6870 Poor Crossfire Performance

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
Picked up a pair of 6870's today to check em out and do some benchmarks.. I'm getting really poor performance and I'm not sure whats up. For example, getting between 35/40 FPS in Crysis Warhead on gamer settings @ 1920x1080 with no AA.. Other games are performing equally as crappy.. Before these cards, I had a GTX470 SOC in this rig. Uninstalled the drivers and Nvidia software, put these cards in and installed the latest ATI drivers from the website.

In GPU-Z It looks like the clock on the second card doesnt go up very high and the memory clock on either doesnt get about 280 or so. Checked BIOS, no special settings to change. Am I missing anything?

Edit: I notice in 3DMark Vantage, after installing the 6870's, my CPU performance dropped from 50,000 to 11,000?
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
Other specs are as follows..

Processor: Phenom II X4 965, 8GB Patriot Sektor RAM, 850 Watt Corsair PSU, Windows 7 64 bit Ultimate

The second card is working, I took one card out last night and ran a few benchmarks. 3DMark Vantage with one card is around 15000 GPU score, with the second it goes up to about 23000 GPU points.. My single 470 SOC gets 22600 with 5x the CPU score. I get nearly identical frame rates with 1 card as opposed to two cards in Crysis and the average with the single 470 is about 6-10 fps higher. In the lost planet 2 benchmark, I've gained 5 FPS average over the 470. Doesn't make any sense, either my 470 is performing INCREDIBLE or the 6870's aren't doing what they should. Today I'm going to do a clean OS install.

As a side note, since I installed the 6870's 3 games have crashed. I've never had a game crash with the 470. Frustrating to say the least.

As an update to the first post, I've gotten the clock the second card usage to match the first card usage in crossfire compatible games.. Still getting crappy performance though.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
In the process right now.. Did a clean format..

Another funny thing I noticed. In crysys, I did three benchmarks.. The 470 on Gamer, the 6870's on Gamer, and the 6870's on enthusiast.. The 470 averaged 38 FPS on gamer. The 6870's on gamer average 36 FPS in crossfire, and 34 FPS on enthusiast in crossfire. The res' were at 1920x1080. Def something wrong, can't say whats causing it though.
 

96Firebird

Distinguished
Jan 5, 2008
125
0
18,690
Just wanted to point out, don't lean on the 3DMark Vantage score difference, unless you disabled Physx. The GTX 470 will cause an inflated CPU score because of the Physx aspect. To correctly compare the scores, disable PPU.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630


Before the format I tried in Lost planet.. I gained 6 FPS from the 470 with all the same settings.. I gained quite a bit in Resident Evil 5. I saw some gameplay benchmarks on youtube with crysis at 1920x1280 with 4xx AA on and it looked like they were getting 40 something FPS.. Figured 1920x1080 with no AA would push that into the 50's pretty steadily.. Maybe I was completely overestimating the 6870's..



I noticed that, I wasn't sure if the 470 would have that effect on the CPU only test.. Thanks for confirming it would.. I assume thats what bloated the first score.

So if this reformat doesn't change anything, we can assume that TWO 6870's = 1 GTX 470 SOC.
 

notty22

Distinguished
No the benchmarks show 2 -6870's ahead of the gtx 580 in some games
http://www.legitreviews.com/article/1461/7/
Did you install the crossfire profiles after the driver install.
They are separate and need to be regularly updated
http://sites.amd.com/us/game/downloads/Pages/crossfirex-app-profiles.aspx
avp.jpg

jc2.jpg



 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
Those are the numbers I was expecting... The format just finished.. I just installed my mobo drivers.. Getting ready to install the video card drivers and crossfire profiles (yes, I isntalled them the first time as well) and the test out a few games.. Wish me luck.. I'll report back the findings.
 
Please do report back.

Once I was trying to track down why an ATI vid card outperfromed a more powerful ATI card. Took a while to find that the settings in CCC were not the same, and I was relying on the in-game settings when in reality CCC had control.

Just a thought.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
not a tremendous difference.

Update: Still no real luck with this.. Got one more of the games to work that wouldn't work before. Metro 2033 won't open. Frame rates are still so so for the setup. I'm frustrated for the night, so I'll keep trying tomorrow.. I opened a ticket with HIS online support so we will see what they say.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
It may be too early to be certain that it's fixed, but I think I may have gotten to the bottom of it.. Or at least what I did to fix it.

Pissed off after a day of saying WTF, I figured I would play some Champions Online.. Not incredibly graphical and lots of settings that can be changed to adjust for performance.. I opened it up and was disgusted instantaneously.. With the 470, I was getting 38-45 FPS.. With the two 6870's Xfired, I was getting 18-23?!

That was the last straw, so I disabled Xfire and re-opened the game.. I got about 55 FPS.. Pissed off, I closed everything and removed the profiles that I installed earlier.. Forgetting to enable the second card, I installed the profiles for the 4th time.. Then I enabled the second card and opened up Champions again.. Bam, 65 FPS.. Opened Crysis, got 34as a MINIMUM with an average of about 45-62 on "gamer" settings.

Have N O idea what the hell happened, all I can think of was I installed the profiles with the second card disabled and THEN enabled Xfire. I'm beat now, so going to crash, but when I get up I'm going to run some benchmarks and whatnot to tell if it's fixed for sure or if it's just a fluke.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
Wow that's some good info. Thank you for it.. That maybe whats happenin..As I play with it more, it seems like it's hit and miss.. For instance, after having it fixed, I started to play Champions again.. Then out of nowhere I got framerates in teens. I disabled crossfire, played and it was back to high 50's. Re-enabled crossfire and back to the 70-80's.. Who knows.. I'm headed to bed.. Back at in the AM.

PS- Just heard my 580 is comin within the next few days ;).. Looking forward to checkin that out too.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
Talk about frustrating.. An update for any of those interested..When I thought I had this issue licked last night, turns out it was just a fluke. VERY inconsistent. I'm going to make a few videos and upload them to illustrate exactly what is happening. The performance seems to based on the cards mood at that particular moment.

Example: I ran a benchmark for Just Cause 2.. First time around on the Desert scene, I average 73 FPS..For the second benchmark, NOTHING CHANGED.. I just closed out of the game, re-opened that game and did the same exact test with the same settings and I didn't get above 35 FPS. Been happening like that in all the games.
 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630


BIOS is the latest.. I was hoping it would be that simple.. That's the first thing I did.. All other chipset drivers are up to date.. Thanks for the good suggestion though, I appreciate the link as well..

Still working on it though, not ready to give up on Crossfire just yet.. Tell ya this though, this almost never happens with a single card setup..
 

notty22

Distinguished
Lol everyone forgot about OP's cpu.Online benchmarks were done with overclocked i7 which will destroy OP's cpu in game.Your cpu is just too weak to keep up with your powerful gpus.Clear case of severe bottlenecking.

No its not.
There is something going on here.
OP, can you go in to your bios and turn of cool and quiet.
There are situations where load is not forced. I see it happen much more often with the 6 cores. Its as if the turbo gets tricked, and the system responds as if its not in a rush.
Best way to describe it.
I've seen benchmarks of the new gtx 580 on AMD systems, return expected numbers.


 

dimurof82

Distinguished
Oct 19, 2010
80
0
18,630
When I get new mobo's and the like, during the initial installation I disable any "eco friendly" or hardware control systems.. Cool N Quiet is disabled, as are all conservative power controls in Windows and software that came with the mobo. The windows power control profile is on performance. I thought for a few seconds it could be CPU bottlenecking, but then I thought back and remembered seeing 5970 benchmarks, 480 benchmarks, 5870x2, all on Phenom 2 quad cores.

However, I'm willing to explore it as an option.. Can you think of a way to test CPU bottlenecking without having to buy a new mobo/processor? I would hate to shell out 5-600 on a new processor/mobo for nothin..
 
If a 965 can't drive 2x6870, then AMD has no game at all lol. This isn't a cpu bottleneck.

But if you want to test, I think if you turn down the level of detail and get an increase in FPS you have shown the cpu is not the limiting factor.