I should confess: I didn't buy a second GTX 970. But as of writing this article, I have two GTX 970 Gaming cards in possession. The second card is a replacement for the first card. I actually went ahead with RMAing the first one as it exhibited some coil whining. Although it is not as heavy as some of the other reports as I've already mentioned in a previous post, I did not want it to develop to a utter irritation. So I decided to play safe.
So, I informed Amazon that I am going ahead with the replacement. They sent me a second card and asked me to ship the old one back to them before a deadline which is quite far away. Being the opportunist that I am, I thought of making full use of this.
A little bit of background for the late-comers
When I shrunk my old ATX PC to MICRO-ATX and not MINI-ITX, I wanted to keep my options of going for a Multi-GPU setup open. When NVIDIA launched the GTX 900 series cards with minuscule power draw compared to what it was replacing as well as the competition, that became almost a reality...well, at least a feasible option. When I assessed the way I should go about upgrading the graphics subsystem in the future, SLI appeared economically leaps and bounds ahead of going with a single GPU high-end card. So, SLI is happening people; just the budget needs to be passed.
But is it only the budget?
But there's a big issue. Will it blend? No, seriously, will it work inside the limited confines of my Silverstone SG09 chassis? Even more than the chassis itself, the PCI-E slot layout of the Maximus VII Gene board that I own would become an issue. This is because, there won't be free slots in between the cards for when installed. The card on top will run extremely hot because it won't get enough air to breathe and even that small amount of air will be warm. This is why people steer clear of graphics cards with open type coolers in a SLI configuration especially if it is MICRO-ATX.
But I wanted to see this from my own eyes!
When the second card arrived, I quickly moved the old card to the bottom position and installed the new one on the top position. This was done so that when it was finally time for me to return the first card, I don't have to switch the slots. (It was just a matter of doing it at the beginning or doing it at the end.)
Here is a pic for your enjoyment.
As you can see, the cards are almost kissing each other - means, there is hardly any breathing room for both cards. The top card was in a bad shape because the heat from the back of the bottom card would directly enter the cooler. I heard that the PCB of these cards heat up considerably.
Now I must first reveal what fan setup was.
- 180mm fan on the roof as exhaust
- 120mm fan on the rear as exhaust
- 92mm x 2 + 120mm on the side as intake
I ran 3DMark Firestrike to see what kind of temps I would get. I did not really care about the card on the bottom because it would always run cooler. I thought that the top card would not reach 85C, but to my amazement, it hit 91C!!! And this was just one round - not an extensive gaming session. And the GPU clocks would only boost to around 1430MHz with the same settings which would have boosted the old card past 1500MHz.
I knew it would be bad, but I didn't think it would be this bad.
So what I did next was reversing the 2x92mm fans as exhaust. The reason for my thinking was that while directly feeding cool air into the cards would seem like it would improve the temps, it would also trap the hot air between the cards as these 92mm fans are not powerful. Turning them around would let the trapped hot air exhausted out of the case, and the vacuum effect would bring cool air from somewhere.
I did the alteration and retested. I could now feel the hot air flowing over my hands. Despite my thinking which I think is logical enough, the max temps of the top card went up by 1C.
There are a couple of things that I could do.
- I can restore the fan on the roof back to how it originally was supposed to operate - as an intake fan. This might lower the temps if the cards are suffocating. The two small guys can exhaust the warm air.
- Make the card in the bottom the primary card might lower the temps of the card on the top because it is a popular belief that the card which the display is connected to has to do more work this running hotter.
So I changed the orientation of the fan and tested only to find out that it did absolutely nothing to improve the temps. The GPU still hit 92C while running 3DMark Firestrike.
Then I made the second card the primary card. To do this, I had to first disable SLI in the driver, shutdown the PC, plug the display cable to the DVI port of the card in the bottom and reboot. If I switched the display while the PC was booting up, it would not be recognized. So I had to change the ports first and then power the PC on. I'm not sure if this is just how this particular motherboard works or if this is the case in general. If I hadn't turned off SLI, I would get a blank screen when you reach the welcome screen. I also took this as an opportunity to update to the latest 344.75 drivers.
Did it improve the temps? I sure did NOT! Max temps were unchanged. That god-awful 92C was still being registered as the max temps.
That 92C might be actually lower than it should be?!
There is a driver issue haunting the SLI setups of GTX 900 series cards. That is the secondary card always run at a lower voltage than the primary card. The MSI card ramps up the voltage up to 1.2060V as per GPU-Z when you don't touch the voltages in MSI Afterburner. But the card on the bottom would only hit around 1.137V. That's 75mv less, which is significant. Therefore it does no boost as high as it would when it was operating in solitude.
The fact that the temps of the card on top ran at 92C regardless of whether it was the primary card or not, means that the temps of the card was equally affected by the voltage of itself as well as the voltage of the other.
- When it was the primary card, it would be running at 1.2060V when boosted. But the other card would be running at a much lower voltage so the heat dissipated from it is lower. I saw that it never hit 70C at that voltage. It would have hit 75C or so if it was running properly, and that additional heat would affect the temps of the card on top.
- When it was running as the secondary card, despite running at a lower voltage (which I couldn't check but it should be around 1.137V) it still hit 92C which means the extra heat from the card on the bottom, which is now the primary card, was equally influencing. If this SLI issue wasn't there, the card on top would run at a higher voltage which would have in turn increased the temps further.
Bottom line is that it could have been worse. But it does not mean that the current situation is more or less usable.
SLI with two MSI GTX 970 is impossible in a MICRO-ATX case.
One more thing. The top card was sagging a lot compared to the bottom one and this made the cards get closer as you moved in the direction towards the front of the case from the back. They were almost touching as the edge. No wonder that the top card was running so hot. I tried to add a spacer in between the cards but it was rushed up and didn't really work as expected. But I doubt that an additional 1cm would result in dramatically different results.
So I removed the card from the case and packed it in its original box to be shipped back to Amazon. Now I have to think about what I'm going to do.