mammoth-1 · FP16
This page aggregates 1 real-world runs of mammoth-1 (FP16) on NVIDIA H20 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
18.55 tok/s
Decode
Decode speed
67.93 tok/s
Prefill
Prefill speed
0.56 s
TTFT
Time to first token
92.15 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
13 days ago
Last verified
Performance
- llama.cpp · FP16 (current)Decode 18.55 · Prefill 67.93 ·
- vLLM · IQ4_XS Decode 45.95 · Prefill 201.36 ·
Core figures
- Decode (avg)
- 18.55 tok/s
- Prefill (avg)
- 67.93 tok/s
- TTFT (avg)
- 0.56 s
- VRAM (avg)
- 92.15 GB
- MTP acceptance rate
- —
- TTFB
- 0.15 GB
- Power draw
- 362.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- mammoth-1
- Quantization
- FP16
- Framework
- llama.cpp
- Version
- v0.49
- Context length
- 2048 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)
- 92.15 GB
- OS
- Ubuntu 24.04
- Driver
- 570.86
- CUDA
- 12.4
- Power draw
- 362.2 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
v0.49 · Ubuntu 24.04 · CUDA 12.4 · 2048 ctx
18.55 tok/s
Decode
67.93 tok/s
Prefill
0.56 s
TTFT
92.15 GB
VRAM
—
MTP
362.2 W
Power draw
L2 batch seed 2026-10-22: llama.cpp on hw_id=56 running model_id=94 at FP16