Back to search results
DeepSeek-R1-Distill-Qwen-14B · IQ3_XS
NVIDIA RTX 509032GBvLLM v0.40L2 Cross-framework verified
This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (IQ3_XS) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
290.25 tok/s
Decode
Decode speed
1,286.52 tok/s
Prefill
Prefill speed
1.62 s
TTFT
Time to first token
20.33 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
Today
Last verified
Performance
- llama.cpp · IQ1_S Decode 313.16 · Prefill 1103.72 ·
- vLLM · IQ3_XS (current)Decode 290.25 · Prefill 1286.52 ·
- vLLM · Q5_K_S Decode 193.3 · Prefill 1031.83 ·
- vLLM · Q5_K_M Decode 146.5 · Prefill 1098.8 ·
- vLLM · Q4_K_M Decode 179.2 · Prefill 1344 ·
- llama.cpp · Q5_K_M Decode 134.4 · Prefill 1008 ·
- llama.cpp · Q4_K_M Decode 164.4 · Prefill 1233 ·
Core figures
- Decode (avg)
- 290.25 tok/s
- Prefill (avg)
- 1,286.52 tok/s
- TTFT (avg)
- 1.62 s
- VRAM (avg)
- 20.33 GB
- MTP acceptance rate
- —
- TTFB
- 2.32 GB
- Power draw
- 403.6 W
Configuration
Member-level fields are taken from the most recent run
- Model
- DeepSeek-R1-Distill-Qwen-14B
- Quantization
- IQ3_XS
- Framework
- vLLM
- Version
- v0.40
- Context length
- 32768 tokens
- Batch size
- 8
- GPU layers
- 999
- Flash Attention
- On
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)
- 20.33 GB
- OS
- Ubuntu 22.04
- Driver
- 570.86
- CUDA
- 13.0
- Power draw
- 403.6 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.40 · Ubuntu 22.04 · CUDA 13.0 · 32768 ctx
290.25 tok/s
Decode
1,286.52 tok/s
Prefill
1.62 s
TTFT
20.33 GB
VRAM
—
MTP
403.6 W
Power draw
L2 batch seed 2026-10-18: vLLM on hw_id=13 running model_id=71 at IQ3_XS