Catalyst 15.5 For Linux Brings Some Performance Improvements

[pete910]

Good AMD, good AMD. Metro X performance has been fixed . But, where is support for Fiji, Grenada and other members of new (???) Rx 300 series?
Can Cat15.5 handle new Fury X? I didn't find Fiji in OpenCL driver. where is it?
Ahhh. Yes, yes in OpenGL driver (but not in OpenCL ).

He's failing to mention the huge improvement in dying light too, Borderland though played well anyway has also double in perf for me

[dungeon]

Yea, I have a 290x. Not quite sure why I have seen a bigger jump in perf to what Michael has though. Whish he would use a common rez though. I have run the valley/heaven benchmark @ 1080p against a 970 but the results refuse to upload/show up for some reason. This is what i used http://openbenchmarking.org/result/1...BE-1506107DE90

screen shot of my results here

Cool, so you are beating nvidia 970 in Unigine Valley

Who knows for Michel maybe fglrx does not play well with Unity or even on Ubuntu for some reason...

[djdoo]

Guys I did not read anyone to mention but the bus width value in Catalyst Control Center for driver 15.5 is not marked as "PCI" anymore but now it reports "PCI Express 3.0" for my Tonga Radeon R9 285 card!
I had seen this value from pictures Michael posted at various phoronix articles testing the Radeon R9 290 for instance, whereas Bus width also said "PCI" where it should report "PCI Express 3.0".
All previous drivers were reporting "PCI" value for me... I don't know if this was a bug of the panel reporting trash values or it was indeed the driver not using PCI Express 3.0 or 2.0 lanes of the motherboard at the bus, maybe that could explain METRO improvements?? Who knows...

[nadro]

It's something wrong with this benchmark or conclusion:

However, there's still a ways to go with the performance when running these titles at 2560 x 1600 as the NVIDIA GeForce graphics cards are still much faster on the same system with the proprietary NVIDIA driver.

because when we check this benchmark:
http://www.phoronix.com/scan.php?pag...x-22gpus&num=1
Radeon R9 290 had 25 FPS on Catalyst 14.12, but in the latest benchmark this card with Catalyst 14.12 has only 10 FPS, so it looks like a different settings was used for those two benchmark or something strange was appear in the latest test. In this situation we should decrease GeForce results too or increase Radeon results (IMO Michael used GF results from the first benchmark in upper conclusion) and after that Radeon score looks really good, why?
25 FPS = 40 ms per frame (test 1: http://www.phoronix.com/scan.php?pag...x-22gpus&num=1)
10 FPS = 100 ms per frame (test 2: http://www.phoronix.com/scan.php?pag...55-linux&num=1)
as we can see in the first test a performance was 2,5 better, so we should divide by 2,5 frame time from test 2 when we want to compare those results with test 1. In this situation we'll have a following result for R9 290:
30 FPS = 33,33 ms per frame
33,33 ms / 2,5 = 13,3 ms per frame
13,3 ms = 75 FPS
When we'll compare 80 FPS (GF 970) and 75 FPS (R9 290 with Catalyst 15.5) we can say that R9 290 is just little slower than GF 970 and this is really good result IMHO. I used frame time in ms instead of FPS for this calculation, because frame time is linear, FPS not.

[Kano]

What you do is stupid magic. You can not calc the benchmark results!