from Infra
- A100 SXM4 connector production ends, forcing new tooling investment — EAccelerate_42 · 2026-09-26
- Two local AIs talk to each other with no cloud: hands-on with Braid — Scobleizer · 2026-09-26
- The handiest GPU this dev ever bought is a ~$300 Intel Arc A310, not NVIDIA or AMD — TheZachMueller · 2026-09-26
- Running image generation in the browser on local hardware: 10s pixel art on an RTX 3060 — Bartholomheow · 2026-09-26
- Alibaba's T-Head unveils Zhenwu V900 chip: 216GB per card, sales in Q1 2027 — shashib · 2026-09-26
- llama.cpp fork dedups repeated prompts losslessly, cutting 108k to 71k tokens in agent loops — Odd_Cauliflower_8004 · 2026-09-26