wrapped·Jul 02, 2026
/r/r_6ahy-dq0f_0your run
qwen2.5-coder
on RTX 3090
69.2
tok/sdecode
RTX 3090 (24GB)
rank in tier
14/25RTX 3090 runstop 56%
best workload
chat-short
where the rig flew
slowest workload
agent-trace63.6 tok/s
where the rig struggled
backend
ollamaollama is llama.cpp under the hood; pure llama.cpp tends to nudge a few % faster.
faster than
- starcoder2-15b-4bit on M3 Ultra64.2 tok/s
- Ripple-65B-Preview-AD on RTX 509056.3 tok/s
- Muse-Glimmer-30B-ctx8k on RTX 409053.7 tok/s