Clear Linux Makes Caffe Deep Learning 10% Faster; Also Discovers XFWM4 Compositor Bug

Intel's Clear Linux distribution hasn't been at rest this summer but they've continued working hard on various performance optimizations and improvements to their distribution.

The Intel Open-Source Technology Center developers working on Clear Linux have begun a weekly report to highlight the changes made over the past week to their updated-daily operating system.

In addition to covering updated packages like their just-landed Mesa 12.0, they also cover performance optimizations.

Eva Hutanu noted in this week's initial report, "This week, we increased the performance of the Caffe machine learning framework by 10% by improving our linear algebra library and by increasing the use of SSE and AVX vector instructions. Caffe is one of the top 5 ML frameworks in the industry, and its performance is one of our PNP focus areas." Of course, they measured their Caffe benchmarking performance with the Phoronix Test Suite.

They also dug into my recent 7-Way Linux distribution comparison results and discovered why they weren't winning as much... They use the Xfce desktop enviornment and in digging deep into my test data they ended up discovering a bug in the XFWM4 compositor. Eva explained, "we root caused this issue to a bug in the XFWM4 compositor. Currently, a short term workaround is deployed, while the search for a sustainable long term solution continues."

More information via the Clear Linux dev list.