Overclock SLI 970s?

[king3pj]

First, I play at 1440p and my system performs well at this resolution. The only game where I have had to turn down any settings is Fallout 4. I was getting dips into the 40s in some cities but I tweaked a couple graphics settings and I'm locked at 60 FPS now.

The point is that performance hasn't really been an issue for my PC. Fallout 4 was the only exception and it was easily fixed. I still like to tinker with my build so I was considering overclocking my 970s anyways. Here are my specs.

i5-4690k at 4.4GHz
8GB 1600MHz RAM
Gigabyte GA-Z97X-UD3H-BK
2X EVGA ACX 2.0 SC GTX 970
EVGA G2 Supernova 850W PSU

I had a couple concerns about overclocking. First I have never overclocked a GPU before but I have experience overclocking both Intel and AMD CPUs. I assume that overclocking a SLI build is more complicated because you want both GPUs to remain at identical speeds and it might be tough to tell which one is causing instability. Like I said, I've never done this so I may be wrong.

Second, I've read in other threads here that a typical GPU overclock will only give you an extra 2-5 FPS. Is this true? If that is the case I'm not sure it would be worth the extra heat, fan noise, and power draw.

Lastly, Is my power supply even strong enough to handle an overclocked CPU and 2 overclocked 970s. Everything has been running well so far but I'm not sure how much the power requirements go up when you start overclocking GPUs.

 

[dudmont]

Nvidia experience is the place to start. Set the system to run on only one card and tune it up. Keep tuning it till you're unstable, saving all your settings as you go. When you get unstable, set it back to last stable settings. Set experience back to SLI, and try both cards at the fastest stable settings of card 1. If it's not stable(and it probably won't be), keep dialing back.

 

[davmazin]

if its for syntetic test that has a point but if its for gaming sli is useless i am still dissapointed in sli and i dont think i will ever go back to it to be honest it fell stupid that a single radeon 4870 could destroy my 8800gt setup dunno how are thinigs now but it still stings

 

[king3pj]

The process seems simple enough. I guess I'm just not sure if my power supply is good enough for an overclocked CPU and 2 overclocked GPUs.

I'm also not sure if it will make an actual performance improvement. I have read threads here claiming that it will only increase by 2-5 FPS in graphically intense games. I'm hoping for a bigger boost than that to make the increased heat and fan noise worth it.

 

[dudmont]

I've got a P2 1200 in mine, 4790k(uses a good bit more power than your 4690k) and a pair of 780tis(they are both major hogs each has a pair of 8pin power connectors). I bet my system would run on your psu, as long as I didn't overclock.
Will it help? Only 1 way to really know, and that's to do it. Think about this, you might get 10-20% more clock speed and a boost to ram speed too. I think you'll do better than you think.

 

[king3pj]

if its for syntetic test that has a point but if its for gaming sli is useless i am still dissapointed in sli and i dont think i will ever go back to it to be honest it fell stupid that a single radeon 4870 could destroy my 8800gt setup dunno how are thinigs now but it still stings

SLI is not useless. I have been using SLI 970s for around 6 months now and it has been a huge improvement from when I had a single 970. Benchmarks show that SLI 970s are equal to a 980Ti. I know from experience that a single 970 has problems with some games on my 1440p monitor but when I run 2 in SLI I can get 60 FPS on all my games.

The problems with SLI tend to be greatly exaggerated on this forum. Everyone tells you how bad it is but my experience couldn't be more different. Maybe it was a nightmare on old 8800gt cards but it doesn't seem to be the case anymore.

Sure, if you are buying new today a 980Ti would be a better choice since it costs the same as 2 970s and it's a single card solution. If you already have a single 970 and want a performance boost, grabbing a second one for SLI is a much more cost effective solution than throwing down $650 on a 980Ti.

Most AAA games have SLI support on day one and the ones that don't, like Fallout 4, tend to get it within a week or two. Nvidia is very good about releasing Game Ready drivers for SLI support these days. Arkham Knight is the only major release from this year that I played that didn't get SLI support and we all know what a mess that game is on PC.

 

[king3pj]

I've got a P2 1200 in mine, 4790k(uses a good bit more power than your 4690k) and a pair of 780tis(they are both major hogs each has a pair of 8pin power connectors). I bet my system would run on your psu, as long as I didn't overclock.
Will it help? Only 1 way to really know, and that's to do it. Think about this, you might get 10-20% more clock speed and a boost to ram speed too. I think you'll do better than you think.

Thanks. I'll try it out tonight and see what happens.

 

[davmazin]

its nice if the sli support has grow when i had sli it was pretty sparse and most games didnt support it i am to the idea that sli was more for syntetic test if that has really changed its good to know but i am still a bit hesitant to get a sli setup especially since heat is a problem for me cuz i dont have air conditioner at home and my gtx 970 has reference nvidia heat which is very hot (80C or so)

 

[king3pj]

Well I messed with this overclocking project for a couple hours tonight and I was not able to come up with a stable GPU overclock. I ran Unigine Heaven at 2560x1440 with ultra settings, extreme tessellation, and 4X AA to do stress testing and benchmarking.

I followed this overclocking guide and I should note that I did not increase voltage because they didn't go into that in the guide. I did increase the power percentage to it's max of 110% though.

http://www.pcgamer.com/how-to-overclock-your-graphics-card/#page-2

I never saw any artifacting no matter how high I clocked my VRAM. I stopped at +400MHz because my temperatures were starting to rise.

I never saw artifacting when adjusting my core clock either but I found that anything over +100MHz would eventually crash my display during the stress testing. It would last long enough to finish a benchmark but within 5 minutes of it finishing the display would go black if I left Heaven running. The computer itself never shut down. I would just lose display.

At stock speeds my FPS was 75.8 and my score was 1909.

At +100MHz core and +400MHz VRAM my FPS was 83.0 and my score was 2090 but display crashed during stress testing after the benchmark.

At +89MHz core and +300MHz VRAM my FPS was 81.9 and my score was 1455.5 but display crashed during stress testing after benchmark again.

I realize that I could have kept decreasing my numbers but I didn't feel that any less FPS gain would have been worth the increased heat and fan noise.

I also realize that I might have been able to make this stable or even push the overclock higher if I started increasing the voltage but my top GPU was already reaching a max of 78C during stress testing without increasing the voltage. I assume that if I were to increase the voltage the card wouldn't be able to stay under the 80C limit I set without throttling the boost clock down.

 

[king3pj]

I messed with it a little more this morning before calling it quits for good. Without increasing voltage I was able to get my cards stable at +80 core clock and +375 memory clock. This increased my FPS in the Heaven Benchmark at 1440p by 7 FPS.

This is not as good as what other people with my exact card were able to achieve but anything higher than that was not stable in my system. Maybe it has something to do with me trying to get the overclock on 2 cards. Either way it doesn't appear to be a huge gain but it's still a free performance boost.

 

[king3pj]

Either this made a much bigger difference in gaming than it did in the Heaven Benchmark or Star Wars Battlefront has released a major performance patch since the last time I played.

At 2560x1440 with max settings I used to get 95-105FPS depending on which map and mode I was playing. I just played for roughly 2 hours and was in the 120-130 FPS range.

My boost turbo clock was at 1440MHz and memory clock is at 1940MHz. I'm highly skeptical that the relatively small overclock I did made a 25FPS difference.

I will test out Fallout 4 later today since I have been playing that game much more often recently. If I stop seeing dips into the low 40s with shadow distance turned back to maximum I will know this overclock made a difference.

 

[dudmont]

With high settings at high resolution, that memory bump, might be the cause for the frame rate bump. Evga cards are usually clocked fairly high to start with. The fact you didn't get that big of a gpu speed bump doesn't suprise me.

 

[king3pj]

With high settings at high resolution, that memory bump, might be the cause for the frame rate bump. Evga cards are usually clocked fairly high to start with. The fact you didn't get that big of a gpu speed bump doesn't suprise me.

Maybe. I didn't see much difference in Fallout 4. With shadow distance on maximum I still get framerate dips into the 40s while in the cities and a solid 60 fps everywhere else. From what I've read that seems to happen with any level of hardware though. I'm interested to see the results in other games.

 

[Jayhawker32]

I messed with it a little more this morning before calling it quits for good. Without increasing voltage I was able to get my cards stable at +80 core clock and +375 memory clock. This increased my FPS in the Heaven Benchmark at 1440p by 7 FPS.

This is not as good as what other people with my exact card were able to achieve but anything higher than that was not stable in my system. Maybe it has something to do with me trying to get the overclock on 2 cards. Either way it doesn't appear to be a huge gain but it's still a free performance boost.

Well before you give up I have a similar rig with an i7-4790k which I'm realizing now that if I want to OC I'm going to probably have to bump up to a 1000 W PSU. However, in testing I found that one of my cards boosted to 1404 stock and the other only to 1342 stock so I set in MSI Afterburner to allow different core and mem frequencies. This should help with the display driver crashing.

 

[king3pj]

I messed with it a little more this morning before calling it quits for good. Without increasing voltage I was able to get my cards stable at +80 core clock and +375 memory clock. This increased my FPS in the Heaven Benchmark at 1440p by 7 FPS.

This is not as good as what other people with my exact card were able to achieve but anything higher than that was not stable in my system. Maybe it has something to do with me trying to get the overclock on 2 cards. Either way it doesn't appear to be a huge gain but it's still a free performance boost.

Well before you give up I have a similar rig with an i7-4790k which I'm realizing now that if I want to OC I'm going to probably have to bump up to a 1000 W PSU. However, in testing I found that one of my cards boosted to 1404 stock and the other only to 1342 stock so I set in MSI Afterburner to allow different core and mem frequencies. This should help with the display driver crashing.

Yeah, when running Folding@Home on my 970s they work independently of each other instead of using SLI. Without my overclock applied I can see that my boost clock on one 970 is 1392MHz but the other is only 1366MHz. Even though they are supposed to be identical models one is clearly better than the other.

I don't think what you are talking about would work with SLI in a game though. When using SLI both cards work together and run at the clock speed of whichever card is slower. That's why when I check my clock speeds while gaming both run at 1366MHz boost speed.

 

[Jayhawker32]

I messed with it a little more this morning before calling it quits for good. Without increasing voltage I was able to get my cards stable at +80 core clock and +375 memory clock. This increased my FPS in the Heaven Benchmark at 1440p by 7 FPS.

This is not as good as what other people with my exact card were able to achieve but anything higher than that was not stable in my system. Maybe it has something to do with me trying to get the overclock on 2 cards. Either way it doesn't appear to be a huge gain but it's still a free performance boost.

Well before you give up I have a similar rig with an i7-4790k which I'm realizing now that if I want to OC I'm going to probably have to bump up to a 1000 W PSU. However, in testing I found that one of my cards boosted to 1404 stock and the other only to 1342 stock so I set in MSI Afterburner to allow different core and mem frequencies. This should help with the display driver crashing.

Yeah, when running Folding@Home on my 970s they work independently of each other instead of using SLI. Without my overclock applied I can see that my boost clock on one 970 is 1392MHz but the other is only 1366MHz. Even though they are supposed to be identical models one is clearly better than the other.

I don't think what you are talking about would work with SLI in a game though. When using SLI both cards work together and run at the clock speed of whichever card is slower. That's why when I check my clock speeds while gaming both run at 1366MHz boost speed.

You can and it does work. I have to do it because one of my GPU's boost to 1405 stock while the other only to 1342 stock both the same model.

These images aren't at load so the cards have downclocked since they aren't being used.
It works because you can overclock one card to be at the same core clock as the second one without overclocking it. It will run at the speed of the lowest yes, but if you can increase the speed of the lowest one then they will run at the same speed.

UPDATE:
Here they are with a graphical load.

 

[king3pj]

I guess I just don't see the point of overclocking my better card past the point where the weaker one is stable. I overclocked both of them a bit but the max stable core clock I could get was +80MHz.

If both cards are going to run at the clock speed of the slower card while gaming what would I gain by independently clocking the stronger card higher?

 

[king3pj]

I think I understand what you are talking about now. You are saying I should overclock my slower card and not my faster one.

I can see how that would make sense. However, with my +80MHz overclock I'm able to run on both cards they both boost to around 1440MHz. That is higher than the 1392MHz boost clock I get on my faster card without overclocking.

I think if I were to only overclock my slower card they would both be limited to 1392MHz boost because that is what my faster card gets at stock speeds.

Then my faster card would become the limiting factor and I would be running almost 50MHz slower than I am able to achieve at boost speeds (1440MHz) when I overclock both cards to the same stable, +80MHz.

I'm still not sure I see the point in that. I haven't been running any overclock because I didn't see enough improvement in most of my games to justify the extra heat. If I wasn't able to hold 60FPS at 1440p I would run my overclock for specific games but that hasn't come up yet.

I suppose if I were to just overclock the slower card so both run at the 1392MHz boost speed my faster card is able to maintain at stock speeds I would see a 26MHz increase in SLI speed without much heat increase.

 

[Jayhawker32]

I think I understand what you are talking about now. You are saying I should overclock my slower card and not my faster one.

I can see how that would make sense. However, with my +80MHz overclock I'm able to run on both cards they both boost to around 1440MHz. That is higher than the 1392MHz boost clock I get on my faster card without overclocking.

I think if I were to only overclock my slower card they would both be limited to 1392MHz boost because that is what my faster card gets at stock speeds.

Then my faster card would become the limiting factor and I would be running almost 50MHz slower than I am able to achieve at boost speeds (1440MHz) when I overclock both cards to the same stable, +80MHz.

I'm still not sure I see the point in that. I haven't been running any overclock because I didn't see enough improvement in most of my games to justify the extra heat. If I wasn't able to hold 60FPS at 1440p I would run my overclock for specific games but that hasn't come up yet.

I suppose if I were to just overclock the slower card so both run at the 1392MHz boost speed my faster card is able to maintain at stock speeds I would see a 26MHz increase in SLI speed without much heat increase.

Basically what I did was overclock each differently so that I achieve the same core clock without it trying to set one higher than the other. If I didn't overclock each card separately my system was unstable and the driver would often crash.