llama.cpp PR Uses AVX2 to Speed Up Large Batch IQ Quantization
pmttyji · reddit · 2026-08-20
A new PR in llama.cpp uses AVX2 instructions to significantly speed up inference for IQ-quantized models at large batch sizes. Benchmarks on Qwen3.6-27B and 35B-A3B show token processing speed improvements of 6x to over 12x for IQ series quantizations (e.g., IQ3S, IQ2XS) with negligible impact on perplexity.
More from Infra
- Haight Street Compute Market: a one-day trading floor for GPU deals in SF — ctjlewis · 2026-10-16
- Temporal Serverless Workers on Cloud Run: queue-driven autoscaling, scale to zero — rseroter · 2026-10-16
- LMA brings day-0 model support to TPU for the first time, spanning NVIDIA, AMD, Intel via SGLang — ying11231 · 2026-10-16
- Semtech's analyst-day slide cleanly explains the four signal-integrity impairments — BenBajarin · 2026-10-16
- Fortytwo Engine boosts Qwen3.8-27B accuracy 7% and cuts tokens 25% via Swarm Inference — rohanpaul_ai · 2026-10-16
- Berkeley scholars: concentrated chip supply chain makes a hardwired frontier AI pause feasible — wfithian · 2026-10-16