Leaderboard Β· All comparisons Β· LynxBenchAI 26Q3 Index

AMD Instinct MI300X vs NVIDIA H100 80GB HBM3 for AI

Device pages: AMD Instinct MI300X Β· NVIDIA H100 80GB HBM3

On the LynxBenchAI 26Q3 Index, the AMD Instinct MI300X scores 1.05x the NVIDIA H100 80GB HBM3 overall (Training 1.03x, Inference 0.78x, Compute 2.10x). It leads on Training and Compute and trails on Inference.

Scores on the LynxBenchAI 26Q3 Index

Ratio = the AMD MI300X's score divided by the H100 80GB HBM3's
ScoreRatioAMD MI300XH100 80GB HBM3
Overall (GT)1.05x986.72939.45
Training1.03x727.23709.35
Inference0.78x1304.041678.11
Compute2.10x1023.28487.13

When to Choose Which

  • Level on training: the two are within 10% (1.03x).
  • Choose the H100 80GB HBM3 for inference (1.29x the AMD MI300X).
  • Choose the AMD MI300X for raw compute (2.10x the H100 80GB HBM3).

From the LynxBenchAI 26Q3 Index category scores, which are not raw throughput; a lead under 10% counts as level.

Per Model and Precision

LynxBenchAI score per model and precision; ratio = AMD MI300X / H100 80GB HBM3
ModelPrecisionRatioAMD MI300XH100 80GB HBM3
compute
Dense MatMulBF160.93x1227.871315.06
Dense MatMulFP160.98x1238.821264.27
Dense MatMulFP321.07x1095.581026.69
Dense MatMulFP641.04x432.02414.37
Sparse MatMulFP328.21x1322.97161.05
Sparse MatMulFP642.76x774.09280.45
inference
DistilBERTFP161.08x2816.122611.72
DistilBERTFP81.12x1524.791362.71
DistilBERTINT80.87x946.391090.29
DistilGPT2FP160.87x3318.553797.65
DistilGPT2FP80.88x1689.361920.23
DistilGPT2INT80.73x1084.291489.31
Flan-T5 SmallFP160.88x3458.973919.63
Flan-T5 SmallFP81.04x1719.401658.94
Flan-T5 SmallINT80.77x1070.691389.31
GraphGPS (Peptides)FP161.07x1139.441060.04
GraphGPS (Peptides)FP81.06x582.67548.07
GraphGPS (Peptides)INT81.03x536.07520.52
MobileNetV3FP160.26x733.832832.98
MobileNetV3INT80.55x515.38931.43
training
DistilBERTBF161.18x1524.761291.98
DistilBERTFP321.07x1583.231480.99
DistilGPT2BF161.40x911.36651.85
DistilGPT2FP321.36x1571.281159.01
Flan-T5 SmallBF161.39x862.82622.63
Flan-T5 SmallFP321.35x1482.851095.97
GraphGPS (Peptides)BF160.81x313.78386.94
GraphGPS (Peptides)FP321.14x426.90374.00
MobileNetV3BF160.45x116.95257.12
MobileNetV3FP320.49x275.35564.13

Related Comparisons

Each figure is the first device's overall score divided by the second's.

Measure your own machine: pip install lynxbench-ai Β· Install guide β†’

How to Read These Results

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).

Questions

Is the AMD Instinct MI300X faster than the NVIDIA H100 80GB HBM3 for AI?
On the LynxBenchAI 26Q3 Index, the AMD Instinct MI300X scores 1.05x the NVIDIA H100 80GB HBM3 overall (Training 1.03x, Inference 0.78x, Compute 2.10x). It leads on Training and Compute and trails on Inference.
Should I choose the AMD Instinct MI300X or the NVIDIA H100 80GB HBM3?
Level on training: the two are within 10% (1.03x). Choose the H100 80GB HBM3 for inference (1.29x the AMD MI300X). Choose the AMD MI300X for raw compute (2.10x the H100 80GB HBM3). These follow the LynxBenchAI 26Q3 Index category scores, which are not raw throughput.
Where is the difference between the AMD Instinct MI300X and the NVIDIA H100 80GB HBM3 largest?
On the LynxBenchAI 26Q3 Index, the ratio is highest on Sparse MatMul at FP32 (Compute), where the AMD Instinct MI300X scores 8.21x the NVIDIA H100 80GB HBM3, and lowest on MobileNetV3 at FP16 (Inference), at 0.26x.
What does this result tell me about my own PyTorch model?
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.

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.

Get Started