Qwen3.8-27B · IQ2_XS
This page aggregates 1 real-world runs of Qwen3.8-27B (IQ2_XS) on NVIDIA H20 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
42.1 tok/s
Decode
Decode speed
229.32 tok/s
Prefill
Prefill speed
2.23 s
TTFT
Time to first token
29.32 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · IQ2_XS (current)Decode 42.1 · Prefill 229.32 ·
- Strata · Q2_K Decode 52.83 · Prefill 205.52 ·
- Strata · Q4_K_M Decode 39.14 · Prefill 148.55 ·
Core figures
- Decode (avg)
- 42.1 tok/s
- Prefill (avg)
- 229.32 tok/s
- TTFT (avg)
- 2.23 s
- VRAM (avg)
- 29.32 GB
- MTP acceptance rate
- —
- TTFB
- 0.34 GB
- Power draw
- 387.3 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.8-27B
- Quantization
- IQ2_XS
- Framework
- llama.cpp
- Version
- v0.50
- Context length
- 2048 tokens
- Batch size
- 256
- 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)
- 29.32 GB
- OS
- Ubuntu 24.04
- Driver
- 572.83
- CUDA
- 12.8
- Power draw
- 387.3 W
Get started
Original model resources
Base model resources may not match this quantization. Use the verified deployment weights above when available.
Sources & evidence
1 measured records in total, each traceable to its original source
v0.50 · Ubuntu 24.04 · CUDA 12.8 · 2048 ctx
42.1 tok/s
Decode
229.32 tok/s
Prefill
2.23 s
TTFT
29.32 GB
VRAM
—
MTP
387.3 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=56 running model_id=5 at IQ2_XS