OpenForBC-Benchmark is a suite of ready-to-run benchmarks that execute in an automated manner along with reporting of test results, detection of installed system software/hardware, and other features.
This standalone tool was developed to benchmark performances of various partitioning options on industrial-grade GPUs, as a part of OpenForBC efforts to develop a framework capable of presenting a common interface on top of GPUs to overcome the barriers introduced by payware environments such as VMware ESXi and CITRIX and providing extended support for Linux KVM, more of which can be found here.
The framework is compatible with Windows, Linux, and macOS given the benchmark supports the tester operating system.
- Python: >= 3.9
- pip: >= 10.0 (can be updated by running
pip3 install -U pip)
ML benchmarks depends on tensorflow, which requires NVIDIA CUDNN and some cuda libraries to be installed, specifically:
- cuda-cudart
- libcublas
- libcufft
- libcurand
- libcusolver
- libcusparse
Please refer to TensorFlow's guide for GPU support if you are using ubuntu or to NVIDIA's documentation for CUDNN installation and CUDA toolkit installation.
| User guide | Developer guide |
|---|---|
 |
 |
Currently, the following benchmarks are implemented: