Dual 3090 Qwen3.8-27B: DFlash2 Speed Test and Full Config Guide
Old_Ad_6033 · reddit · 2026-08-29
A Redditor shared a full recipe for running the Qwen3.8-27B INT4 model on dual RTX 3090s (24GB). Using a stack of vLLM 0.28.0, LMCache, and DFlash2, the setup leverages NVMe as a Level 2 KV cache to support 256k context length. Benchmarks show that with DFlash2 enabled, decode speed for Code tasks hits 268 tok/s (3.8× speedup) and real agent jobs reach 165 tok/s. The author detailed the environment specs, LMCache server startup commands, and vLLM launch parameters, noting that specific patches are required to prevent cache corruption.
More from coding & agent
- How to hook Claude Code or Codex up to iMessage in 3 steps — jacob_posel · 2026-10-16
- Are the fastest GPU kernels now written by AI? Kernels are hyper-verifiable, argues one dev — fleetwood___ · 2026-10-16
- Anthropic lead engineer's 4-question prompt makes agents more ambitious, now a free skill — daniel_mac8 · 2026-10-16
- ChainForge dev shows 'feature map' for orchestrating subagents on large PRs — IanArawjo · 2026-10-16
- aiDotEngineer benchmarks agent orchestration pipelines, staging humans as Foreman and workers — vikvang1 · 2026-10-16
- Developer mass-produces 200 videos in hours using Claude Code + Hyperframes — mcraddock · 2026-10-16