Crossfired 5850's Issue

[erocker]

Here it is:

Current system specs:
x4 965BE @ 4ghz / 2740mhz NB / 2100 HTT
8gb DDR3 1400mhz 6 6-6-18
HD 5850(s) @ 825/1000 Using Catalyst 10.8 w/ 10.8a App Profile
Corsair 850TX
Crosshair IV


Single 5850:


CrossFire 5850's:

[cadaveca]

Good for you, not so for us.

Different res..might affect things, but not so likely.

Only other thing...are you using just the FC2 bench, or the full app? They are NOT the same, unfortuantely, due to updates since release. This might be a big one here.

Also, because you are overclocked, it might be possible that the driver is far more cpu-dependant than before. Still a few things left to look at.


I'm just downloading FarCry2 again, to the new OS. Shouldn't be too long, and if I'm lucky, i'll be getting the same results as you, e.

[erocker]

Originally Posted by cadaveca

Good for you, not so for us.

Different res..might affect things, but not so likely.

Only other thing...are you using just the FC2 bench, or the full app? They are NOT the same, unfortuantely, due to updates since release. This might be a big one here.

Also, because you are overlcocked, it might be possible that the driver is far more cpu-dependant than before. Still a few things left to look at.


I'm just downloading FarCry2 again, to the new OS. Shouldn't be too long, and if I'm lucky, i'll be getting the same results as you, e.

I'm using just the FC2 bench. One thing that I know has helped me with CrossFire is increasing the frequency of the north bridge (memory controller) but even so, you shouldn't be getting identical results to a single card. I'm a bit stumped with your problem.

Good luck to you both!

[cadaveca]

Originally Posted by erocker

I'm using just the FC2 bench. One thing that I know has helped me with CrossFire is increasing the frequency of the north bridge (memory controller) but even so, you shouldn't be getting identical results to a single card. I'm a bit stumped with your problem.

Well, that's still good info...I will try the bench and see how that fares as well. Maybe just an issue with the profiles, or something.


Also makes me think that AMD isn't testing with full-version software, for alot of these apps. And if that's the case, they are such amateurs...and need to be fired, IMHO. Using benches to compare performance for full-version software is just so inaccurate it isn't funny, and they should know better. And if the bench works, but the full app doesn't, then 100% this is the case.

[ArmoredCavalry]

Thanks for posting your results erocker, at least that tells me something must be wrong with my individual setup. No clue what.... but something...

I was just playing BC2 for a round, and it crashed twice during the game.... Don't know if that explains anything...

[erocker]

I'm starting to think you guys should take a good hard look at your PSU's being the problem... maybe? 4 x 18 amp rails vs. my single 70 amp rail.. I wouldn't think it would be an issue but it's something you both share...

[ArmoredCavalry]

Originally Posted by erocker

I'm starting to think you guys should take a good hard look at your PSU's being the problem... maybe? 4 x 18 amp rails vs. my single 70 amp rail.. I wouldn't think it would be an issue but it's something you both share...

Yeah, and we both have 3 hard drives and a solid state hooked up to our psu's as well.... I dunno, I just can't believe this PSU supposedly 'crossfire ready' can't handle 2 5850's. I mean, 850w should be ok right...?

[erocker]

Originally Posted by ArmoredCavalry

Yeah, and we both have 3 hard drives and a solid state hooked up to our psu's as well.... I dunno, I just can't believe this PSU supposedly 'crossfire ready' can't handle 2 5850's. I mean, 850w should be ok right...?

I'm also using 3 HDD's, a SSD and a 850w PSU. Yes, it should be ok, but maybe your rails just aren't cutting it... Try swapping around the PCI-E power cables perhaps.

[cadaveca]

Originally Posted by erocker

I'm also using 3 HDD's, a SSD and a 850w PSU. Yes, it should be ok, but maybe your rails just aren't cutting it... Try swapping around the PCI-E power cables perhaps.

I have tried swapping the cables. I have the 1200W TPQ, with 6 rails @ 38a. I really doubt PSU is the issue, considering I have each card on it's own 38a rail.(I have considered this, and will pick up a new unit tomorrow to ensure it's not the issue, based on the rest of my testing. If the bench works fine, AMD's driver team sucks for improper testing.)

Just finished testing 10.8 driver without profiles... same results. But swap results, as singlecard is faster. Exact same FPS.

Going to install profiles, reboot, and retest, then I will download the self-contained bench.

[erocker]

Bottom of this page: http://www.hardocp.com/article/2010/...a_performance/

..is the 10.8a application profile. Just so we are using the same thing.

I agree, your PSU is not the issue.

[cadaveca]

Yep, used the same app profile, no change. BTW, this app profile is now on the AMD.GAME website too. Downloading self-contained bench...


I'm really hoping I get better results, and all this is explained, or I dunno what to say.

Installed OS, updated OS, Installed DirectX, installed ahci drivers, installed chipset drivers, checked windows update, installed vga drivers, installed FarCry2(fresh download), installed app profiles(10.8a), actually got slightly better results without the profile.

[erocker]

I just don't get it...We are running very similar systems. Do you have PCI-E Link State Power Management disabled in Windows?

[cadaveca]

Yep. turned off screensaver and monitor sleep, as well as HDD sleep. Will gladly turn stuff back on if it helps, but I doubt it will.

I have been running into situations like this for months, unfortunately. Spent most of the day today installing new OS and testing in old and new OS, same results...I'm not the only one that gets these sort of results, either, so I'm at as much of a loss as you are. i've tried almost everything over the past year since getting these cards, except selling 'em ot get non-reference cards. Eyefinity6 card behaves the same.

It's why I've offered my systems to AMD...I will do without them and let them use them, if it helps, because I tell you, it's pretty frustrating, to say the least.

I mean, the last thing left is that the driver is cpu-limited, and your OC helps. So after trying the self-contained bench, I'll try OC and see if that makes the difference.

[cadaveca]

Originally Posted by erocker

One thing that I know has helped me with CrossFire is increasing the frequency of the north bridge (memory controller) but even so, you shouldn't be getting identical results to a single card.

Bang on here, e. upped cpu to 3.6ghz, nb to 2600mhz, htt to 2400mhz, ram from 1333 CAS9 to 1600 CAS6.


I guess this means that Crossfire is cpu limited with a 965BE @ stock. Now I've got to go over my other issues and see what happens..if all of this is a lack of cpu power, AMD still sucks, for not being able to optimize for non-overclocked systems. I bet whoever is working on their driver is working on an overclocked system...

Results:

[ArmoredCavalry]

Well, after reading your post, I oc'ed my q9550 to 3.40ghz with the following results:


At first I thought it might just be the cpu providing extra frames, but when I check msi afterburner, the GPU utilization on each card is definitely higher than it was before (about 70% now).

I never would have guessed the problem was my cpu bottlenecking.... not sure I will be able to overclock the q9550 enough to get it to a point where both gpu's are approaching 100% use. :\

[erocker]

Sandy Bridge is getting very tempting now... I really had no idea, I just always have my system OC'd.

[cadaveca]

Originally Posted by erocker

Sandy Bridge is getting very tempting now... I really had no idea, I just always have my system OC'd.

If you read my sig, I think it's asinine that you need to OC to benefit from Crossfire with Cypress-based cards.

Plain and simple, it's the DRIVER that is bottlenecking the cpu, and IT SHOULDN'T.

Further testing shows that going back to stock cpu, but UPPING MEMORY SUBSYSTEM SPEED provides near identical results, so really, it's not even about cpu, it seems. I'll have to test further to make that 100% for sure though.

I've been saying for MONTHS that Crossfire takes 3.6ghz cpu...looks like I was close to being 100% right there, but that I failed to look at how performance had increased in other parts of the system.


That still leaves me with the verdict that Crossfire should NOT be recommended for ANYONE, as getting it to work properly requires VOIDING YOUR WARRANTY.

I wonder what the lawbooks would say about that.


And thanks for your help on this, e, as without you, coming to this conclusion as to what's really going on would not have been so easy. As your handle infers..YOU ROCK!

[ArmoredCavalry]

Here are some charts showing 5850 crossfire performance w/ a q9550 at 4.0ghz vs an i7-960 at 4.0ghz:
http://forums.vr-zone.com/news-aroun...ll-hd-rec.html

Seems they are getting much better scaling than I am, at 3.7ghz. I can't get up to 4.0ghz without causing my system to become unstable, so looks like thats as good as it gets.

I think I'm gonna probably end up selling/returning that second 5850, I don't think the performance increase is worth the trouble/money, it certainly isn't what I was expecting from lots of benchmarks that I've seen.

Thanks to anyone who posted, especially erocker/cadaveca, I appreciate it.

[erocker]

Well, here's looking forward to ATi coming out with a big nasty single GPU card in the future.

[cadaveca]

Originally Posted by ArmoredCavalry

Thanks to anyone who posted, especially erocker/cadaveca, I appreciate it.

NO problem, dude. I have the time to test(Even install a new OS, as I did today) every single day, and that's a large part of why I'm on these forums. I spend all day at home with my kids, and have rigs in my livingroom so that I can pay attention to them while posting and doing simple things like running benches.

I just feel good knowing that not only does this issue not have anything to do with your system, or mine, it also was a good exercise to show how much high-end ATi gpus need high-end cpus.

Because of how I have increased frames here, CPU L3 speed has increased as well. I'm going to spend some more time testing this this weekend...I have a feeling that basically even single app will have this issue, and if thats the case, then AMD really does have some serious work to do with thier drivers.

I've very purposely been running my systems very close to stock or at stock, trying to sort out the cursor corruption issues with Eyefinity. Now I wonder if this issue isn't partly caused by the driver not being able to update fast enough, because of a lack of cpu grunt....and if AMD guys are developing drivers on overclocked systems, this would be a good reason as to why they have such issues replicating this...and again, if they ARE using overclocked systems for development of drivers, they really do need to be fired.

Originally Posted by erocker

Well, here's looking forward to ATi coming out with a big nasty single GPU card in the future.

I thought something was up with Crossfire results when they weren't as "stable" an singlecard...now to overclock to the point that Crossfire results become stable like singlecard(at which point, obviously, cpu limitations are gone). Might have to pull out the LN2...ugh...


Might explain why 5770 cards get such better scaling...very interesting situation, this one.

[mastrdrver]

That really surprises me how much the cpu is limiting performance.

I went and ran the FC2 bench too just to make sure since I'd been messing with my overclock on my 920 since I got a second 5870. I kicks the clocks back to stock except forced 21x cpu multiplier constantly, ram at 1600 7-7-7-20, uncore 2x, and qpi 3600. Windows 7 hd power off after 30 mins and moderate power savings on PCIe link are the only power saving options checked.

My scaling is ~75% on the bench (small ranch) from one card to two. Only setting I changed was set AI to advance and AA to multi sample. Everything else I use in game settings. Went and used CCC to overclock the cores to 900 left memory alone.

I'm using 10.4 Cats with 10.7 profiles fwiw.

[cadaveca]

Originally Posted by mastrdrver

That really surprises me how much the cpu is limiting performance.

I went and ran the FC2 bench too just to make sure since I'd been messing with my overclock on my 920 since I got a second 5870. I kicks the clocks back to stock except forced 21x cpu multiplier constantly, ram at 1600 7-7-7-20, uncore 2x, and qpi 3600. Windows 7 hd power off after 30 mins and moderate power savings on PCIe link are the only power saving options checked.

My scaling is ~75% on the bench (small ranch) from one card to two. Only setting I changed was set AI to advance and AA to multi sample. Everything else I use in game settings. Went and used CCC to overclock the cores to 900 left memory alone.

I'm using 10.4 Cats with 10.7 profiles fwiw.

It's not the cpu, exactly. @ 3.4ghz, i get max 90% cpu load, avg 81%. And that's after upping the NB/memory speed.

Up to 3.6ghz makes for a small gain(another 4 FPS), and cpu loading goes down(roughly 5 %). If the raw mhz was the issue, the cpu load wouldn't have lessened. 3.4 ghz is enough, and the bottleneck lies elsewhere.

Fact of the matter is that the driver's use of cpu sucks( jsut like how cpu's use of driver sucked in BFBC2, and that's what AMD has reported as fixing for the one map with this 10.8a app profile). There's no reason for this to happen, to me, as they fixed it in BFBC2...just looks like they aren't looking at the big picture, and they need to apply the same fix across the board. but I've alread said that in another thread...

your results are better, most likely, due to triple-channel mem.

I'll find the exact bottleneck...I'm on a mission, now. Might have to pick up an INtel cpu and board tomorrow to see how that affects things.

[mastrdrver]

Give me a minute and I'll take a memory stick out but I don't think triple channel would help. Be back...

[Mussels]

i'm not entirely surpised to see its CPU bottlenecking, remember that all these DX9 and DX10 games have only one thread for graphics - so you're being CPU limited by one thread, it cant use your other cores.


my 4.2GHz xeon bottled my 2x4870, so i can imagine 2x5850 requiring some beef to max it out... i just figured you'd get more performance than a 10% or so boost :P



time to OC, both of you!

[cadaveca]

Originally Posted by Mussels

my 4.2GHz xeon bottled my 2x4870, so i can imagine 2x5850 requiring some beef to max it out... i just figured you'd get more performance than a 10% or so boost :P

I do not see a single thread maxing out a core though. I wish it was so simple...but it's not. I think AMD would really have no excuse if that was truly the case.


And that last bit..OC...really, should NOT be nessecary. I was fine with my qx9650@ 3.6ghz, and was getting 100% load on both of my 4890's, but again, AMD's driver has DRASTICALLY changed since those times. It's that change that has lead to the super-small gains to be had with stock cpus.

I have good idea as to what's going on, but I'm to willing to go into it until I'm 100% for sure. At least I now have some direction though...I'll be posting ALOT more about this exact issue in the days to come.