inference/research
verifiedlow · 0.50updated 2026-10-14

Hardware Corner — RTX 5090 32GB local LLM llama-bench run (Qwen3 Q4_K_XL context sweep)

benchmark

Fields

id
sources/hardware-corner-rtx-5090-llm-benchmarks
Publisher
Hardware Corner (Allan Witt)
Kind
benchmark
Accessed
2026-10-14
notes
llama-bench build 466c1911 on RTX 5090 32GB GDDR7, EPYC 7642 host, Ubuntu 24.04, -fa 1 -ngl 99 -p 1024/2048 -n 128, -d context-depth sweep 4k..131k on Qwen3 8B/14B/30B-A3B/32B Q4_K_XL + gpt-oss 20B. Per-request batch-1.

Notes

llama-bench build 466c1911 on RTX 5090 32GB GDDR7, EPYC 7642 host, Ubuntu 24.04, -fa 1 -ngl 99 -p 1024/2048 -n 128, -d context-depth sweep 4k..131k on Qwen3 8B/14B/30B-A3B/32B Q4_K_XL + gpt-oss 20B. Per-request batch-1.

Cross-references

This record references no other records.