Back to search results
Llama-3.2-3B-Instruct · Q4_K_M
NVIDIA RTX 509032GBStrata v0.35L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (Q4_K_M) on NVIDIA RTX 5090 with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.
367.38 tok/s
Decode
Decode speed
1,255.45 tok/s
Prefill
Prefill speed
1.35 s
TTFT
Time to first token
7.27 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
Today
Last verified
Performance
- llama.cpp · Q2_K Decode 325.57 · Prefill 1657.68 ·
- llama.cpp · Q4_K_S Decode 376.12 · Prefill 1436.24 ·
- Strata · Q4_K_M (current)Decode 367.38 · Prefill 1255.45 ·
- Strata · Q5_K_S Decode 289.95 · Prefill 1022.85 ·
- llama.cpp · Q8_0 Decode 436.8 · Prefill 3276 ·
- llama.cpp · Q5_K_M Decode 635.3 · Prefill 4764.8 ·
- llama.cpp · Q4_K_M Decode 776.5 · Prefill 5823.8 ·
Core figures
- Decode (avg)
- 367.38 tok/s
- Prefill (avg)
- 1,255.45 tok/s
- TTFT (avg)
- 1.35 s
- VRAM (avg)
- 7.27 GB
- MTP acceptance rate
- 0.57
- TTFB
- 2.94 GB
- Power draw
- 452.8 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3.2-3B-Instruct
- Quantization
- Q4_K_M
- Framework
- Strata
- Version
- v0.35
- Context length
- 4096 tokens
- Batch size
- 512
- GPU layers
- 8
- 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)
- 7.27 GB
- OS
- Ubuntu 22.04
- Driver
- 560.94
- CUDA
- 12.8
- Power draw
- 452.8 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.35 · Ubuntu 22.04 · CUDA 12.8 · 4096 ctx
367.38 tok/s
Decode
1,255.45 tok/s
Prefill
1.35 s
TTFT
7.27 GB
VRAM
0.57
MTP
452.8 W
Power draw
L2 batch seed 2026-10-18: Strata on hw_id=13 running model_id=70 at Q4_K_M