Shareable technical brief · July 2026

H100 prefill throughput and board-energy comparison.

Measured head-to-head on the same hardware, same shape. Version 100 internal measurement.

Summary

On a matched prefill-heavy workload, the LuxiEdge energy/throughput path is both faster and more energy-efficient (lower board joules per token position) than a current open serving stack (vLLM), while holding deterministic dual-run behavior on the Luxi path.

QuestionAnswer
Faster?Yes - about 1.17–1.18× higher throughput
Better or equal energy per token?Yes - about 10–14% lower board J per position
Determinism held (Luxi TRADE path)?Yes - dual-run score 1.0

Peer stack used greedy sampling (temperature = 0), not a separate “deterministic product mode.” Token accounting is matched prefill positions.

Open the public data pack →

Protocol (what was compared)

SettingValue
Model classQwen2-7B-Instruct (FP16)
Sequence length128
Batch sizes16 and 32
WorkloadPrefill-heavy (one generated token per iteration; positions = iterations × batch × 128)
HardwareSingle NVIDIA H100 80GB-class GPU
ComparisonSequential arms (not concurrent on the same GPU)
MetricsThroughput (positions/s), board energy (NVML joules / position), dual-run determinism

Peer stack: vLLM (current release in test), tensor parallel 1, prefix cache off, same token accounting.

Luxi path: production energy/throughput configuration (Flash-class attention control + device-resident multi-layer stack + FP16 weight residency). Not the high-fidelity AUDIT receipt lane.

Results

BatchSystemThroughput (pos/s)Board J/positionDeterminism
16LuxiEdge~41,700~0.01711.0
16vLLM~35,800~0.01901.0
32LuxiEdge~43,900~0.01581.0
32vLLM~37,100~0.01821.0

Ratios (Luxi / vLLM)

BatchThroughput ratioEnergy ratio (lower is better)
161.17×0.90×
321.18×0.86×

Board power was measured under load via standard GPU power/energy interfaces (NVML). Energy is board joules, not facility wall-plug AC.

How to read this

  • Throughput counts prompt positions under a fixed sequence length and batch (prefill-oriented), not chat tokens/s under a full OpenAI-compatible server.
  • J/position is board energy per useful position under that same definition.
  • Determinism is dual-run agreement on the measured path (same inputs → same reported behavior).
  • This is an absolute, same-day head-to-head on one GPU class - not a claim against every serving recipe, every model, or every sequence length.

Why it matters

For operators, the useful question is rarely “who wins a microkernel.” It is:

  1. Do we move more useful work per second at commercial batch?
  2. Do we spend fewer joules per unit of that work?
  3. Can we still stand behind reproducible behavior?
On this shape, the answer to all three is yes relative to the peer stack tested.

Non-claims

  • Not a full multi-tenant serving stack comparison (scheduling, continuous batching product surface, multi-GPU TP/PP).
  • Not wall-plug or PUE.
  • Not bit-exact identity between energy path and high-fidelity audit path (those remain separate product lanes).
  • Not a claim of higher open-chat quality than the peer model implementation.

Contact

LuxiEdge · e@ewaller.com
For deeper diligence packs (methods, multi-run series, broader shapes), contact for a controlled evaluation.