Qwen3-8B BF16 on 1× H200 (Lium, Jackson) with vLLM 0.33.0
Quick-mode measurement on a Lium Secure-tier marketplace node (f092246d, Jackson, United States), 2026-11-10, default vLLM settings, $5.30 per GPU-hour (the host's listed price). Same model, workload and latency target as our other Qwen3-8B reports; the node ran tokenwatt-check first (700 W power limit, its default; 668 TFLOPS BF16; 4.30 TB/s copy bandwidth). $0.37 per million output tokens at that price. Lium provided the credit for this run and checked the node details before publication. Energy is GPU board power.
Results
Each value is the median of 2 runs.
SLO goodput
Tokens per joule
Decode bandwidth utilization
Axis runs to 100% of rated bandwidth.
GPU board power
GPU board power only — excludes CPUs, memory, NICs, fans and PSU losses.
All numbers
Cards for equal goodput = 64 × reference goodput ÷ configuration goodput, rounded up.
| Metric | lium-f092246d-qwen3-8b |
|---|---|
| Engine build | vLLM 0.33.0 · KV BF16 |
| Decode bandwidth utilization (MBU) | 40% |
| GPU board power | 0.69 kW |
| SLO goodput | 4,014 tok/s |
| Tokens per joule | 5.80 |
| TTFT p99 | 417 ms |
| TPOT p99 | 36.9 ms |
| SLO met | yes |
| Accuracy (gsm8k-5shot) | 92.0 |
| Accuracy Δ vs. reference | reference |
| Cards for equal goodput | 64 |
Setup
Any change to these fields makes it a different report.
- Accelerator
- 1× NVIDIA H200
- Rated HBM BW
- 4.8 TB/s per GPU
- Driver / runtime
- 580.159.03
- Host
- Lium Secure-tier pod (marketplace host), node f092246d, Jackson, US
- Engine
- vLLM 0.33.0
- Kernel libraries
- vllm/vllm-openai:v0.33.0
- Model
- Qwen3-8B (Dense)
- Quantization
- BF16 weights, BF16 KV cache
- Parallelism
- 1 replica × TP1
- Workload trace
- chat-mix-v1
- Input tokens
- median 1,024 · p90 4,096
- Output tokens
- median 256 · p90 1,024
- Arrival
- Poisson, rate swept to SLO boundary
- Power source
- GPU board power sum (NVIDIA) · 1 Hz · 120 s steady state
- Runs per config
- 2 (variance band ±1.4%)
- Accuracy gate
- gsm8k-5shot · max drop 1.0 pts
Not covered
Do not extrapolate this report to the following.
- Whole-node power (no BMC access on this host)
- Context lengths beyond the chat-mix-v1 trace
- Quick mode: short windows, few repeats — validation only, not for publishing
Reproduce
Run on the same hardware and software versions. Results should fall within ±1.4%.
tokenwatt run --spec lambda-qwen3-8b-tp1-dp1.spec.yaml --label 'lium-f092246d-qwen3-8b' --role other --out runs/lium-f092246d-qwen3-8b.json tokenwatt report runs/lium-f092246d-qwen3-8b.json --title 'Qwen3-8B BF16 on 1× H200 (Lium, Jackson) with vLLM 0.33.0'