RDNA 3, Driver 23.1.2 fixes AMD Mesh Shader performance issue, 7900 XT/XTX (Update: Maybe Not)

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
7,262
There was a concern about a possible hardware issue with Mesh Shaders with the 7900 XT/XTX due to rather poor results in 3DMark Mesh Shader feature test. With driver 23.1.1 with Mesh Shader on, it was only 11.9% faster then with it off. With 23.1.2, look at the difference between the two:

MeshShader.jpg
AMD did not indicate this was updated or fixed in most recent notes dealing with 23.1.2
https://www.amd.com/en/support/kb/release-notes/rn-rad-win-23-1-2-kb

Also note the GPU speed on the XFX Merc 310 7900 XTX, this is default clock settings.
 
Interesting, I wonder if there are notable performance increases to actual games that efficiently use mesh shaders?
 
Interesting, I wonder if there are notable performance increases to actual games that efficiently use mesh shaders?
Good question, 3DMark new Speed Way benchmark uses Mesh shaders and the following:
The test measures graphics card performance with a combination of real-time ray tracing and traditional rendering techniques. The scene features ray traced reflections, real time global illumination, mesh shaders, volumetric lighting, particles and post-processing effects. The rendering resolution is 2560 × 1440.​
Using Speed Way and comparing the two drivers, one that Mesh Shaders are working much better, 23.1.2 on the left side and the one not so good 23.1.1 -> really made no difference:

SpeedWay.jpg
Now comparing the 7900 XTX Speed Way results to a 4080 paints a rather bad picture from the benchmark perspective. How it relates to an actual game??? The bottom box, yellow orange outline is my result, top graph shows where that performance lies with all the 4080 data results using a 5800 X3D CPU system (yellow orange line is my score, the blue line shows the 4080 results). I hope that does not indicate future performance in games, I doubt it but do wonder.



SpeedWay4080.jpg
 
I think the only game using them yet is maybe Fornite? As I think UE5's Nanite system can make use mesh shaders if the hardware is present.
 
Comparing the Driver 23.1.2 to the newest available driver 23.2.2 gives some very weird results on this benchmark. Look at the difference with Mesh Shader Off compared to On with the two drivers and the overall difference in FPS with it on. What is even more strange during the Mesh shader test while running with Mesh Shaders On, it will go to 74fps when a larger number of objects are being shown, the same as when it is off and then jump to over 1000fps when the number of objects decrease, as if switching between on and off depending upon the number of meshes. Not sure what to make of this, how applicable this test to real world games but it definitely looks wonky.

MeshShader.jpg
 
Speedway which uses Mesh Shaders, comparing 23.2.2 to 23.1.2 drivers lost about 2.3% performance in the test. Insignificant but looks to be measurable.

SpeedWay.jpg
 
Back
Top