[Benchmark][Dota 2] GCN1.1 FGLRX VS RADEON

[master_andreas]

(Please ignore Typos)[Dotaradeon -> Test PC]

FGLRX is 18% faster than the open source Driver

FGLRX is 10 kelvin degrees lower

So am i going to recommend the radeon driver to anyone with GCN 1.1 now. NO

Though i have to say i am impressed with the open source driver 1 year ago it was a lot slower and it might outperform the FGRLX driver maybe in a couple of year it will be better that FGRLX

Question for the readers Taking into consideration those results which driver would you use

[Tids]

and now please try a game without optimized driverpathes for the propritary drivers. like tomb raider
but yeah its just normal that fglrx is still on top because its optimized, while mesa is just at "get it working".

[master_andreas]

tomb raider

i am super poor not going to afford that so sorry i can not check it

but yeah its just normal that fglrx is still on top because its optimized

at the end of the day that is what matters to 99% of the users(if the performance Delta was't so bad i would have kept using the free drivers but i could notice it)

[BS86]

FGLRX is 18% faster than the open source Driver

a game without optimized driverpathes for the propritary drivers

well, I guess it highly depends on the card used and does not apply to amdgpu

I am only getting 30-40fps and regular drops to <15fps on fglrx on my Fury X with vsync, but on amdgpu, I get stable 60fps vsynch'ed

Tomb Raider has graphic erros on fglrx and performance <10fps, but works with 20-30fps on medium to high settings with amdgpu.

fglrx uses OpenGL 4.5, while radeon/amdgpu only use 4.3, so I guess if the benchmark would also adjust to this, I suppose that radeon would outperform the fglrx
Can you make the same benchmark with the next release of llvm and mesa (they should enable OpenGL 4.4 or 4.5 support for the opensource drivers)

[Tids]

20-30fps on medium to high settings with amdgpu.

Thats a little strange. Even on my R9 380 (a much weaker card) i'm on 35++ fps on "ultra" settings at 1080p with mesa drivers. You should have always your 60fps with a fury (well.. as long as you're not a cinnamon user or 4K Desktop user )

[BS86]

thanks for beating me to it ^^ (I am using KDE on a FullHD screen)
With Vsync disabled I get 50-55 fps on ultra settings now. (I have Vsync enabled by default because a lot of games run at 200+ fps without it which causes heavy coil whine)
I was playing with Steamenvironment disabled due to some startup errors, that I had since months. Did not think that the fix from the patchnotes would fix it for me because I had it for ages before it was added to the patchnotes.
Now steam runs with steamenvironment turned on, and Tomb Raider has quite good fps, but some slowdowns when it needs to load data from the disk (I should really try to install it on my SSD and not on my old rotational data disk)

[master_andreas]

well, I guess it highly depends on the card used and does not apply to amdgpu

the point of this those benchmarks is to show performance on amd GCN GPU what do not support amdgpu
which unclude but are limeted to:
Oland
Cape Verde
Pitcairn
Tahiti

Bonaire
Hawaii

Temash
Kabini
Liverpool
Durango
Kaveri
Godavari
Mullins
Beema
Carrizo-L

[master_andreas]

Can you make the same benchmark with the next release of llvm and mesa (they should enable OpenGL 4.4 or 4.5 support for the opensource drivers)

yes i am planing to

[Tids]

We really should call the opengl drivers "mesa/radeonsi" and "fglrx" it really sucks that its always not clear what someone means by writing "AMDGPU", since there is no such opengl driver..

"Driver Marketing name" -> "OpenGL driver"
Catalyst -> FGLRX
AMDGPU-Pro -> FGLRX
AMDGPU -> Mesa-RadeonSI
Radeon -> Mesa-RadeonSi

[BS86]

@Tids now you are mixing things up/complicating things

amdgpu and radeon are both opensource
amdgpu-pro is amdgpu + fglrx
fglrx is catalyst is nonfree

both radeon and amdgpu are using Mesa and RadeonSi, thats where you are correct.

@master_andreas I was referring to you saying that opensource has less performance than fglrx - it is only true for radeon and the card you benchmarked

[master_andreas]

I was referring to you saying that opensource has less performance than fglrx - it is only true for radeon and the card you benchmarked

Correct me if am wrong but shouldn't all GCN 1.1 GPU perform similarly (talking about the lerative performance of FGLRX and MESA)

[Tids]

now you are mixing things up

the OpenGL module of "AMDGPU" and "Radeon" is Mesa/RadeonSI
the OpenGL module of "Catalyst" and "AMDGPU-Pro" is fglrx

since we only talk about opengl performance we just talk about fglrx or radeonsi ^^

[BS86]

since we only talk about opengl performance we just talk about fglrx or radeonsi ^^

that leaves us with the problem of how to tell the difference between amdgpu and radeon - which is existing performance-wise and technology-wise... better to talk about the kernel-modules

Correct me if am wrong but shouldn't all GCN 1.1 GPU perform similarly (talking about the lerative performance of FGLRX and MESA)

I would not think so, but I also have no perfomance benchmark at hand to prove anything.

[Tids]

which is existing performance-wise and technology-wise

by a minimal up and down. i would say you just can ignore it
https://openbenchmarking.org/result/1608170-HA-1608175HA54
https://openbenchmarking.org/result/1608178-HA-1608175HA04

shouldn't all GCN 1.1 GPU perform similarly (talking about the lerative performance of FGLRX and MESA)

Nope. Mesa tries to support the "standard" OpenGL while fglrx has some OpenGL profiles for some games to reach better performance. Thats one of the "professional" features you'll only get with the propritary drivers.