Poor SLi Performance

[shamus8888]

Hi all,

I just finished building my i7 rig and the specs are in my sig.

The problem is I am getting bad 3DMark Vantage scores. With my highest overclock at the moment (also in sig) I only managed to get around 13k marks @ Performance settings in 3DMark.

I also noticed the GTX275s runs a bit hot. I know that the first card would be hotter since its sucking in hot air from the second card, but even with my cooling sorted , it would idle @ around 69oCelcius (Room temp around 26oC).

I made sure the case (CoolerMaster ATCS840 with custom made 120mm side fan exhaust) has proper cooling but nothing seems to help the temp.

But the main thing I am worried about is the performance. I have seen single GTX275s getting better scores. The cards are utilizing the first and second PCEe x 16 lanes on my mobo (rampage 2 extreme) as it seems to be the only option.

Any ideas why?

 

[OCGuy]

What is your score with SLI disabled?

Are both cores ramping up to 3d mode like they are supposed to?

 

[shamus8888]

Originally posted by: OCguy
What is your score with SLI disabled?

Are both cores ramping up to 3d mode like they are supposed to?

Interesting, never tried them with SLi off.

How would I check if they are ramping like they are supposed to?

I am using EVGA Precision tools to OC and monitor temps since I have problems with Rivatuner with Vista 64.

 

[OCGuy]

Originally posted by: shamus8888

Originally posted by: OCguy
What is your score with SLI disabled?

Are both cores ramping up to 3d mode like they are supposed to?

Interesting, never tried them with SLi off.

How would I check if they are ramping like they are supposed to?

I am using EVGA Precision tools to OC and monitor temps since I have problems with Rivatuner with Vista 64.

It should show you your core speed. Idle should be ~300 and then it should ramp up to full in 3d apps.

See what your score is with only one card enabled. If it is the same score, switch the card positions and test again.

If you have 2 cards that you know work by themselves, than it could be your mobo or bridge or.....?

 

[shamus8888]

Pretty sure both cards work since I had to take the whole thing apart once and have switched them around. Besides, I got the option to enable SLi so that also means that both cards are detected.
I will use EVGA to monitor my core speeds and see if they ramp up. Problem is when I run benchmarks in 3Dmark its in fullscreen so its kinna hard to monitor the core speed might try to load a game or something...

 

[shamus8888]

Hmm, at the moment I am idle on desktop. I have also reset the clock speeds back to stock (633Mhz core, 1134 Memory and 1404 Shader). It shows that GPU1 has 633 clock speed while GPU2 has around 301. Values for shader and memory are also much lower on GPU2.
Will try to launch and instance of WoW in windowed mode and see how things are.

edit : GPU2's clock values doesn't change. It doesn't ramp up and stayed at a much lower value. What's wrong here :S

 

[yh125d]

Your side fan should be intake, not exhaust. If it's exhausting then it probably explains your temps

 

[cusideabelincoln]

GPU2 not ramping up would be the problem. Tried reinstalling the drivers?

 

[Soccerman06]

Originally posted by: shamus8888
Hmm, at the moment I am idle on desktop. I have also reset the clock speeds back to stock (633Mhz core, 1134 Memory and 1404 Shader). It shows that GPU1 has 633 clock speed while GPU2 has around 301. Values for shader and memory are also much lower on GPU2.
Will try to launch and instance of WoW in windowed mode and see how things are.

edit : GPU2's clock values doesn't change. It doesn't ramp up and stayed at a much lower value. What's wrong here :S

The card is overheating and defualts to 2d temp so it wont die

 

[OCGuy]

Originally posted by: shamus8888
Hmm, at the moment I am idle on desktop. I have also reset the clock speeds back to stock (633Mhz core, 1134 Memory and 1404 Shader). It shows that GPU1 has 633 clock speed while GPU2 has around 301. Values for shader and memory are also much lower on GPU2.
Will try to launch and instance of WoW in windowed mode and see how things are.

edit : GPU2's clock values doesn't change. It doesn't ramp up and stayed at a much lower value. What's wrong here :S

OK go into Rivatuner. There should be a "Force 3d mode" option. This should force both cards to be at the 633mhz at all times. Test your scores to make sure that was the problem.

Note: I am not saying you should have to do this to enjoy your system to its full potential, I am just trying to be sure your 2nd card is not defective.

You should play around with different drivers and see if that helps.

 

[shamus8888]

Originally posted by: yh125d
Your side fan should be intake, not exhaust. If it's exhausting then it probably explains your temps

My GFX card is a Palit gtx 275 which has custom coolers that exhaust hot air inside the case. So I drilled a 120mm in my case to exhaust the hot air out. I tried intake but that just blew the hot air back in.

Am I doing it wrong?

 

[shamus8888]

Alright, I updated to the latest nvidia driver (186.18) and things seemed to be improving on stock speeds w/o OCing the card. I am hitting the 16k mark now in Vantage (although I had to manually adjust some performance settings in the Nvidia control panel). Temps are still high (second card reached 105 for 25 seconds) but I suspect thats because of the overall ambient temp. It is around 27oC in my room at the moment. Very hot night ><

Weird thing is that before I can successfully install the newest driver I had to uninstall the old drivers first. Normally I would just install over the old ones but when I did this it crashed the whole system.

Gonna do further tests to see if things improves more.

 

[shamus8888]

Right, noticed another odd thing... my 1st GPU doesn't ramp down to 2D settings it seems. It's always on max core clock ><

 

[shamus8888]

no one knows? :0

 

[yh125d]

How much room is between the fan intake on the top card and the PCB of the lower card? Since its a non reference cooler (which has a notch to make intake easier with SLI) the fans on the top card are possibly choking really bad. I think all three slots on your motherboard are full bandwidth, so you could try using the lower slot to see if that helps temps. I also saw that your GPU fans are at a constant 1200rpm (which is kind of low for 80mm), so they don't ramp up with temperature. Add to that that you've overclocked your GPUs a lot, and I can definitely see how you'd get high temps

As for the side fan, there's two ways to look at it really. It could be exhausting some of the hot air, but I think it would be better used to supply the fans with fresh air from outside, as well as blowing the hot air away from your GPU's, for the other exhaust fans to deal with. I'd try it out reversed an see which works better in your scenario

 

[Woofmeister]

Originally posted by: yh125d
How much room is between the fan intake on the top card and the PCB of the lower card? Since its a non reference cooler (which has a notch to make intake easier with SLI) the fans on the top card are possibly choking really bad. I think all three slots on your motherboard are full bandwidth, so you could try using the lower slot to see if that helps temps. I also saw that your GPU fans are at a constant 1200rpm (which is kind of low for 80mm), so they don't ramp up with temperature. Add to that that you've overclocked your GPUs a lot, and I can definitely see how you'd get high temps

As for the side fan, there's two ways to look at it really. It could be exhausting some of the hot air, but I think it would be better used to supply the fans with fresh air from outside, as well as blowing the hot air away from your GPU's, for the other exhaust fans to deal with. I'd try it out reversed an see which works better in your scenario

This.

Take it from somebody who's tried it both ways. You will get much better temps in most cases by using a side fan to intake cool air to your graphics solution rather than exhaust hot air. In contrast to some of the older NVIDIA heatsink designs, your 275s actually will exhaust hot air out the back of the PCI slots. Just make sure that any fans at the back of the case are set to exhaust rather than intake to avoid pulling exhausted hot air from the video cards back into the case.

 

[shamus8888]

Originally posted by: yh125d
How much room is between the fan intake on the top card and the PCB of the lower card? Since its a non reference cooler (which has a notch to make intake easier with SLI) the fans on the top card are possibly choking really bad. I think all three slots on your motherboard are full bandwidth, so you could try using the lower slot to see if that helps temps. I also saw that your GPU fans are at a constant 1200rpm (which is kind of low for 80mm), so they don't ramp up with temperature. Add to that that you've overclocked your GPUs a lot, and I can definitely see how you'd get high temps

As for the side fan, there's two ways to look at it really. It could be exhausting some of the hot air, but I think it would be better used to supply the fans with fresh air from outside, as well as blowing the hot air away from your GPU's, for the other exhaust fans to deal with. I'd try it out reversed an see which works better in your scenario

Thanks for your reply. The spacing between the two cards are very tight unfortunately. Around 10 cm or less from the 1st card's intake fan to the 2nd cards PCB (bottom side). I am thinking of putting a non conductive material between them so that the heat from the second PCB will be dissapated elsewhere instead of being sucked in. But I am afraid the 1st card's fans will suck that material in instead :S

I will try to make the side fan intake. The thing is during summer it can be quite hot in my room. Ambient of around 25oC upwards. So I am afraid its just sucking in warm air to the cards.

Latest update: I managed to bring the temps down and performance up. It seems that I need to mess with the global settings in the NVidia control panel a bit. Also the reason why the second card if not ramping down is because I have a second display (LCD TV) connected to the card for watching movies. If I disable the second display it ramps down. Only weird thing is sometimes it wont ramp down again immediately after I run a round of 3D Mark. It takes around 5-10 min for it to ramp back down.

One thing I noticed is if I set the Multi display/mixed GPU acceleration to Single Display the second card will fail 3DMark Vantage (temps reached over 105 degrees). I need to leave it in Multiple Display Performance Mode.

So in conclusion, what I have found so far is :

If you have the problem of one of the cards not ramping down to 2D mode in SLi setup, disable your second display if you have one.

 

[shamus8888]

Originally posted by: Woofmeister

This.

Take it from somebody who's tried it both ways. You will get much better temps in most cases by using a side fan to intake cool air to your graphics solution rather than exhaust hot air. In contrast to some of the older NVIDIA heatsink designs, your 275s actually will exhaust hot air out the back of the PCI slots. Just make sure that any fans at the back of the case are set to exhaust rather than intake to avoid pulling exhausted hot air from the video cards back into the case.

Hi, thanks for your reply.

My cards are Palit GTX 275s. Some pictures I am not 100% sure (since I can't find a proper air flow diagram for it) but I have a hunch the hot air exhausts from the side fins.

The ATCS840 case also have a custom exhaust box attached to the GPU slots behind the case to suck out hot air but I am not sure how efficient it is since I don't even know if this card exhausts any hot air from the rear.

The clearance between the cards are really bad. This is the best pic I can find. I have two cards instead of three.

Can I insert some sort of non conductive material between them so the top (1st) card won't suck in hot air from the PCB of the 2nd card?

I really do not want to resort to water cooling since I have zero experience with them and it seems very complicated ><

 

[yh125d]

I wouldn't bother with putting any material between the cards, if anything I think that would be likely to worsen the situation. You could try using some masking tape on the side vents (on the two long sides of the cards) to direct more airflow out the back instead of out the side


Couldn't you just drop the lower card to the white PCI-e slot though?

 

[shamus8888]

Originally posted by: yh125d
I wouldn't bother with putting any material between the cards, if anything I think that would be likely to worsen the situation. You could try using some masking tape on the side vents (on the two long sides of the cards) to direct more airflow out the back instead of out the side


Couldn't you just drop the lower card to the white PCI-e slot though?

Hmm, according to the manual it's best to use 1st and second slots for duo SLI. I will try to find out more.

Also now I noticed that whenever I try to run any benchmarks my second card's temp will shoot right up to 105oC while the 1st (top) card is only 88oC... On idle the top card hovers around 60 while second around 56...

Really don't know whats wrong now... any idea where I can get help on this matter? Google seems to be failing me.

edit: right found this thread. Thing is if I put my cards in the 1st PCIe and last PCIe slot would it still run at 16/16? Doesn't seem answered there :S

 

[Woofmeister]

Originally posted by: shamus8888

Originally posted by: Woofmeister

This.

Take it from somebody who's tried it both ways. You will get much better temps in most cases by using a side fan to intake cool air to your graphics solution rather than exhaust hot air. In contrast to some of the older NVIDIA heatsink designs, your 275s actually will exhaust hot air out the back of the PCI slots. Just make sure that any fans at the back of the case are set to exhaust rather than intake to avoid pulling exhausted hot air from the video cards back into the case.

Hi, thanks for your reply.

My cards are Palit GTX 275s. Some pictures I am not 100% sure (since I can't find a proper air flow diagram for it) but I have a hunch the hot air exhausts from the side fins.

The ATCS840 case also have a custom exhaust box attached to the GPU slots behind the case to suck out hot air but I am not sure how efficient it is since I don't even know if this card exhausts any hot air from the rear.

The clearance between the cards are really bad. This is the best pic I can find. I have two cards instead of three.

Can I insert some sort of non conductive material between them so the top (1st) card won't suck in hot air from the PCB of the 2nd card?

I really do not want to resort to water cooling since I have zero experience with them and it seems very complicated ><

Didn't realize the Palit used a custom cooling solution. Nice big fans, but no way to exhaust the hot air off the cards out of the case. More card clearance will help a little, but I wouldn't hold my breath.

I'd definitely explore getting a case with a good side fan before I went to the expense and complexity of water cooling. Check out the Enermax Chakra or the Cooler Master RC-690-KKN1 if you don't have a massively large CPU cooler.

 

[dust]

+1 for changing the exhaust fan into intake. I fixed three intake fans on the removable side of the case and my cards reach 55c under load( ATI cards run pretty hot usually). It helps keeping the CPU temp lower as well.I'd check also the temp for the SB,NB. Dunno abt Extreme, but Formula comes with a small fan you can fix to push air on the NB. It brings the temp down by at least 5c for the NB.

 

[yh125d]

I'm pretty sure the third slot will auto arrange lanes to 16x for both slots. Even if it can't, you'll have at least 16x/8x, which is enough bandwidth for GTX275. Use the top blue and bottom white slot

 

[shamus8888]

Thanks for all the kind replies!

Unfortunately, as my case features a removable mobo tray, I can't really utitlize the third PCIe lane So I guess that option is out. I am running the cards with the side panel off at the moment with ambient room temp of 22oC. GPU idles at 55 and GPU at 48. According to some thats still too high :S

Still getting the CPU overheating in FurMark issue. I tried switchin the cards around but still same. I have also tried single cards. I get idling around mid 40s while load around 89-92 (which panels off)

I have spoken to Scan (scan.co.uk) whom I have bought the cards from. According to them the will issue an RMA notice and test my card but if its not faulty I will have to fork up the courier charge to return the cards to me. The cost which could total up to £90!!!

Really stumped now... I am 90% positive they will just say the card is not faulty since they will probably test the card individually (non SLI) and theorectically it works...just annoyingly high temps.