Should we be expecting any sort of performance benefit with OpenCL applications when using the new 14.12 driver? (Particularly in the Linux environment) There is a bunch of marketing out there about improved performance, but nothing mentioned OpenCL except for the OpenCL 2.0 support...
You may be asking.... is there a performance improvement just from the new driver, without doing anything else? I don't know, but I can start asking around, see if I can get some light on that.
Is there potential improvement, if you take advantage of OpenCL 2 features? For sure.
One blog we have up has comparative benchmarks: http://developer.amd.com/community/blog/2014/10/24/opencl-2-shared-virtual-memory/
a binary tree search of 5M nodes took 18 milliseconds using SVM vs. >310 milliseconds without.
There might be some benchmark data in the other OpenCL 2.0 blogs, but if so it has slipped my mind.
Yeah, please ask around. I am using OpenCL 1.2 as development started several months ago and we don't have the budget to rewrite everything at the moment!
all OpenCL 1.2 application that I've tested with Catalyst 14.12 run slower than with Catalyst 14.9 (20%-50%). The performance of Catalyst 14.12 is comparable to the OpenCL 2.0 beta driver. Hopefully the performance regressions will be fixed in the future by AMD.