A3B cross-machine — Predator side
TL;DR
12 request records in the retained JSONL. Predator · GTX 1060 6 GB · Qwen3-30B-A3B IQ2_M: 3.87 tok/s mean; median request 101.9 s. Pavilion and Predator used different quantizations. These results do not isolate hardware or offload effects.
What this run shows
This Predator run recorded 12 request records with Qwen3-30B-A3B IQ2_M. The summary reports 3.87 tokens per second mean and a 101.9-second median request. The paired Pavilion record uses IQ2_XXS, so the difference cannot establish a hardware speed ranking or isolate an offload bottleneck.
Methodology
Request counts were checked against the retained JSONL on 21 October 2026, excluding telemetry and skipped records. Timing aggregates remain those published in the source. Count audit and original summary counts.
See Public methodology and measurement limits
for the full procedure.
Recorded source revision (where available): —.
Results
| Cell | tok/s mean | tok/s p50 | tok/s p95 | duration p50 | calls |
|---|---|---|---|---|---|
| qwen3:30b-a3b-iq2m-think500 | 3.9 | 4.0 | 4.1 | 1m41s | 12 |
tokens per second — mean · p50 · p95
Cold start vs warm
Cold and warm are the source harness labels for first and subsequent calls in a sequence. The published labels alone do not prove a fresh process, disk reload or cache reset.
| Cell | cold n | cold tok/s | cold p50 | warm n | warm tok/s | warm p50 | warm/cold |
|---|---|---|---|---|---|---|---|
| predator:llamacpp:qwen3:30b-a3b-i… | 3 | 3.5 | 1m37s | 9 | 4.0 | 1m46s | 1.16× |
By prompt difficulty
Tokens per second by prompt class. hello is a trivial
one-line prompt; P-MEDIUM and P-HARDGeneration rates and durations by prompt class. These show the observed differences; they do not by themselves isolate prompt-processing versus generation bottlenecks.
| Cell | hello | P-MEDIUM | P-HARD |
|---|---|---|---|
| predator:llamacpp:qwen3:30b-a3b-i… | 3.6 tok/s 16.1s · n=4 | 4.0 tok/s 1m41s · n=4 | 4.0 tok/s 3m16s · n=4 |
Reasoning vs answer
Thinking models split their output into a hidden reasoning trace and a visible answer. The ratio shows how much of the budget the model spent thinking vs answering.
| Cell | reasoning chars | answer chars | reasoning / answer |
|---|---|---|---|
| predator:llamacpp:qwen3:30b-a3b-i… | 1081 | 835 | 1.29× |
Per-call timeline
Every call placed during this run, in order, colored by phase. Width is proportional to the call’s share of the cell’s wall-time. Hover any segment for the prompt id and tok/s.
Raw data
Every run gets its JSONL, log, summary, and metadata published. Clone the archive; re-run it; tell us where we got it wrong.
Cite
Margetic, S. et al. (2026). benchmarks.weeyuga.com/benchmarks/5fb2913d.html Public benchmarks of the Weeyuga cluster. Run id: 5fb2913d-6500-4ecf-9e97-d43f7dd61145. SHA —.