TOPS (Tera Operations per Second) has become convenient shorthand to describe the deep learning capabilities of a component. However, there are a myriad of variables that affect how hardware handles artificial intelligence. In a bid to level the playing field, from CPUs to smartphones, Geekbench has launched its new AI benchmark.
Plainly called Geekbench AI, this test allows you to put your computer through multiple benchmarks to check its handling of AI tasks. At the end of the tests, your system receives a score that you can free compare to other devices. You can download the app for yourself now, on Windows, Linux, macOS, iOS, and Android.
Geekbench AI includes multiple computer vision workloads, namely, Image Classification, Image Segmentation, Object Detection, Face Detection, Pose Estimation, Depth Estimation, Image Super Resolution, and Style Transfer. You can learn more about each of these in Geekbench’s documentation.
Currently, the Windows version features two AI frameworks for its tests: OpenVINO and ONNX. While both the best CPUs and graphics card can tackle the latter, only processors can compete in the former ring. Both are open-source with authors including Facebook, Microsoft, and Intel.
Since I can’t miss an opportunity to benchmark my PC, I naturally had to download Geekbench AI and give it a spin. Without further ado, here are the results of my Ryzen 5 7600X (running in ECO mode) and Radeon RX 7900 XT:
| Ryzen 5 7600X | Radeon RX 7900 XT | |
|---|---|---|
| Single Precision Score (ONNX) | 2,888 | 34,121 |
| Half Precision Score (ONNX) | 832 | 38,752 |
| Quantized Score (ONXX) | 4,565 | 26,476 |
| Single Precision Score (OpenVINO) | 5,674 | – |
| Half Precision Score (OpenVINO) | 5,735 | – |
| Quantized Score (OpenVINO) | 15,265 | – |
- As expected, my scores with one of the best graphics cards on the market dwarf those of my entry-level processor. Note that GeekBench calibrates its AI tests against Intel’s Core i7-10700, with a baseline score of 1,500.
Want to test your PC? Download the app from Geekbench (597MB) and give it a run, it only takes a couple of minutes to finish. Meanwhile, if you like to compare your AI performance to other devices, check out the ML Benchmark Chart from developer Primate Labs.