Back to search results
DeepSeek-R1-Distill-Qwen-14B · IQ4_XS
NVIDIA RTX 309024GBllama.cpp v0.42L2 Cross-framework verified
This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (IQ4_XS) on NVIDIA RTX 3090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
79.26 tok/s
Decode
Decode speed
346.57 tok/s
Prefill
Prefill speed
2.5 s
TTFT
Time to first token
22.8 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
17 days ago
Last verified
Performance
- llama.cpp · IQ4_XS (current)Decode 79.26 · Prefill 346.57 ·
- llama.cpp · Q3_K_M Decode 75.68 · Prefill 415.45 ·
- vLLM · Q3_K_M Decode 80.11 · Prefill 318.86 ·
- vLLM · Q2_K Decode 95.91 · Prefill 524.52 ·
- Ollama · IQ4_XS Decode 73.36 · Prefill 312.95 ·
- vLLM · Q5_K_M Decode 76.5 · Prefill 573.8 ·
- vLLM · Q4_K_M Decode 93.6 · Prefill 702 ·
- llama.cpp · Q5_K_M Decode 70.2 · Prefill 526.5 ·
- llama.cpp · Q4_K_M Decode 85.9 · Prefill 644.3 ·
Core figures
- Decode (avg)
- 79.26 tok/s
- Prefill (avg)
- 346.57 tok/s
- TTFT (avg)
- 2.5 s
- VRAM (avg)
- 22.8 GB
- MTP acceptance rate
- —
- TTFB
- 0.63 GB
- Power draw
- 343.1 W
Configuration
Member-level fields are taken from the most recent run
- Model
- DeepSeek-R1-Distill-Qwen-14B
- Quantization
- IQ4_XS
- Framework
- llama.cpp
- Version
- v0.42
- Context length
- 16384 tokens
- Batch size
- 512
- GPU layers
- 99
- Flash Attention
- Off
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 3090
- Nominal VRAM
- 24 GB
- Measured VRAM (avg)
- 22.8 GB
- OS
- Windows 11 24H2
- Driver
- 565.57
- CUDA
- 12.6
- Power draw
- 343.1 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.42 · Windows 11 24H2 · CUDA 12.6 · 16384 ctx
79.26 tok/s
Decode
346.57 tok/s
Prefill
2.5 s
TTFT
22.8 GB
VRAM
—
MTP
343.1 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=2 running model_id=71 at IQ4_XS