7950GX2 Only working on 1 GPU problem. Help Appreciated

The Asgard · 3 Sep 2006 at 09:53

Greetings,
While I have been playing with my new toy I have noticed occasionally that the performance suddenly drops by 40-50% for no apparant reason. Its like the SLI is breaking or one of the GPU's is shutting down. Intially this problem only occurred if I went into the Nvidia control panel but I updated the Bios on my P5W DH from 1001 to 1305 and it appeared the fix the problem. But benching this morning I have found that I still have the problem all be less frequent.

For example in normal operation 3D mark 2006 scores 55-60 on proxy, when its in this 1 gpu mode I get 30-33. Its the same in Fear, Normal 44-69-392, 1 GPU mode is 20-36-79.

All I have been doing this morning is monitoring the GPU temp via the Nvida Temp Monitoring page and putting some extra cooling over the card Via a 120mm fan.

Also when its in this state the fans don't spin up while underload. Very strange and I'm not sure it would be apparant unless you could actually see your FPS. The only way to get the card fully working when its in this state is to reboot.

I'm running a P5W DH with 1305 Bios. Latest Nvidia Official Drivers. Fresh Install of XP. Idle temp's around 55'c with load temps about 70-75'c

Anyone else having problems? Could just be a faulty board?

The Asgard · 3 Sep 2006 at 10:35

Coolasmoo said:
Sounds like a power issue, but ive only used the power hungry ATi range so i cant comment to much on Nvidia.

I was going to suggest setting PEg link mode to slow in bios but i cant see that doing much really. You could try running slot power to heavy maybe.

These sort of problems you could do with testing in a second PC really to compare , but its not always viable.

Well I was running an X1900 XT-X previously with no issues. My Tagan 580W is pretty amazing on all the rails with a slight overvolt on all even underload. Common on all Tagan 580's. I have a Seasonic 600W I could try but it doesn't feel like a PSU issue. I know there are a few using the P5W DH with no reports of problems. Very strange. Never used Sli before apart from a quick play with 2 6600 a couple of years ago.

Only got the card Friday so I will just send the thing back if I can't resolve today.

Its really strange that the card doesn't crash the OS when it does this. Its almost like its a controlled shutdown due to temps

Cheers for the input m8 will give your suggestions a try

The Asgard · 3 Sep 2006 at 12:41

Ping said:
what sort of temps are u looking at on both gpu,s. Could be throttling down the speed due too heat? Everest shows both gpu temps. It might be worth a look.

Will have a look at Everest cheers

The Asgard · 3 Sep 2006 at 12:48

Morfik said:
Asgard, I have actually experienced problems like this too. That's part of the reason I was checking the Link Width in the other thread as I don't know what's wrong.

Whilst I haven't experienced it with benchmarking (probably as I dont do much of it) it tends to happen when gaming. Whilst I realise games are much less of a controlled environment for system load and it could be a million things I tend to notice that during a session of gaming there are prolonged times when my framerate cuts in half for no apparent reason. No I don't mean during a fire fight, I can be alone in an area of the map that I normally get 90fps in and for no obvious reason i'm only getting ~40.

I had several theories some of which may be complete rubbish and I dont really have any evidence for any but here goes:

a) A problem with my MSI 7950 gx2 - Seems unlikely as you have a different make

b) A problem with my motherboard - Possible, it's brand new and the bios is only at its first revision and it's known to have issues with Quad SLI so perhaps single 7950 gx2 performance is bugged too

c) Card overheating - I have good airflow through my case so it seems unlikely. As mentioned above Everest is the only program that can monitor both GPU cores though I have no idea if it's accurate as not even Rivatuner or the Official Nvidia Control Panel can do this. Temps of the inner GPU were hitting 110c or so during load, the cards brand new so i can only assume this is normal?

d) Conroe throttling - I had overheating problems so this crossed my mind but as you're water cooled and don't have this problem it can't be that

e) Power supply - I have a 700w FSP PSU so it's unlikely to be this

As I said, no idea if any of these ideas carry any weight, if you manage to solve it let me know.

Sounds exactly the same problem m8. When your in this mode just try running 3D mark and looks at the FPS I bet its halfed. Try doing 3D Mark 06 from first boot and make a note of the FPS in Proxy.

I need to do some more testing but it looks like an overheating problem. I'm going to check with everest. I have asked the Nvidia drivers to reports any thresholds exceeded but so far I have had none reported. Very strange but depending on the game this problem might be happening a lot but people are not seeing it because even 1 card is still fast enough for most things.

The Asgard · 3 Sep 2006 at 12:50

Amidar said:
I would say thats it's got nothing to do with the PSU,go into the conrol panel and make sure 3D Mark is set to "Single display performance mode" in the programs settings,as when you first install the driver it sets to "Multiple display performance mode" you find it set to this in all the games & benchmarks in the program settings as this seem's to be Nvidia's default setting.So what ever game or benchmark you use that has an SLI profile make sure you change it to "single display performance mode".
The other thing i have noticed is by using the classic control panel it seems half the SLI profiles are missing so if possible use the new control panel.

I tried that but it had no effect. Only way of getting out of this mode is a reboot. Default settings work fine until this borked mode kicks in.

The Asgard · 3 Sep 2006 at 13:42

locutus12 said:
given all the 7950GX2 is is 2 X 7900GT`s bolted together in an ugly design and linked with an SLI connector bridge to allow them to run through one PCI-E socket, ... is it not possible that one of the boards is simply dead on arrival or at least faulty?

Possibly. One card could be faulty certainly some issue with mine. The one thing I don't like about Sli is you don't really know if its working. Especially with some games.

The Asgard · 3 Sep 2006 at 19:03

w3bbo said:
I had very similar problems.

The way I sorted it was to ensure I had 16x pciE lanes open in one slot. Before I only had 8 + 8.

How did you do that m8?

Did a fresh install today

The Asgard · 3 Sep 2006 at 21:44

Running some more test and a run at 3D Mark 2006 with Everest Logging

Overclocked 550/700

GPU1 °C Min 55 Max 80
GPU2 °C Min 64 Max 86

Compared to my X1900 under air they are slightly less

Still working fine.

The Asgard · 3 Sep 2006 at 22:26

Right its back in its borked state and I have just ran 3D mark 06 with everest running.

GPU1 °C Min 62 Max 80
GPU2 °C Min 67 Max 70

Looks like GPU2 is not working

The Asgard · 3 Sep 2006 at 23:15

Morfik said:
That looks pretty conclusive Asgard. The temps in the 1st instance weren't hot enough to cause it to slowdown the second GPU though - why is it doing it?

I have not got a clue.

If I force the driver back to single gpu and then back to sli it all starts working again. Something is causing the Sli to drop out.

Just checked the event log, nothing.

Just going to send the card back obviously there is a fault. Not sure what to do for a replacement now.

The Asgard · 3 Sep 2006 at 23:19

MeatLoaf said:
Do you have another mobo to try it in? Or a mate with a PCI-E system with a compatible mobo?

I have access to loads m8. I even have a spare P5W DH in my case, lol. I'm at the point now where I am pretty sick of messing around now. Spent quiet a few hours testing, reinstalling ...... I had no problem with my X1900 XT-X and I know other people have had no problems with the P5W.

Might give a X1900 Xfire system a go

The Asgard · 4 Sep 2006 at 06:56

redeye said:
I've had exactly the same problem as The Asgard. I'm running an XFX 7950GX2 on an Asus P5BDLX WiFi, E6700. Problem started in similar circumstances, running 3Dmark06 on Proxycon when the frame rate dropped from 50+ to around 25, games became slow etc.

I tried everything- re-installed the drivers, re-installed windows, checked the temps (all seemed ok). So spent 30mins on the phone to XFX support in the states (it was late at night and they were really helpful) who could only suggest everything I had already done. They thought it very unlikely it was a card issue as they are apparently all individually tested before they leave the factor. So I changed the power supply (was running a Silverstone ST60F 600W), changed the motherboard for another P5BDLX and bought another 7950. I re-installed windows and guess what, it works perfectly. I benchmarked it for days and still is fine. I put the old 7950 in and it was still around 50% slow so definately the card is borked!

I've returned the card and the manufacturer has agreed it sounds like one of the cores has died. There seems to be an alarming high failure rate with these cards, probably due to the shocking cooling on them!

Was it an intermittent fault or had one board/gpu died from the start?

Just had mine on this morning for 3 hours running Orthos and Oblivion in a nice tree/grass area. Expected it to fail and it was still working. Max temps were 80/90 on gpu 1/2. This is the worst kind of fault.

Where did you get the info from for high failure rates on these cards m8?

The Asgard · 4 Sep 2006 at 14:00

redeye said:
I've just read a lot of threads on various forums with people having similar problems. Probably a bit OTT of me to say "alarming high", I've just had enough of these cards after the amount of time, effort and money spent to try and sort it out.

Well mine is going back and I have requested a refund. I pretty much spent all weekend messing with the thing. While its an amazing card when it was working I can't help thinking how reliable it would be in the long run. Far too much heat on the PCB for my liking.

Not sure what to buy now. The X1900XT-X was a great card but I needed more which is why the 7950 is so attractive. ummm

The Asgard · 4 Sep 2006 at 14:19

MeatLoaf said:
How about a pair of theose 7900GTOs, assuming you have an SLI mobo.

I'm on a P5W DH m8

The Asgard · 4 Sep 2006 at 15:11

MeatLoaf said:
Ahh.

Only really 3 choices then, 7900GTX/GTO, X1900XT/XTX or an X1950XTX

Only option at the moment really is back to a X1900 or X1900 Xfire.

The Asgard · 5 Sep 2006 at 11:57

Elvis Himselvis said:
Is there any way to conclusively check whether both GPU cores are operating on the 7950 GX2?

With SLi set-ups you could have an on-screen bar which showed the load split between the two cards (indicating both cards were performing), this doesn't seem possible with the 7950 (or is it a driver issue?). Riva Tuner, cool bits etc seem to detect it as a single card/core rather than as an SLi solution (which I ssume is normal for this card) as a result it's hard to know exactly what is or is not happening.

Everest shows varying temps on the two cores, with one much hotter than the other, but does this prove anything? Bearing in mind the design of the card I would expect the lower GPU to get warmer as airflow is more restricted.

Surely there must be some way to check the two GPUs are working.

Temps are the key. I used everest to confirm 1 GPU was not working. Determine Idle temps for the core and if its withing a few degree's under load it isn't working.

The Asgard · 5 Sep 2006 at 12:36

there did used to some load balancing graphics you could enable on SLI that appeared within the game

The Asgard · 5 Sep 2006 at 23:25

Morfik said:
Been away for the past few days so wasn't able to do anymore testing, but now I'm back i'm fairly sure what's wrong with my card.

I don't think the fan on the inner GPU is working at all. Under load the temperature of the inner GPU goes up to 115c and i'm guessing is being shutdown to prevent damage which is why my performance suddenly drops. Even several minutes after when the outer gpu has had time to cool from around 76c under full load to 66c mostly idling (in windows) the inner gpu is still at 105c. Something's clearly wrong with its cooling.

Can I RMA the card direct to MSI do you think? Suffice to say that returning anything to the company I bought it off (that I can't name) is an incredible hassle and just not worth the effort so i'd prefer to deal with MSI direct.

Glad you have sorted the cause of your problem. Wished mine was that easy.

I would always deal with the supplier first. I know some manu won't deal direct. Best give MSI a ring first and make sure you get an RMA and make sure your insured so if the card goes missing your covered.

The Asgard · 6 Sep 2006 at 16:14

Well its looking like mine was just faulty. fornowagain bought an EVGA version and so far its working perfectly. Early days because he only got it last night but it was pretty obvious in a couple of hours of testing that mine was faulty.

Really have to say that when my 7950 was actually working it was an awesome card. I have pretty much decided to give the first round of DX10 cards a miss and get something to tied me over until the 2nd gen. I can't think of a better card than the 7950GX2 to do this. My next card will definately be an EVGA because of the warranty and option for upgrade in 90 days time.

I would say go for the 7950GX2

The Asgard · 6 Sep 2006 at 21:26

I would probably look into a way of extending the bridge/sli link. Its a pity Nvidia didn't give you 2 different size bridges with some more spaces.