On Windows, new LLM inference engines all lose to plain llama.cpp in this RTX 5090 test
No-Bite2555 · reddit · 2026-11-17
A Reddit user with an RTX 5090 (24GB) Windows 11 laptop benchmarked many new inference engines — nInfer, basalt, rusting strata, exLlamaV3 and llama.cpp forks — and found all noticeably slower than his stack of Strata for Qwen 3.8 Flash Next and llama.cpp for everything else. Even using Claude and Codex to tune configs didn't help, leading him to suspect these engines only hit their paper performance on Linux.
More from Infra
- Beyond solar capture: the case for 'power liquidity' at point of use — edgarpavlovsky · 2026-11-17
- Drogue Raises Pre-Seed From Garuda VC to Power AI Data Centers With Ocean Thermal Energy — arian_ghashghai · 2026-11-17
- Cloud Cost Rule of Thumb: $1,000/Month Equals About $0.11/Hour — StewartalsopIII · 2026-11-17
- Analyst: Data Center Builder $BE Can Scale Operations Faster Than Expected — BenBajarin · 2026-11-17
- NVIDIA unveils Vera CPU rack-scale HPC platform with 2x x86 energy efficiency — luisceze · 2026-11-17
- Microsoft engineer on cloud outages: rigor and redundancy feed a failure loop — tianyin_xu · 2026-11-17