Machine

RX 9070 9070 Xt 9070 Gre · Ryzen 5 4500 #2d8cf276

89 runs of 28 models since Oct 4, 2026. Last run Oct 29, 2026.

ShareRedditX
89
runs
28
models
44
configs
20
configs run 2+ times

Hardware

GPU
AMD Radeon RX 9070/9070 XT/9070 GRE 16 GB
CPU
AMD Ryzen 5 4500 6-Core Processor 6 cores
RAM
23 GB
OS
Ubuntu 24.04.5 LTS
Tools
llama.cpp 81 · Ollama 8
Client
0.4.98
All AMD Radeon RX 9070/9070 XT/9070 GRE benchmarks

Models on this machine

One dot per model, at its best config. Up is better answers, right is faster. Time to finish counts every token the model wrote, thinking included, so a model that talks a lot ends up further left.

Best trade-off: quality 60+, and nothing else here is both faster and betterOther models · dot size = number of runs
below 20 tok/s: slower than you read4050607080901002050100200tok/s (log scale) →quality →

Medians of each model's best config, 27 models.

Best config per model

sort

Runs with the same model, quant, tool, context, thinking and KV cache count as one config. Numbers are the median of its runs, and each dot is one run on the 0–100 quality scale. ★ marks the best config with at least 2 rated runs.

ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF
5 runs · 3 configs
★ bestIQ3_XXS-mtpllama.cpp32k ctxthinking oneffort xhigh
×3 runs
85.672.5–87.0
varies
44.1 tok/s
42.5–44.5
ukisai/Swift-1.5-Qwen3.8-27B-GSQ-RCO-GGUF
5 runs · 3 configs
★ bestIQ3_S-mtpllama.cpp32k ctxthinking oneffort xhigh
×3 runs
85.283.3–85.3
steady
44.2 tok/s
43.9–44.4
prism-ml/Ternary-Bonsai-2-27B-gguf-PQ2_0_Vulkan
1 run · 1 config
★ bestQ2_0llama.cpp32k ctxthinking oneffort medium
84.8
1 rated run
43.9 tok/s
peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF-MTP
2 runs · 2 configs
★ bestUD-IQ3_XXSllama.cpp32k ctxthinking on
84.6
1 rated run
88.3 tok/s
unsloth/Muse-Glimmer-30B-GGUF
5 runs · 2 configs
★ bestUD-IQ3_XXSllama.cpp32k ctxthinking on
×3 runs
84.582.8–88.4
steady
37.5 tok/s
37.5–37.6
pfeifferj/Ornith-1.5-35B-A3B-GSQ-RCO-GGUF
10 runs · 4 configs
★ best3bitllama.cpp32k ctxthinking default
×5 runs
84.462.2–86.7
2 outliers1 slow run
88.2 tok/s
56.8–104
2 runs of the best config scored 65.3, 62.2, the other 3 sit between 84.4 and 86.7. The median isn't thrown off by them.
ConwayResearch/Underdog-Saluki-27B-1.0
1 run · 1 config
★ bestllama.cpp32k ctxthinking oneffort xhigh
84.3
1 rated run
43.5 tok/s
prism-ml/Ternary-Bonsai-2-27B-gguf-PQ2_0
1 run · 1 config
★ bestQ2_0llama.cpp32k ctxthinking oneffort medium
82.9
1 rated run
47.8 tok/s
IsValorum/Ornith-1.5-35B-A3B-APEX-I-NanoPlus-GGU
1 run · 1 config
★ bestllama.cpp32k ctxthinking default
82.2
1 rated run
75.6 tok/s
IsValorum/Ornith-1.5-35B-A3B-APEX-I-NanoPlus-GGUF
3 runs · 1 config
★ bestllama.cpp32k ctxthinking default
×3 runs
82.181.4–84.3
steady
110 tok/s
93.5–110

Compare with another machine

Run history

Show all 89 runs ▾