Back to search results
Qwen3-8B · IQ2_M
NVIDIA RTX 509032GBvLLM v0.31L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-8B (IQ2_M) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
325.14 tok/s
Decode
Decode speed
1,583.07 tok/s
Prefill
Prefill speed
0.99 s
TTFT
Time to first token
10.84 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · IQ4_XS Decode 256.42 · Prefill 1150.82 ·
- llama.cpp · Q5_K_M Decode 221.44 · Prefill 1381.14 ·
- vLLM · IQ3_S Decode 285.95 · Prefill 1301.24 ·
- llama.cpp · Q6_K Decode 184.6 · Prefill 725.28 ·
- vLLM · IQ2_M (current)Decode 325.14 · Prefill 1583.07 ·
- vLLM · Q2_K Decode 357.71 · Prefill 1294.78 ·
- vLLM · IQ4_XS Decode 252.51 · Prefill 829.1 ·
- Ollama · Q5_K_M Decode 172.9 · Prefill 940.17 ·
- llama.cpp · Q8_0 Decode 162.5 · Prefill 1218.8 ·
- llama.cpp · Q4_K_M Decode 285.3 · Prefill 2139.8 ·
Core figures
- Decode (avg)
- 325.14 tok/s
- Prefill (avg)
- 1,583.07 tok/s
- TTFT (avg)
- 0.99 s
- VRAM (avg)
- 10.84 GB
- MTP acceptance rate
- —
- TTFB
- 2.6 GB
- Power draw
- 295.8 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-8B
- Quantization
- IQ2_M
- Framework
- vLLM
- Version
- v0.31
- Context length
- 8192 tokens
- Batch size
- 32
- GPU layers
- 999
- Flash Attention
- Off
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 5090
- Nominal VRAM
- 32 GB
- Measured VRAM (avg)
- 10.84 GB
- OS
- Ubuntu 22.04
- Driver
- 570.86
- CUDA
- 12.8
- Power draw
- 295.8 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.31 · Ubuntu 22.04 · CUDA 12.8 · 8192 ctx
325.14 tok/s
Decode
1,583.07 tok/s
Prefill
0.99 s
TTFT
10.84 GB
VRAM
—
MTP
295.8 W
Power draw
L2 batch seed 2026-10-16: vLLM on hw_id=13 running model_id=53 at IQ2_M