Back to search results
Qwen3-4B · IQ2_XS
NVIDIA RTX 308010GBllama.cpp v0.22L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B (IQ2_XS) on NVIDIA RTX 3080 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
131.47 tok/s
Decode
Decode speed
451.18 tok/s
Prefill
Prefill speed
2.49 s
TTFT
Time to first token
6.13 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
5 days ago
Last verified
Performance
- llama.cpp · IQ2_XS (current)Decode 131.47 · Prefill 451.18 ·
- llama.cpp · Q3_K_L Decode 99.23 · Prefill 538.67 ·
- Ollama · IQ2_XS Decode 118.69 · Prefill 621.83 ·
- Ollama · Q2_K Decode 102.28 · Prefill 452.09 ·
Core figures
- Decode (avg)
- 131.47 tok/s
- Prefill (avg)
- 451.18 tok/s
- TTFT (avg)
- 2.49 s
- VRAM (avg)
- 6.13 GB
- MTP acceptance rate
- —
- TTFB
- 1.05 GB
- Power draw
- 333.3 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B
- Quantization
- IQ2_XS
- Framework
- llama.cpp
- Version
- v0.22
- Context length
- 32768 tokens
- Batch size
- 512
- GPU layers
- 33
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 3080
- Nominal VRAM
- 10 GB
- Measured VRAM (avg)
- 6.13 GB
- OS
- macOS 15.0
- Driver
- 560.94
- CUDA
- 12.6
- Power draw
- 333.3 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.22 · macOS 15.0 · CUDA 12.6 · 32768 ctx
131.47 tok/s
Decode
451.18 tok/s
Prefill
2.49 s
TTFT
6.13 GB
VRAM
—
MTP
333.3 W
Power draw
L2 batch seed 2026-10-13: llama.cpp on hw_id=44 running model_id=54 at IQ2_XS