Back to search results
gemma-4-12B-it · IQ2_XS
NVIDIA H2097GBllama.cpp v0.30L2 Cross-framework verified
This page aggregates 1 real-world runs of gemma-4-12B-it (IQ2_XS) on NVIDIA H20 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
63.1 tok/s
Decode
Decode speed
298.51 tok/s
Prefill
Prefill speed
1.38 s
TTFT
Time to first token
14.2 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · IQ2_XS (current)Decode 63.1 · Prefill 298.51 ·
- vLLM · Q3_K_L Decode 49.15 · Prefill 244.04 ·
Core figures
- Decode (avg)
- 63.1 tok/s
- Prefill (avg)
- 298.51 tok/s
- TTFT (avg)
- 1.38 s
- VRAM (avg)
- 14.2 GB
- MTP acceptance rate
- —
- TTFB
- 0.5 GB
- Power draw
- 327 W
Configuration
Member-level fields are taken from the most recent run
- Model
- gemma-4-12B-it
- Quantization
- IQ2_XS
- Framework
- llama.cpp
- Version
- v0.30
- Context length
- 8192 tokens
- Batch size
- 512
- GPU layers
- 20
- Flash Attention
- Off
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA H20
- Nominal VRAM
- 97 GB
- Measured VRAM (avg)
- 14.2 GB
- OS
- Ubuntu 24.04
- Driver
- 560.94
- CUDA
- 12.6
- Power draw
- 327 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.30 · Ubuntu 24.04 · CUDA 12.6 · 8192 ctx
63.1 tok/s
Decode
298.51 tok/s
Prefill
1.38 s
TTFT
14.2 GB
VRAM
—
MTP
327 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=56 running model_id=65 at IQ2_XS