gpu4.pics
data via LocalScore
NVIDIA RTX 5090
32 GB VRAM NVIDIA 575 W TDP
running Meta Llama 3.1 8B Instruct· Q4_K_M
66.3
tokens / sec
1,193 · Excellent #6 of 370
Generation
66.3tok/s
higher is better
Prompt / prefill
6,297tok/s
higher is better
Time to first token
266 ms
lower is better
Composite score
1,193
Model size
8B
Runtime
gpu4.pics/r/ls-nvidia-rtx-5090-meta-llama-31-8b-instruct-q4-k-m

All benchmarks on this rig

ModelQuantGenPromptTTFTPowerPerf/W$/MtokScore
Meta Llama 3.1 8B InstructQ4_K_M66.36,297266 ms1,193

How it stacks up — Meta Llama 3.1 8B Instruct

NVIDIA RTX 5090
66.3
NVIDIA H100
120
NVIDIA A100 80GB
110
NVIDIA RTX 3090 Ti
110
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
103
NVIDIA RTX 3090
103
NVIDIA RTX A6000
90.5