NVIDIA · GPU · LynxBenchAI 26Q3 Index, recorded 2026-10-28
On the LynxBenchAI 26Q3 Index, the NVIDIA L40S ranks 17 of 48 devices overall with a GT of 395.82 (Training 316.04, rank 17; Inference 528.12, rank 17; Compute 344.33, rank 12); overall it scores 0.42x the NVIDIA H100 80GB HBM3 and 0.88x the NVIDIA GeForce RTX 4090.
Ada Lovelace architecture · 48 GB GDDR6 · PCIe card (NVIDIA specifications)
Rank among all 48 devices; further right is higher.
| Compared with | Overall | Training | Inference | Compute |
|---|---|---|---|---|
| NVIDIA H100 80GB HBM3 | 0.42x | 0.45x | 0.31x | 0.71x |
| NVIDIA GeForce RTX 4090 | 0.88x | 0.94x | 0.80x | 0.94x |
| Score | Next above | Ratio | Next below | Ratio |
|---|---|---|---|---|
| Training | NVIDIA GeForce RTX 4090 | 0.94x | NVIDIA L40 | 1.02x |
| Inference | NVIDIA A100 80GB PCIe | 0.90x | NVIDIA L40 | 1.02x |
| Compute | NVIDIA H100 NVL | 0.99x | NVIDIA L40 | 1.02x |
| Model | Precision | Sustained |
|---|---|---|
| MobileNetV3 | FP32 | 376.00 |
| MobileNetV3 | BF16 | 248.23 |
| Flan-T5 Small | FP32 | 470.80 |
| Flan-T5 Small | BF16 | 263.33 |
| DistilGPT2 | FP32 | 391.29 |
| DistilGPT2 | BF16 | 223.21 |
| DistilBERT | FP32 | 492.41 |
| DistilBERT | BF16 | 405.03 |
| GraphGPS (Peptides) | FP32 | 207.50 |
| GraphGPS (Peptides) | BF16 | 184.43 |
| Model | Precision | Sustained |
|---|---|---|
| MobileNetV3 | FP16 | 1375.69 |
| MobileNetV3 | INT8 | 403.39 |
| Flan-T5 Small | FP16 | 1009.63 |
| Flan-T5 Small | INT8 | 503.01 |
| Flan-T5 Small | FP8 | 566.59 |
| DistilGPT2 | FP16 | 752.18 |
| DistilGPT2 | INT8 | 420.32 |
| DistilGPT2 | FP8 | 457.31 |
| DistilBERT | FP16 | 898.53 |
| DistilBERT | INT8 | 438.15 |
| DistilBERT | FP8 | 514.72 |
| GraphGPS (Peptides) | FP16 | 268.17 |
| GraphGPS (Peptides) | INT8 | 138.11 |
| GraphGPS (Peptides) | FP8 | 137.60 |
| Model | Precision | Sustained |
|---|---|---|
| Dense MatMul | FP32 | 303.04 |
| Dense MatMul | FP16 | 243.74 |
| Dense MatMul | BF16 | 243.02 |
| Dense MatMul | FP64 | 8.68 |
| Sparse MatMul | FP32 | 575.74 |
| Sparse MatMul | FP64 | 577.96 |
LynxBenchAI 26Q3 Index · GT = LynxBenchAI Global Score. Every precision result is sustained throughput under a declared, bounded optimisation budget — no collapsed single number below the category level.
Each figure is the first device's overall score divided by the second's.
A LynxBenchAI score is sustained throughput on a fixed catalogue of PyTorch models, measured after a warm-up at a batch size raised until throughput stops improving, so it is what the machine holds under load, not its best instant. Training, Inference and Compute are three separate scores; GT combines them (Training and Inference weigh 0.4 each, Compute 0.2).
Every run uses the stock, publicly released PyTorch stack, so a result is broadly indicative of the out-of-the-box PyTorch experience on that hardware, and so of custom PyTorch models too. It does not run your own model, and it is not a vendor-tuned peak like an MLPerf submission.
Scores compare only within one Index: the 26Q3 Index fixes the models, precisions, correctness thresholds and scoring, and a later Index may change them. Each test is one timed window, the scores have no 0-100 scale, and a precision a device cannot run scores zero rather than being skipped.
LynxBenchAI is built by TechnoLynx, which the European Commission engaged as an independent expert advisor on AI performance and benchmarking for AI Gigafactories under the EuroHPC Joint Undertaking (case study).
Measure your own NVIDIA L40S: pip install lynxbench-ai · Install guide →
The Personal Edition submits your benchmark results to TechnoLynx's servers as part of participating in the public leaderboard. If that isn't acceptable for your use case, contact TechnoLynx about Press, Pro, or Enterprise licenses.
LynxBenchAI End-User License Agreement · LynxBenchAI Privacy Policy