Conclusion Up Front: It appears the Catalyst 14.4 Release Candidate driver for Linux provides huge performance gains over both previous iterations of the proprietary AMD Catalyst (fglrx) driver and the new proprietary AMD Catalyst 14.4 (not the Release Candidate) driver for Linux when running Fedora 20 with an AMD A10-6800k Richland 4.1 GHz APU (see benchmarks below). If anyone can verify the contents of this post by replicating the results, I would be very interested in knowing about it. See this post for a complete set of patching and installation instructions.
Benchmarks:
So, if you’ve read the last post regarding benchmarks and the Catalyst 14.4 Release Candidate driver, you know that the benchmark pattern has gone as follows:
From 3/30/2014 with Catalyst 14.3:
Unigine Heaven v4.0 @ 1920×1080: 6.02 fps
Unigine Heaven v4.0 @ 1600×900: 8.14 fps
Unigine Tropics v1.3 @ 1920×1080: 24.01 fps
Unigine Tropics v1.3 @ 1600×900: 28.73 fps
And then on 4/25/2014 with the Catalyst 14.4 Release Candidate, a stunning improvement:
Unigine Heaven v4.0 @ 1920×1080: 6.14 fps (+0.12 or +1.99%)
Unigine Heaven v4.0 @ 1600×900: 8.74 fps (+0.60 or +7.4%)
Unigine Tropics v1.3 @ 1920×1080: 30.86 fps (+6.85 or +28.5%)
Unigine Tropics v1.3 @ 1600×900: 38.44 (+9.71 or +33.79%)
Sure, the test is limited, but each test is performed three times and the average performance is reported, and it sure seemed hard to believe that it was some sort of fluke. I played some TF2 and certainly seemed to notice the significant performance improvement.
Today, after updating to Catalyst 14.4 (not the release candidate):
Unigine Heaven v4.0 @ 1920×1080: 6.01 fps (-0.13 or -2.12%)
Unigine Heaven v4.0 @ 1600×900: 8.16 fps (-0.68 or -7.78%)
Unigine Tropics v1.3 @ 1920×1080: 24.78 fps (-6.08 or -19.70%)
Unigine Tropics v1.3 @ 1600×900: 29.58 (-8.86 or -23.05%)
The regression is not complete, but it is nearly so. What a bummer! There’s not enough information to know exactly what’s going on here, so I will definitely attempt to install the Catalyst 14.4 Release Candidate (employing the patch described in the previous post) on my current build to see if I can regain those performance enhancements. If that doesn’t work, I’ll head back to the 3.13.10-fc200.x86_64 kernel to see if I can repeat the anomalous results.
Update: Catalyst 14.4 Release Candidate on Fedora 20 with Kernel 3.14.3-200.fc20.x86_64
Unigine Heaven v4.0 @ 1920×1080: 6.30 fps (+0.29 or +4.82%)
Unigine Heaven v4.0 @ 1600×900: 8.74 fps (+0.68 or +7.11%)
Unigine Tropics v1.3 @ 1920×1080: 30.89 fps (+6.07 or +24.66%)
Unigine Tropics v1.3 @ 1600×900: 38.44 (+8.86 or +29.95%)
Well, would you look at that? Not only have I gathered some positive evidence for the vast improvement offered by the Catalyst 14.4 Release Candidate for my APU (an improvement which seems to have gone unnoticed with discrete GPUs and low-end APUs) but the potential causes of this improvement have been narrowed. It seems that something peculiar about the Release Candidate is causing the massive gains, as these gains remain constant across kernel versions and there is a serious performance regression when upgrading from Catalyst 14.4 Release Candidate to Catalyst 14.4.
Can anyone else out there replicate these results? As it stands, I’ll advise anyone running a high performance APU from AMD (Kaveri, Richland, etc.) to use the Catalyst 14.4 Release Candidate rather than the Catalyst 14.4 driver when running Fedora 20. Just follow the steps detailed in this post to patch and install the driver on Kernel 3.14.*
Engineering Walden 