Qwen3.8-27B runs 100k-token physics simulation on a 16GB GPU
1000_bucks_a_month · reddit · 2026-08-26
A developer ran a heavily quantized Qwen3.8-27B model (IQ3XXS) on an older 16GB Quadro RTX 5000, tasking it with implementing the coherent optical transfer-matrix method (TMM) for multilayer films from scratch. The session lasted about 100 minutes, generating 108,000 output tokens and undergoing three context compaction attempts.
Key Details:
- Hardware: NVIDIA Quadro RTX 5000 (16GB), Intel Xeon Silver 4116, 256GB RAM.
- Config: Used Unsloth Dynamic V3 quantization with a custom DFlash2 llama.cpp build and Flash Attention enabled.
- Process: The model produced 43,000 tokens in the first 27 minutes and wrote tmm.py. The core implementation was correct early on, but the model spent the remaining time debugging its own broken validators, repeatedly re-deriving conventions and getting stuck in contradictions.
- Conclusion: Despite heavy quantization and VRAM constraints, the model can handle hardcore scientific computing over extremely long contexts, though its self-consistency verification process can be highly inefficient.
More from coding & agent
- Dev bills $20,000 for an app built with $200 of Claude Code — IgorCarron · 2026-10-15
- Dan Shipper: run product and labs in parallel, even with a labs team of one — danshipper · 2026-10-15
- Gemini 4 Argon wrote 5,000+ lines of pure Python to build its own ray-tracing 3D engine — aitrendz_xyz · 2026-10-15
- Minecraft fully ported from Java to Rust, file by file, server and client included — Vjeux · 2026-10-15
- mitsuhiko builds a full Python IDE for iOS in an afternoon with new AI models — mitsuhiko · 2026-10-15
- How to get Gemini 4 Argon to build games that actually look decent — chongdashu · 2026-10-15