Benchmark search · Arao AI

Try these searches

3 matching benchmark runs

ModelQwen3.8-27BHardware: AllFramework: AllQuantization: AllEvidence level: AllSource: All

Qwen3.8-27B · Q6_K

L0 Self-reported

AMD Radeon RX 7900 GRE + NVIDIA GTX 1080Ti · 27GB · llama.cpp

Today

12.53 tok/s

Decode

116.21 tok/s

Prefill

— s

TTFT

23 GB

VRAM

Config (latest run):

Ubuntu Q6_K llama.cpp 10453

Sources: Reddit · 1 runs · 1 independent sources

CompareView details

Qwen3.8-27B · Q4_K_M

L0 Self-reported

AMD Radeon RX 7900 GRE + NVIDIA GTX 1080Ti · 27GB · llama.cpp

Today

11.13 tok/s

Decode

71.46 tok/s

Prefill

— s

TTFT

17.2 GB

VRAM

Config (latest run):

Ubuntu Q4_K_M llama.cpp 10453 Flash Attention

Sources: Reddit · 1 runs · 1 independent sources

CompareView details

Qwen3.8-27B · UD-Q8_K_XL

L0 Self-reported

2x NVIDIA RTX 3090 · 48GB · llama.cpp

GitHub

Today

43.65 tok/s

Decode

1,600.18 tok/s

Prefill

0.42 s

TTFT

44.12 GB

VRAM

Config (latest run):

Ubuntu 26.04 LTS CUDA 13.3 UD-Q8_K_XL llama.cpp b10236

Sources: GitHub · 1 runs · 1 independent sources

CompareView details

3 results