GPU Efficiency Rankings
Raw TFLOPS do not tell the whole story. A GPU that draws 600W to deliver 100 TFLOPS is less efficient than one that delivers 80 TFLOPS at 300W. We rank every GPU in our catalog by three efficiency metrics using real spec data.
TFLOPS per Watt measures performance per unit of power — critical for large clusters and sustainability.TFLOPS per GB measures compute density — important when memory bandwidth is the bottleneck.Combined divides per-watt by VRAM for the ultimate efficiency score.
Choose an efficiency lens
Filter the catalog
Efficiency ranking
| Rank | GPU | TFLOPS/W |
|---|---|---|
| #1 | AMD Instinct MI300X CDNA 3 | 1.743 |
| #2 | NVIDIA L4 Ada Lovelace | 1.681 |
| #3 | NVIDIA RTX 6000 Ada Ada Lovelace | 1.217 |
| #4 | NVIDIA B200 Blackwell | 1.125 |
| #5 | NVIDIA L40S Ada Lovelace | 1.046 |
| #6 | NVIDIA GB200 Blackwell | 0.938 |
| #7 | NVIDIA T4 Turing | 0.929 |
| #8 | NVIDIA B300 Blackwell Ultra | 0.846 |
| #9 | NVIDIA A10 Ampere | 0.833 |
| #10 | NVIDIA A100 Ampere | 0.780 |
| #11 | NVIDIA RTX 5090 Blackwell | 0.729 |
| #12 | NVIDIA H100 Hopper | 0.707 |
| #13 | NVIDIA H200 Hopper | 0.707 |
| #14 | NVIDIA H200 NVL Hopper | 0.697 |
| #15 | AMD Instinct MI250X CDNA 2 | 0.684 |
| #16 | NVIDIA RTX 5080 Blackwell | 0.625 |
| #17 | NVIDIA RTX A4000 Ampere | 0.550 |
| #18 | NVIDIA RTX A6000 Ampere | 0.517 |
| #19 | NVIDIA A40 Ampere | 0.500 |
| #20 | NVIDIA GH200 Hopper | 0.495 |
| #21 | NVIDIA RTX A5000 Ampere | 0.483 |
| #22 | NVIDIA A10G Ampere | 0.417 |
| #23 | NVIDIA V100 Volta | 0.417 |
| #24 | NVIDIA RTX 4090 Ada Lovelace | 0.367 |
| #25 | NVIDIA RTX 3090 Ampere | 0.203 |
| #26 | NVIDIA A16 Ampere | 0.144 |
| #27 | NVIDIA P100 Pascal | 0.076 |