I experience half the FPS of one card with a crossfire R9 290 setup.

[JaredzzC]

Hi, Everyone.

When I am in crossfire mode, in the games I play, I experience roughly half the FPS from one card...

E.g. Crysis 3 one card: 80-100fps
two cards: 40-60fps
Battlefield 3 one card: 80+fps
two cards: ?40fps

I've tried drivers 14.6 and 14.7
I've tried restarting the computer many times
I've tried reseating the cards
I've also tried swapping the cards around so that each card has had some time in the primary slot, and each card works fine on its own.

i5-2500 @ 3.3GHz
12GB @ 1333MHz
XFX 850W Silver
Corsair H55 Liquid Cooling
2x R9 290 XFX DD
ASRock Z77 Extreme 4
Resolution: 1920x1080

Any suggestions?

Thanks.

Regards,
Jared.

 

[Mussels]

please fill out your *complete* system specs in your user control panel.

 

[RCoon]

Run GPU-Z and check that each card has at least an X8 PCI-E lane

 

[Mussels]

now that the specs are in, cant see any issues with the hardware. should be good enough.


GPU-Z or afterburner would be neccesary here to make sure the cards are running at the correct clocks, and the right PCI-E bandwidth.


you're running fullscreen yes, not windowed?

 

[JaredzzC]

Hi, guys...

Thanks for your replies and trying to assist me with my problem.

The cards are running at x8 each, so that shouldn't be a problem.
The clocks seem to be fine, too - no issues there, and yes, I am running things at full screen : )

 

[Mussels]

care to take some photos of the setup? show us where the monitor is connected, the cards in the slots, etc.

try and make them clear, because blurry photos wont help us at all.

 

[JaredzzC]

I hope these are clear enough and that they're what you were looking for.
I apologise for the lack of cable management. I could do more, but there isn't much space in that case.

 

[Mussels]

damn, i cant see anything wrong there. its all connected as it should be.


all we can go by from now is GPU-Z - card 1 was running at 1x1.1 ? unless its at idle, thats a problem.

 

[basco]

like mussels said first card in gpu-z running at pciex1-1.1
mussels was too quick

 

[Mussels]

try these:


1. Advanced Power Settings in control panel

2. Disable ULPS


Both of those *could* help get the PCI-E back up to the right speeds.

 

[JaredzzC]

It does seem to change between x8 2.0 and 1x1.1 randomly. Maybe because ULPS is active? I was only browsing the internet at the time, so maybe that's why.
It does work fine when it's under load, though. I ran the render test in GPU-Z and it was reporting both cards at x8 2.0

 

[basco]

did ya flash one card with 290x bios?

 

[RCoon]

the 1x1.1 is a power saving feature for idle or low clock states. Literally, I see nothing wrong in GPU-Z or in the way the PC is set up. Visually, there is absolutely nothing wrong, and software isn't reporting issues either.

It may be worth having Afterburner open, and check the GPU Usage figures for each GPU core during a render test or something. Then we can see if each GPU is at 99% usage during the render, despite only providing half the performance, or if each GPU is being bottlenecked down to 50% during the test.

EDIT: By the way, there is a VRM_1 issue with XFX 290 DD cards. After reaching a certain temperature, DD 290/290X's will severely throttle performance. They literally crap themselves because the VRM cooler doesn't even exist. May be worth noting the temps. Maybe airflow is bad.

 

[JaredzzC]

In reply to 'basco': I haven't flashed the cards... I haven't even overclocked any since I received them both. They've been kept at their factory clocks.
I ran the GPU-Z render tool and I had afterburner up, too. The GPUs seem to be fine - they were at 100% ( at times they dropped a few percent, but they would always be either 100 or near 100, but I guess that's normal)

I then went ahead and played Crysis 3 for about 15-20minutes.... There is still no difference
I even edited the cfg file a while back so that I could get more fps (which is why it may look a bit different than what it usually looks like)
I should note that this is being played in full screen (1920x1080) and the graphics are on the lowest settings.

The temperature for GPU 1 maxed out at about 77/78 degrees
The temperature for GPU 2 maxed out at about 70/71 degrees

The GPU usage whilst in-game fluctuates quite a lot, but as far as i'm aware, that's quite normal.

I still have no idea why it's being like this :/
Thanks for all those who have replied so far. Hopefully I will get more replies in the future...

 

[RCoon]

By the looks of those graphs, Crysis 3 CPU usage is bottlenecking your Crossfire. Your GPU's are using whatever they are fed, and unfortunately Crysis 3 is not so nice to CPU's even at the high end. My best guess is the GPU's are starved of data due to the IPC and clock of the CPU. Mulitple GPU's always need a stronger CPU than just one.

You don't see this in the render test, simply because it is not CPU bound in the slightest, so the crossfire is fed all the data as efficiently as possible.

EDIT:
"Originally Posted by Lowki View Post
if it is a pirated copy of crisis 3 that doesn't have patches you have to disable vsync and renable it for crossfire to kick in. That's how the game was for me vanilla when it first came out. Another reason I don't pirate games.
Or alt-tab out of the game then enter the game again, same thing. A simple reset ofr display-related stuff for CrossFire to kick in on Fullscreen with this game. And it happens with me even after it's fully updated. I have it on Origin.

I've searched a number of forums regarding crossfire on Crysis 3, and they all complain of similar issues. Single GPU 90%, multi GPU 50%. CPU is bottlenecking everything in Crysis 3 even at low resolutions, I would imagine it is a similar affair in BF3 but not quite so bad.

 

[JaredzzC]

To be honest, that does seem like a plausible explanation....
The CPU wasn't monitored at that time/in those pictures, but I played Crysis 3 again and monitored the CPU usage.

They are pretty much running at maximum all the time.
I guess it doesn't help that my CPU only delivers 2.0 x16 when the R9 290 can utilise (not necessarily fully) 3.0 x16
It's a shame, really. I was told by quite a few people (e.g. posts from Tom's Hardware) that the i5-2500 wouldn't be a problem with crossfire R9 290...

Since I don't want to buy a new motherboard, and assuming that it is the CPU that is the problem, then would an i7-3770k be a good alternative?
That seems to be the best CPU for a third gen range.

 

[zsolt_93]

In the scond GPU-z screenshot that card looks like its not a a 290 but something else like a 290x instead. Maybe that is why they are not playing nice.

 

[JaredzzC]

In the scond GPU-z screenshot that card looks like its not a a 290 but something else like a 290x instead. Maybe that is why they are not playing nice.

Yea, it does that sometimes. As far as I'm aware, the R9 290 and the x version are extremely similar and are based off the same PCB/architecture type which is why it shows those figures(the x version is basically a dumbed down version of the 290). That only happens when the GPU is in idle mode, though. When it is being used, it corrects itself to the specs of an R9 290.

 

[RCoon]

To be honest, that does seem like a plausible explanation....
The CPU wasn't monitored at that time/in those pictures, but I played Crysis 3 again and monitored the CPU usage.

They are pretty much running at maximum all the time.
I guess it doesn't help that my CPU only delivers 2.0 x16 when the R9 290 can utilise (not necessarily fully) 3.0 x16
It's a shame, really. I was told by quite a few people (e.g. posts from Tom's Hardware) that the i5-2500 wouldn't be a problem with crossfire R9 290...

Since I don't want to buy a new motherboard, and assuming that it is the CPU that is the problem, then would an i7-3770k be a good alternative?
That seems to be the best CPU for a third gen range.

I checked 3 forums, all of them detailing a 2500(K) coupled with 3 different sets of crossfire, all of which had the same issue as you. CPU is bottlenecking the crossfire quite badly, but not single GPU's. In ALL other cases, your system is fine for everything. I doubt you will see the same issues anywhere but Crysis and BF. Unfortunately both games are coded to run a million things on the CPU. Even my system doesn't handle Crysis 3 very well on high settings. 3770K would be a good upgrade, depends on whether you think it's worth it for those two games.

 

[Mussels]

so there you go, its CPU limited.


try and close down any running apps in the background that could be making it worse.

 

[FreedomEclipse]

someone give me a free copy of Crysis 3 and i can test the CPU side for you with my behemoth, if you feel a bit rich, some 290s would be awsum to kthnxbai.

 

[Dent1]

JaredzzC,

Try overclocking your RAM to about 1600Mhz+ area, and your CPU to the 4GHz area.

Don't worry if its 100% stable at first, just stable enough to run Crysis 3 and BF3 for about 10 minutes or so. Within that time I want you to conduct the same tests you did above. If your crossfire issue stops, then you know for sure it was due to CPU limitation. Then you can work on stabilising the overclock or upgrading the CPU.

If the FPS issue in CF persists, then you know its another issue, software, drivers OS etc.

 

[FreedomEclipse]

JaredzzC,

Try overclocking your RAM to about 1600Mhz+ area, and your CPU to the 4GHz area.

Don't worry if its 100% stable at first, just stable enough to run Crysis 3 and BF3 for about 10 minutes or so. Within that time I want you to conduct the same tests you did above. If your crossfire slowdown stops, then you know for sure it was due to CPU limitation. Then you can work on stabilising the overclock or upgrading the CPU.

If the CPU slowdown doesn't stop, then you know its another issue, software, drivers OS etc.

Its gonna be harder to OC his CPU because its a non-k model. probably squeeze another 200mhz out of it.

 

[Mussels]

JaredzzC,

Try overclocking your RAM to about 1600Mhz+ area, and your CPU to the 4GHz area.

Don't worry if its 100% stable at first, just stable enough to run Crysis 3 and BF3 for about 10 minutes or so. Within that time I want you to conduct the same tests you did above. If your crossfire issue stops, then you know for sure it was due to CPU limitation. Then you can work on stabilising the overclock or upgrading the CPU.

If the FPS issue in CF persists, then you know its another issue, software, drivers OS etc.

without a K chip, he wont be able to OC that high. i got a pretty awesome OC out of my i5 2400, but ~500Mhz is normal.

 

[Dent1]

without a K chip, he wont be able to OC that high. i got a pretty awesome OC out of my i5 2400, but ~500Mhz is normal.

500Mhz is enough to conduct the experiment. The OP has nothing to lose at this point.