NVIDIA GPU accelerators transform general-purpose servers into AI/ML powerhouses. This guide compares the three most common data center GPUs.
GPU Comparison
| Feature | Tesla T4 | A100 (80GB) | H100 (80GB) |
|---|---|---|---|
| Architecture | Turing | Ampere | Hopper |
| FP32 TFLOPS | 8.1 | 19.5 | 67 (SXM) |
| Memory | 16GB GDDR6 | 80GB HBM2e | 80GB HBM2e (PCIe) / HBM3 (SXM5) |
| TDP | 70W | 300W | 350W (PCIe) / 700W (SXM) |
| Form Factor | Low-profile | PCIe / SXM4 | PCIe / SXM5 |
| Best For | Inference, VDI | Training, HPC | Large model training |
| Server Fit | Any 1U/2U | 2U+ GPU servers | 4U GPU servers |
GPU cross-reference across brands | Find GPUs for your server