AMD Catalyst 14.6 Does Slightly Better With APITest OpenGL Tests

Written by Michael Larabel in Display Drivers on 30 July 2014 at 12:00 PM EDT. Page 1 of 3. 16 Comments.

Last month I published some benchmarks where NVIDIA slaughtered AMD's Catalyst driver in APITest benchmarks, a set of OpenGL 4.x test cases. While these are just micro-benchmarks of modern OpenGL and designed for showing off potential solutions for problems to leading to lower driver overhead, there's been some improvements within the Catalyst 14.6 Beta on Linux. Up today are some new benchmarks of APITest using the latest AMD and NVIDIA drivers.

AMD's Catalyst 14.6 for Linux improves some GL4 support that affects APITest, which is now leading to more of the APITest problems properly running outside of NVIDIA's binary driver. However, AMD still has more work to do as these results will show. Graham Sellers of AMD explained in an email to Phoronix, "We're working on the APITest suite and have made a number of improvements in the last little while. I don't think much of it has hit the public driver yet, and we still have some ways to go, but we're getting there. Expect to see performance improve over the next little while."

While NVIDIA generally is much faster than AMD's Catalyst at APITest, Graham also wishes to reiterate that the focus of APITest's development is not about comparing GPU driver vendors but rather finding fast OpenGL paths for developers. "Unfortunately, due to the pathological nature of the tests, we're spending a lot of effort optimizing for the things that the test suite is designed to encourage people not to do. Again, I can't stress enough that this suite was never meant to compare vendor to vendor, but technique to technique. For example, if you take the "UntexturedObjects" problem and create a graph of the solutions to it - "GLBufferStorage-SDP", "GLMapUnsynchronized", "GLBufferStorage-NoSDP", "GLMultiDrawBuffer-SDP", and "GLMultiDrawBuffer-NoSDP" - you'll see that the performance scaling from the "not recommended" to the "strongly recommended" solutions is similar across vendors. The message is not, "if you do this, you'll go faster on AMD than NVIDIA", it's "if you do this, your application will go faster, regardless of vendor"."

APITest NVIDIA + AMD Linux Perf

Anyhow, for these updated APITest numbers are results from the latest-generation AMD Radeon and NVIDIA GeForce hardware that I have available, which included on the GeForce 700 series side:

- GeForce GT 740
- GeForce GTX 750
- GeForce GTX 750 Ti
- GeForce GTX 760
- GeForce GTX 770
- GeForce GTX TITAN
- GeForce GTX 780 Ti

On the AMD side the Rx 200 graphics cards I own are just the Radeon R7 260X, R9 270X, and R9 290. The latest AMD Catalyst 14.6 Beta driver was in use (fglrx 14.20.7 / OpenGL 4.3.12967) and on the NVIDIA side was the latest 340.24 Linux driver. All of these APITest benchmarks were carried out via the Phoronix Test Suite.


Related Articles