GPU Comparison Chart
from fastest to slowest based on FP32 performance.
| FP32 | Deep Learning | Memory | Memory Bandwidth | |
|---|---|---|---|---|
| DEFINITION | Single precision perf. (TFLOPS) | Tensor perf. / TF32 (TFLOPS) | GPU memory (GB) | GPU memory bandwidth (GB/s) |
| GPU List | ||||
| Nvidia RTX 4090 | 83 | n/s | 24 | 1008 GB/s |
| Nvidia H100 (PCIe) | 51 | 756 | 80 | 2000 GB/s |
| Nvidia RTX A6000 | 38.7 | 310 | 48 | 768 GB/s |
| Nvidia RTX 3090 | 35.5 | n/s | 24 | 936 GB/s |
| Nvidia RTX A5000 | 27.8 | 222 | 24 | 768 GB/s |
| Nvidia RTX A4000 | 19.2 | 153 | 16 | 448 GB/s |
| Quadro RTX 8000 | 16.3 | 130 | 48 | 672 GB/s |
| Titan RTX | 16.3 | 130 | 24 | 672 GB/s |
| Tesla V100 (PCIe) 1 | 14 | 112 | 16/32 | 900 GB/s |
| GeForce RTX 2080 Ti | 13.5 | n/s | 11 | 616 GB/s |
| GeForce GTX 1080 Ti | 11.3 |  |
11 | 484 GB/s |
| Tesla P100 (PCIe) | 9.5 | |
12/16 | 732 GB/s |
| GeForce GTX 1080 | 8.9 | |
8 | 320 GB/s |
| Tesla M60* 2 | 4.8 | |
8 | 160 GB/s |
| Tesla K80* 3 | 4.1 | |
12 | 240 GB/s |
1 AWS P3 Instance
2 AWS G3 instance
3 AWS P2 Instance
* (single GPU)
Octanebench Benchmark
OctaneBench is a popular program that uses Octane Render to benchmark GPU rendering performance.
| OctaneBench Score | |
|---|---|
| Graphics Card | OctaneBench® 4.00 Score |
| GeForce RTX 3090 | 656 (2020.1) |
| Quadro RTX A6000 | 631 (2020.1) |
| Quadro RTX A5000 | 592 (2020.1) |
| Tesla V100 (PCIe) | 354 |
| Quadro RTX 8000 | 319 |
| Quadro RTX 6000 | 304 |
| GeForce RTX 2080 Ti | 302 |
| Quadro RTX 5000 | 245 |
| GeForce GTX 1080 Ti | 216 |
| Quadro P6000 | 192 |
| Tesla P100 (PCIe) | 237 |
| GeForce GTX 1080 | 149 |
| Quadro P5000 | 139 |
Results are taken from official OctaneBench website.
Need a GPU machine for your research/project?
Whether it’s for deep learning, or rendering… You can affordably rent a cloud GPU server from us. Contact us if you can’t find the configuration you’re looking for, or if you’re unsure about which GPU/server to get.