Update6: Never trust Maxwell to be stable. Ever. Nvidia sucks. (newest post)

[Techhog]

I used the Tomb Raider benchmark to test my 970 OC. Very noticeable artifacts when unstable.

I'm gonna use BF4 first (since I haven't played that and thus will have more motivation), then maybe spend more time in Tomb Raider. I've had it crash once before, but I couldn't replicate that crash. For some reason, changing settings in-game causes the driver to crash the first time I do it in some games, but then never again. I'm half-wondering if the real issue there is some kind of bug related to switching from AMD to Nvidia without reinstalling Windows. I'm only seeing the issue in games that I played before.

 

[ocre]

So your sticking with 1540mhz with the voltage on max?

A lot of people use modded bios to get the 1550-1600mhz overclocks.

 

[Techhog]

So your sticking with 1540mhz with the voltage on max?

A lot of people use modded bios to get the 1550-1600mhz overclocks.

I turned the voltage down, actually. I think that the crashes were actually caused by suddenly hitting the power limit. It's at +40mV right now, and that's what it was at when it ran Heaven for over an hour and a half at... I think it was 1539MHz. That might have been 1554MHz. It doesn't really make much of a difference of 15MHz is negligible in actual games. Speaking of actual games, it's time for the BF4 stress test. I'll come back with results. It's still pretty difficult to figure this out when the main sign of instability I'm seeing fixes itself after having the error once. I just don't understand how something like that works. How am I supposed to diagnose a problem that I can't replicate?

 

[flexy]

I wouldn't shoot that high with your OC at first. Seriously, show me the person hitting 1550-1600 (1600, uhm...no!) without limiting and running Heaven on Extreme. It's not that long ago when it made headlines that some people reached 1500, and TBH I haven't seen ANYONE running those cards at 1600 yet.

By the way, nice case...have the same..I love it!

(And my EVGA SC ACX2.0 is fricking going back via RMA, it cannot even hold it's STOCK boost without OC and 110% power target. Not in Heaven and not in BF4. Wish me luck they send me a ACX2.0 SC+ in exchange, I begged them

 

[Techhog]

So, at +40mV, I actually hit the voltage limit and throttle down to 1524MHz. That's... interesting. I'm going to try +50mV and see what happens. BF4 did crash once when trying to open it, since I opened it, remembered that I forget to set the clocks, closed it, set the clocks, and reopened it all within the span of like a minute. I'll see if that happens again before I decide if that was in instability crash, or if it was just from the sudden change in specs and the game didn't even have time to close properly.

I wouldn't shoot that high with your OC at first. Seriously, show me the person hitting 1550-1600 (1600, uhm...no!) without limiting and running Heaven on Extreme. It's not that long ago when it made headlines that some people reached 1500, and TBH I haven't seen ANYONE running those cards at 1600 yet.

By the way, nice case...have the same..I love it!

(And my EVGA SC ACX2.0 is fricking going back via RMA, it cannot even hold it's STOCK boost without OC and 110% power target. Not in Heaven and not in BF4. Wish me luck they send me a ACX2.0 SC+ in exchange, I begged them

I actually read a review where someone hit 1609MHz. Clearly the most golden of golden chips. Also, Heaven on Extreme runs just fine at 1554MHz.

 

[Techhog]

Hm... BF4 just seems to really like 1524MHz. Well, not a big deal. I can see why overclockers don't like GPU boost 2.0, though. I'm sure that adding more voltage would keep it at 1538, but I'm not sure if it's worth it.

 

[ocre]

That's the thing about maxwell and boost. It is nothing like overclocking gpus in the past.

Used to, you push up the clocks until it is unstable then add a little voltage to stabilize. Then repeat until the temps are too high or you just can't get it stable.

But with the gm204, the power limit is right in the center of all this. It is very sophisticated.

See the target TDP will interfere and limit your boost. Actually, boost is designed completely around the target TDP and target temps.
110% is not that much. So raising the voltage can actually cause your boost to suffer, especially if your already close to the power limit already. Some games stress the GPU harder than others, that is the whole reason for this profound change in the way clock speed is handled in gpus. If you are running a game that is demanding, that pushes the TDP and causes you to throttle back or effects your stability, adding more voltage will hurt more than help. Higher voltages mean higher consumption.

So its tricky to inch further.
The power target will throttle back in fractions of a second. Keep this in mind when overclocking. This is why you shouldn't be quick to add voltages.
Yet in the case where your not on the edge the power limit. Where your overclock is limited by voltage, it may help.

This is why people are running modded bios on maxwell. You can have more control and gain higher clocks without worrying about premature power limiting.

Without a bios mod, 1550ish is a decent overclock for the 970. Some chips are more power hungry and they reach their power limit earlier than that. See, the 970 has a very low window for pushing up power limit. The 980 can go up to 125%. So the 970 is limited much more. Just keep this in mind.

MSI already uses custom bios settings and allow more power draw than most anyone else. Your clocks are decent enough and you wouldn't gain all that much with a fully custom bios. 50mhz and your running about 1600mhz. It may not be worth the trouble or risk to you. Just saying.

Btw,
Your random crash starting games is very familiar to me. When I have MSI open, I get crashes like that. It might be the overlay or the 64bit games. It just acts funny with MSI opened. If I start AC unity and tab out then open afterburner and tab back in, the game will crash. Also sometimes I can start bf4 with afterburner open and the game will start but no overlay and crashes soon afterwards. Before I get to play or get to title screen. Crazy thing is, I can restart the game right afterwards and everything is fine. But I don't keep tabbing out and back in the game with MSI opened because I am afraid it will crash.

Its been like that for me for awhile. If MSI is not open, this doesn't happen.
Its not every time but I can say for sure 100%, MSI open does cause me crashes in 64bit games. I suspect it is the overlay and 64 bit apps. Its not been a major issue, but I know it gives me crashes

 

[flexy]

I can tell you that my incredibly poor EVGA SC ACX2.0 has a default P/L of 170W in the BIOS, with 110% it is 187W.

I did a BIOS mod (NOT that easy since it's not as simple as editing the power limit, you also need to overspec what the individual rails will draw), and opened up P/L to 196W default and 206W at 105%. And even with 206W it would still (although rarely) hit the PL in some stress tests, mainly Firestrike Demo. But at least it stopped limiting in BF4 or Heaven.

And by the way, lots of cards don't even allow going higher than 1.212, they're volt locked.

 

[Techhog]

That's the thing about maxwell and boost. It is nothing like overclocking gpus in the past.

Used to, you push up the clocks until it is unstable then add a little voltage to stabilize. Then repeat until the temps are too high or you just can't get it stable.

But with the gm204, the power limit is right in the center of all this. It is very sophisticated.

See the target TDP will interfere and limit your boost. Actually, boost is designed completely around the target TDP and target temps.
110% is not that much. So raising the voltage can actually cause your boost to suffer, especially if your already close to the power limit already. Some games stress the GPU harder than others, that is the whole reason for this profound change in the way clock speed is handled in gpus. If you are running a game that is demanding, that pushes the TDP and causes you to throttle back or effects your stability, adding more voltage will hurt more than help. Higher voltages mean higher consumption.

So its tricky to inch further.
The power target will throttle back in fractions of a second. Keep this in mind when overclocking. This is why you shouldn't be quick to add voltages.
Yet in the case where your not on the edge the power limit. Where your overclock is limited by voltage, it may help.

This is why people are running modded bios on maxwell. You can have more control and gain higher clocks without worrying about premature power limiting.

Without a bios mod, 1550ish is a decent overclock for the 970. Some chips are more power hungry and they reach their power limit earlier than that. See, the 970 has a very low window for pushing up power limit. The 980 can go up to 125%. So the 970 is limited much more. Just keep this in mind.

MSI already uses custom bios settings and allow more power draw than most anyone else. Your clocks are decent enough and you wouldn't gain all that much with a fully custom bios. 50mhz and your running about 1600mhz. It may not be worth the trouble or risk to you. Just saying.

Yeah, all of that is pretty much what I've been able to surmise. I have no interest in using a custom BIOS, though. That's really not worth it, especially since I can't exactly afford to replace the card if something goes horribly wrong. As far as I can tell, my power usage in BF4 is only reaching ~100%, so there is a little bit of headroom to increase voltage. Increasing it to max was too much though, and just caused instability (I'd assume that it crashed instead of throttling due to how rapidly it hit the power limit.)

Btw,
Your random crash starting games is very familiar to me. When I have MSI open, I get crashes like that. It might be the overlay or the 64bit games. It just acts funny with MSI opened. If I start AC unity and tab out then open afterburner and tab back in, the game will crash. Also sometimes I can start bf4 with afterburner open and the game will start but no overlay and crashes soon afterwards. Before I get to play or get to title screen. Crazy thing is, I can restart the game right afterwards and everything is fine. But I don't keep tabbing out and back in the game with MSI opened because I am afraid it will crash.

Its been like that for me for awhile. If MSI is not open, this doesn't happen.
Its not every time but I can say for sure 100%, MSI open does cause me crashes in 64bit games. I suspect it is the overlay and 64 bit apps. Its not been a major issue, but I know it gives me crashes

That explains the crash in BF4, but not the crash when changing settings in certain games for the first time. That's just really weird, and I have no idea how to determine if that's a stability issue or a software because there isn't any consistency to it other than, possibly, it affecting games that I played previously which had vendor-specific features.

 

[Techhog]

I can tell you that my incredibly poor EVGA SC ACX2.0 has a default P/L of 170W in the BIOS, with 110% it is 187W.

I did a BIOS mod (NOT that easy since it's not as simple as editing the power limit, you also need to overspec what the individual rails will draw), and opened up P/L to 196W default and 206W at 105%. And even with 206W it would still (although rarely) hit the PL in some stress tests, mainly Firestrike Demo. But at least it stopped limiting in BF4 or Heaven.

And by the way, lots of cards don't even allow going higher than 1.212, they're volt locked.

It's really sad how gimped these cards are. I guess Nvidia really doesn't want people to be able to reach GTX 980 levels of performance, and they pulled out all of the stops to make it as hard as possible.

 

[flexy]

It's really sad how gimped these cards are. I guess Nvidia really doesn't want people to be able to reach GTX 980 levels of performance, and they pulled out all of the stops to make it as hard as possible.

This might *possibly* play a role too, but with some cards the components are not exactly "military quality" and they are not exactly made for overclocking records either. There are certainly economical reasons too, manufacturers try to save money where they can.

A prime example would be that hideous heatsink on the SC EVGA which they clearly re-used from their 770 which they simply slapped on the 970. On heatpipe is entirely useless and doesn't even touch the GPU since the GTX970 chip is actually smaller. Things like that.

If you have a GTX970 which goes beyond 1500 AND where you can actually supply more voltage I am very jealous, sounds like a overclocker's dream. I hope they send me a SC+ as a replacement and I really hope the EVGA SC+ are better than their SC series. Overall I am really disappointed, not only because it overclocks like shit (obviously it doesn't, it cannot even hold stock boost clocks ...also the heat, fan noise etc. are LESS than overwhelming. At 40% which I need to keep it near 70C it's almost as loud as an old GTX 275 which was a hellish loud card with a blower And then their crappy 4 power phases where any other card has 6 and of course it's volt-locked too, it uses an analog VRM where most others use digital ones... and then of course that heatsink. Not exactly things that I'd expect from a reputable company...just saying. So, consider yourself lucky

 

[Techhog]

This might *possibly* play a role too, but with some cards the components are not exactly "military quality" and they are not exactly made for overclocking records either. There are certainly economical reasons too, manufacturers try to save money where they can.

A prime example would be that hideous heatsink on the SC EVGA which they clearly re-used from their 770 which they simply slapped on the 970. On heatpipe is entirely useless and doesn't even touch the GPU since the GTX970 chip is actually smaller. Things like that.

If you have a GTX970 which goes beyond 1500 AND where you can actually supply more voltage I am very jealous, sounds like a overclocker's dream. I hope they send me a SC+ as a replacement and I really hope the EVGA SC+ are better than their SC series. Overall I am really disappointed, not only because it overclocks like shit (obviously it doesn't, it cannot even hold stock boost clocks ...also the heat, fan noise etc. are LESS than overwhelming. At 40% which I need to keep it near 70C it's almost as loud as an old GTX 275 which was a hellish loud card with a blower And then their crappy 4 power phases where any other card has 6 and of course it's volt-locked too, it uses an analog VRM where most others use digital ones... and then of course that heatsink. Not exactly things that I'd expect from a reputable company...just saying. So, consider yourself lucky

You know, I always thought that "military class" thing was just marketing that doesn't really affect anything. Does it actually have some sort of effect?

And yeah, I heard that fiasco with the original SC cooler. No idea what EVGA was thinking. They lost a purchase from me due to not being up to par with these cards, since before this I always said that I would go to EVGA if I ever bought an Nvidia card. I hope Sapphire doesn't screw the pooch when I want to buy another AMD card.

And yeah, I'm not really complaining. I expected to only get to 1400-1450MHz based on my usual luck in life. Granted, it was already doing 1316MHz out of the box...

 

[Techhog]

So I finally started hitting the power limit for short bursts in the Witcher 2 at +70mV and +204 core offset. Even then, it was only barely hitting the limit. This card is a beast. I'm gonna see if I can squeeze +212 out of it (which lets me hit 1554MHz) and then call it a day.

EDIT: By the way, in The Witcher 2, Geralt's hair had some minor flickering spots from the side at times. I couldn't tell if it was artifacting or just clipping between layers of hair. I think it was clipping because it was only from the sides, it went away during some parts, and I didn't see it happening anywhere else.

EDIT2: I think @ocre might be correct about MSI Afterburner causing crashes, because 3DMark 11 and 3DMark 13 crash any time I try to open it right after changing settings, even if I change to stock.

EDIT3: Uh... I think I broke the card. ._.

EDIT4: TIL don't change OC settings with the 3DMark 11 window open, or else your card will be unstable until you restart the computer

Thank you for riding the rollercoaster that was this post. I'm sticking with +70mV and +204 core offset. That never goes below 1520MHz, so I'd say that's good.

 

[flexy]

What sort of ***es me off is that on another forum, people sort-of looked at me because I want to RMA my card. Yes, true I don't see any crashes (yet) at 100% "power target", but when I go up to 110% the card cannot hold IT'S STOCK CLOCKS, in my case 1380 max. boost. It crashes out my entire PC. Since 110% power target is within specs and WAY within tolerance I do consider this a hardware defect. Not anyone agreed with me and of course not the one who sold me the card. Needless to say there is not much "overclocking" to do if you cannot even up the power target and the card can't even hold 1367/1380 stable.

 

[Techhog]

What sort of ***es me off is that on another forum, people sort-of looked at me because I want to RMA my card. Yes, true I don't see any crashes (yet) at 100% "power target", but when I go up to 110% the card cannot hold IT'S STOCK CLOCKS, in my case 1380 max. boost. It crashes out my entire PC. Since 110% power target is within specs and WAY within tolerance I do consider this a hardware defect. Not anyone agreed with me and of course not the one who sold me the card. Needless to say there is not much "overclocking" to do if you cannot even up the power target and the card can't even hold 1367/1380 stable.

Wait, I'm confused. Was that before or after the BIOS mod?

 

[Techhog]

Uh, looking at some other results, my Unigine scores seem pretty low for a GTX 970 unless memory bandwith has a huge effect there. Other benchmarks are fine.

 

[Techhog]

3DMark 11 is terrible. Either it made my overclock permanently unstable, or it's an unstable program! This stinks!

Well, I guess not permanently, but the driver stays unstable until I restart the computer. Same goes for Vantage. It's really weird.

 

[flexy]

Wait, I'm confused. Was that before or after the BIOS mod?

This is on stock bios with PT to 110%.

 

[flexy]

3DMark 11 is terrible. Either it made my overclock permanently unstable, or it's an unstable program! This stinks!

Well, I guess not permanently, but the driver stays unstable until I restart the computer. Same goes for Vantage. It's really weird.

Uhm...3dm11 works here, what happens at lower clocks? I am still saying you're aiming too high and should gradually get up with your clocks.

 

[Techhog]

Uhm...3dm11 works here, what happens at lower clocks? I am still saying you're aiming too high and should gradually get up with your clocks.

It crashes every time I try to run it, even at stock, and then the driver becomes completely unstable. I just uninstalled it because it seems more like a bug with the program or the system detection. Also, I did go up gradually at the start. If you really think 3DMark 11 matters that much I'll try again, but if I'm stuck at stock because of an outdated benchmark I'm going to be pissed to the point that I'll consider selling the card and getting a console. I really don't feel like starting from scratch again.

I'm going to try it, but I'm almost willing to bet money that any overclock will cause it to crash due to a glitch in system detection.

EDIT: Found the problem. For wherever reason, the first graphics test in 3DMark 11 use WAY more power than any other test. I don't understand why that's the case at all, but I doubt that I could hit even 1450MHz stable in 3DMark 11. I don't know what to do from here...

 

[Techhog]

Yeah, I've confirmed it now. 3DMark 11 (and Vantage, apparently) push voltage harder than anything else does. Why? That's a good question. It's really puzzling. Even Furmark was perfectly stable at max voltage with a +204 core offset. I guess this means that, at the very least, what I've found so far isn't a "24/7" overclock, but I never intended to use it as one in the first place. (I don't like using overclocked settings at times when it's not necessary unless I'm testing stability.) I guess I should see how far it can go. It's good for seeing what my max clock without increasing voltage is, at least.

EDIT: Max clock I can hit in 3DMark 11 is +179MHz offset and +20mV, and I'm hitting the voltage limit at that point. So... what do I do? Every game (in addition to Heaven, Valley, all Fire Strike tests, and even Furmark) so far has been stable at +204MHz and +70MHz. That's a pretty huge difference. Hm... I'm going to keep a close eye on how much power games are using I think.

 

[Techhog]

Okay, I got it now. My standard OC will be the one that 3DMark 11 can handle. If I play a game and it isn't pushing voltage super hard like 3DMark 11 does, I'll switch to the +204MHz profile. That seems like the most logical way to handle it.

 

[Techhog]

*sigh* After thinking about it, I'll take the safe route. My chip can only take 1508MHz at 100% stability. That's a huge disappointment. It can take more than that in most cases, but the problem is that going higher than +20mV means that it's possible for some applications to push voltage super high in a burst, which results in a crash. I don't even understand how 3DMark 11 of all things does this, yet Heaven, Valley, Furmark, and OCCT don't. It makes no sense whatsover, but it does explain the crash when changing settings for the first time... kinda... Not really... Well, either way, I lost 50MHz due to one program which is just a dumb benchmark. I think this will be the last time I ever overclock. It's fun, but really stupid, and I keep losing the silicon lottery. This is why i wanted to spend more on the G1, so I'd have a better chance at a good overclock.

 

[RussianSensation]

*sigh* After thinking about it, I'll take the safe route. My chip can only take 1508MHz at 100% stability. That's a huge disappointment. It's fun, but really stupid, and I keep losing the silicon lottery.

A 1550mhz overclock is only 2.7% higher than what you have. Moving from an i5 to an i7 4790K @ 4.7Ghz will net you more in games. Many 970 cards can't even get to 1.45Ghz.

Your expectations seem a bit unrealistic if you think you should be getting 1560-1600mhz with a non-EVGA Classified/Asus Matrix/MSI Lightning style card. The best overclock Computerbase got was 1528-1566mhz on a Galax GTX970 with a 125% power target, and that's from a whole stack of 970 cards from various manufacturers.

Also, you have to prioritize what you want and find a balance. IMO, the MSI Gaming 970 offers the best balance of performance, overclocking, price and noise levels of any 970 card in the US. You zoomed in so much on that last 40-50mhz clocks but forgot about noise levels.

Idle
MSI Gaming 970 = 27.5 dBA
G1 Windforce 970 = 31 dBA

Load
MSI Gaming 970 = 36.5-37 dBA
G1 Windforce = 38.5-42 dBA

Their GPU temperatures are only 1*C apart.
http://www.computerbase.de/2014-10/nvidia-geforce-gtx-970-test-roundup-vergleich/4/

Unless you place no value on the noise levels, and the MSI card turning off the fans which preserves the fan bearings, I think those 2 factors matter way more than a 50mhz extra overclock over 1.5Ghz over the 2-3 years of GPU card ownership. In the context of base clocks on a reference 970, 1.508Ghz with quiet noise levels is an excellent result. I am very surprised you are not thrilled with the card. Instead of comparing it to a max overclocked 980 that costs $515+, you should be comparing what you got from your old 7950. You are basically getting a 40-50% increase in performance, and even more at 1.5Ghz for $320 or so (- the TW3 value coupon). I wouldn't worry about the last 2-3% of overclocking performance because if you really need extra performance, you can always sell this card for $200 in 12 months and get something faster.

 

[Techhog]

A 1550mhz overclock is only 2.7% higher than what you have. Moving from an i5 to an i7 4790K @ 4.7Ghz will net you more in games. Many 970 cards can't even get to 1.45Ghz.

Your expectations seem a bit unrealistic if you think you should be getting 1560-1600mhz with a non-EVGA Classified/Asus Matrix/MSI Lightning style card. The best overclock Computerbase got was 1528-1566mhz on a Galax GTX970 with a 125% power target, and that's from a whole stack of 970 cards from various manufacturers.

Also, you have to prioritize what you want and find a balance. IMO, the MSI Gaming 970 offers the best balance of performance, overclocking, price and noise levels of any 970 card in the US. You zoomed in so much on that last 40-50mhz clocks but forgot about noise levels.

Idle
MSI Gaming 970 = 27.5 dBA
G1 Windforce 970 = 31 dBA

Load
MSI Gaming 970 = 36.5-37 dBA
G1 Windforce = 38.5-42 dBA

Their GPU temperatures are only 1*C apart.
http://www.computerbase.de/2014-10/nvidia-geforce-gtx-970-test-roundup-vergleich/4/

Unless you place no value on the noise levels, and the MSI card turning off the fans which preserves the fan bearings, I think those 2 factors matter way more than a 50mhz extra overclock over 1.5Ghz over the 2-3 years of GPU card ownership. In the context of base clocks on a reference 970, 1.508Ghz with quiet noise levels is an excellent result. I am very surprised you are not thrilled with the card. Instead of comparing it to a max overclocked 980 that costs $515+, you should be comparing what you got from your old 7950. You are basically getting a 40-50% increase in performance, and even more at 1.5Ghz for $320 or so (- the TW3 value coupon). I wouldn't worry about the last 2-3% of overclocking performance because if you really need extra performance, you can always sell this card for $200 in 12 months and get something faster.

I've since tested it again. The max stability for 3DMark 11 drops every time I try to test it (it's now at <1475MHz), and, frankly, now that I know what the problem is, I know that any added voltage automatically makes the card unstable. I've had it crash at just +10mV, and only in 3DMark 11. I wouldn't be surprised if even factory overclocks could be unstable because of this. Nvidia pulled a dirty trick to lower power consumption: the minimum voltage on Maxwell cards is ridiculously low. This means that any program which pushes the voltage rapidly has a chance of crashing the card. If the card had an option to increase the minimum voltage, everyone would have dramatically higher stability. That's why your point is pretty useless. You need a golden chip with top-class VRMs and caps to be able to add any voltage without a BIOS mod. The added power target doesn't do anything except for allowing a higher max boost; it doesn't affect the offset. As it is now, the voltage slider is 100% useless because an overvolted 970 cannot be stable, no matter how much it seems to be. I want to sell the card now and buy a PS4, honestly.

For now, I have no choice but to keep it at stock. Every game/benchmark affects voltage differently, so there's no way to know that it's stable until I've tested every game in existence. If I knew that something this stupid would happen, i wouldn't have bought this PoS. Never touching Nvidia again. I should have bought the 290.