AMD Driver Overhead and Async Shaders in Mgpu...

from AshesoftheSingularity Forums

Ok. First of all i know that i only have 2 Radeon 7950 for now. (I'm waiting for the red team's latest effort this summer...) And, as they are their first try at Async Shaders, they aren't near as effective as the latest AMD cards (290X, 390X, Fury, etc). But still, it seems strange to me that Async Shadres seems more effective when Mgpu mode is disabled... I mean, it should be twice as effective with 2 cards..? Driver Overhead, driver maturity or game maturity problem maybe..? Crossfire is disabled in AotS Crimson profile...

 

Look at this..

1 Card, Async On...

 

1 Card, Async Off...

 

Now, Compare FPS differences and Driver Overhead with a 7950 Mgpu...

 

2 Cards, Async On...

 

2 Cards, Async Off...

 

Only .2FPS difference with a 7950 Mgpu setup vs 1.5Fps difference with only 1 Card used..? Again, who may be to blame? Driver overhead, driver and\or game maturity, my 3.7Ghz Xeon X5650 not powerful enough (1 core at about 70%, others at +- 50%), PCI-E 16x 2.0 not fast enough..? Driver overhead is really high in Mgpu mode and, like i said, Crossfire is disabled in AotS driver profile... When Enabled, it looks like this... Ehh.. Not good!

 

 

Also strange is the fact that the 2nd card only adds about 10\11Fps when my CPU and GPU 2 arent even maxed out.. GPU 1 is seeing 99% load, GPU 2 between 70-90% load...

 

Just to compare, can anybody post a similar comparaison (Driver Overhead) on a Nvidia setup..?

 

Thanks!

27,034 views 5 replies
Reply #1 from Sins Forums Top

A more parallel input doesn't help a card that can't keep up get faster, it helps a card that can't be fed fast enough get utilized more fully.  If you match it up with a 680 GTX, it's a high bandwidth card with excess computing power, that lacks pretty majorly in texture fill rates by comparison.  It's bottleneck is simply elsewhere, so asynchronous computing just isn't a big help.

Reply #2 from AshesoftheSingularity Forums Top

Do you mean i would get more FPS with 1 Geforce and 1 of my Radeons..? If only i could test it without spending $$...

 

In the meantime, i did more tests and at the same exact time where there is la lot of extra Driver Overhead... Boom GPU load drops! (30-80% typical...)

 

...weird if you ask me...

The Political Machine 2024
Reply #3 from Sins Forums Top

No, I mean that it's limiting factor isn't shader computation, it's weaker in other areas.  It's like putting in more ram, or a bigger processor, it wont improve your performance unless it's a limiting factor.

Reply #5 from AshesoftheSingularity Forums Top

Note you can easily view and re-post your benchmark results by visiting the Ashes Metaverse webpage here.