Caffe AlexNet Deep Learning Benchmark Added
In addition to adding some new OpenCL / CUDA tests this week to the Phoronix Test Suite and OpenBenchmarking.org, Caffe was added too as a deep learning benchmark.
For those unfamiliar with the Caffe project, "Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by the Berkeley Vision and Learning Center (BVLC) and by community contributors." Caffe offers an expressive architecture, extensible code, can process over 60 million images a day on a lone NVIDIA K40, and there's quite a community built up around the project.
This new test profile is exposed as caffe and currently is a timed implementation of the AlexNet model with 100 iterations. The Phoronix Test Suite automatically builds a CPU-only copy of Caffe as well as a CUDA+cuDNN version as well, if you're using a NVIDIA GPU and have the necessary development packages installed.
So with this addition, you can see some deep learning benchmarks on Phoronix. Of course, kicking off those benchmarks will be tomorrow's results for the NVIDIA Jetson TX1.
Another new test profile this week is also gfxbench, the graphics benchmark primarily popular on Android and iOS devices. With this test profile, if you have a copy of GFXBench31_Linux, you can zip it up and toss it in your Phoronix Test Suite download cache and then phoronix-test-suite benchmark gfxbench to fully automate your GFXBench testing process on Linux. Unfortunately I'm unaware of any public build of GFXBench v3.1 for (non-Android) Linux outside of companies.