7950 crossfire issues

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410
Hello.

I just got a second card for crossfire today. And i've been getting odd performance.
One thing that's odd is my gpu usage. I'll have one card that is at 73% and the other at 100%. I don't know if this is normal or if it's rivatuner glitching.
I also only got a 10 fps improvement in crysis 3, and it's choppy at that. Although i run the multiplayer now at 60fps with msaa @4 on very high.

I'll try to post a video in a bit.

Specs:
3570k @4.4
8gb of ram
2x 7950's
 

Darkman69

Honorable
Jun 6, 2013
506
0
11,010

Would Vsync work ?
 

Darkman69

Honorable
Jun 6, 2013
506
0
11,010


Not really but it does add negligible mouse input lag.
 


Yes really. V-sync doesn't let you write except during vertical retrace. That means it cannot have evenly distributed frames. With a 60hz monitor, that means at 45 FPS, you'll have alternating frame times of 16.7ms and 33.3ms. That causes a bit of stuttering.
http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Dissected-Full-Details-Capture-based-Graphics-Performance-Tes-11

I notice it when I use 3D Vision, because vsync is forced on. If my FPS do not maintain 60, it stutters. The same thing happens with normal 2D on 60hz monitors. It is less noticeable on 120hz monitors due to the extra refresh time interval.
 

Darkman69

Honorable
Jun 6, 2013
506
0
11,010


Cite your sources... Vsync inhibits inherent mouse input lag but thats it ....
 


I did on the post you just quoted. Click the link.

Here is a better one that is specifically talking about the stuttering:
http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Visual-Effects-Vsync-Gaming-Animation
The single card and SLI configurations without Vsync disabled look just like they did on previous pages but the graph for GTX 680 SLI with Vsync on is very different. Frame times are only switching back and forth between 16 ms and 33 ms, 60 and 30 instantaneous FPS due to the restrictions of Vsync. What might not be obvious at first is that the constant shifting back and forth between these two rates (two refresh cycles with one frame, one refresh cycle with one frame) can actually cause more stuttering and animation inconsistencies than would otherwise appear.
 


It says almost word for word what I said. What more do you want?

http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Visual-Effects-Vsync-Gaming-Animation
Frame times are only switching back and forth between 16 ms and 33 ms, 60 and 30 instantaneous FPS due to the restrictions of Vsync.

What might not be obvious at first is that the constant shifting back and forth between these two rates (two refresh cycles with one frame, one refresh cycle with one frame) can actually cause more stuttering and animation inconsistencies than would otherwise appear.

Read the whole article. It even has videos to watch so you can see for yourself. Read the comments. Look at the graphics:
BF3_2560x1440_PLOT_6.png
 

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410


I hope this is true. Although, it seems to happen to me in metro too.

Here's a pic of the gpu usage. Freggin' weird.
dwm2013060710222904.png
 

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410


No more stutter since i've been using the frame limiter. I just thought the gpu usage was very odd; frame limiter or no frame limiter.
 


The FPS limiter causes the GPU to not need full power, so it is bouncing around as it is needed or being forced to wait. It does look interesting in the way that it happens, in that it is allowed to go full speed, then stop, then full speed and stop, but that is what a limiter does.
 

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410
So i've discovered that my new 7950 that i got for crossfire is throttling with even the slightest overclock.
If it's anything above 850 core clock it throttles. This is only with my new one. It will hop between 64% and 100%, while my other one stays at average clocks. The only thing it doesn't do it in is heaven bench and BF3.
Is there any way to remedy this?

First one is a sapphire 7950 BOOST, and the other one is a XFX core edition.
Any help would be appreciated.
 


Can you test them individually, with only one in the computer at a time? Your PSU might not have enough power to keep both going under full load when OC'ed. Testing them individually will narrow things down to whether it could be that or just the GPU itself.
 

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410


I turned off crossfire and used the XFX by itself and it still did this. 63 - 100%
The fps doesn't seem to change though, which is the odd thing.
I overclocked it to 1000mhz. I felt that it was a healthy overlcock.
 
So the FPS was the same as the other card, or the FPS didn't change with the OC? If the other card at the same clock gets better scores, that one may not be performing well. You might also change how much power it can use by sliding the slider from 100% to 120% or higher.

If it performs at stock settings without problems, you may just have gotten unlucky with a bit of a dude OCer, which is not something you can do anything about.

You also want to test the card in both PCIe slots, to make sure it is not something going on with the PCIe slot. And again, make sure it is the only card in the machine.
 

Djentleman

Distinguished
Jul 25, 2011
1,045
0
19,410


Okay, so i took at the other card and just kept the problematic one.
I fired up crysis 3 and this is what i got:
dwm2013060821562661.png

This is with 20% power limiter and at 65c.


It does it in bioshock too, but it's not as prevalent:
dwm2013060822033943.png


The funny thing is, that the stock is 900 mhz, but even at that clock speed it gets wonky with the usage. Only at 850 is it normal. I asked the previous owner, and he said he didn't have a problem. (this is ebay mind you). Which he has a very good record so i wouldn't deny it.

I think i have plenty of power too. Considering i used killawatt and it wasn't even close to pulling 600 watts out of the wall. (this is with crossfire). Edit: It was averaging around 520w.

I also had the card in different slots, but it didn't work.
 
Crysis 3 is not a good choice to use, as it gets CPU bottlenecked often. What does it look like using an Unigine benchmark, like heavy or valley? If you are being CPU bottlenecked, you can expect some GPU usage drops, though you'd normally see it stay at less erratic %'s.

And if you could, run the same benchmarks with the other card, while using the same clocks. I'm curious if it is a sensor thing, or if you have different performances as a result.