ASUS GTX 670 Won't Stay at Core OC Clock Speeds

[PersonWithTech]

I've been trying to overclock my (already) factory-overclocked ASUS GTX 670 today using Evga Precision X. I've been following this guide on overclock.net, and keep having an issue where the core clocks would return to the ASUS stock speeds, while running Unigine Heaven. While running the benchmark, the screen would temporarily turn black with the monitor's icon for 'input connected' present in the top left corner. The benchmark would continue at the a lower max frequency.

Checking GPU-Z just shows that the boost clock is back to default. Precision X has no evidence of change, however changing the slider and hitting 'Apply' will work. The problem occurs while running Heaven for longer than a single benchmark, at 1140 MHz (offset +81MHz). The same problem occurs around halfway through the benchmark when at 1160MHz (offset +101MHz). The power target is at 117% and the core voltage is set to 1150mV. USing the fan curve found at the top of guide, temps have never gone over 60 C (140 F), and hang more around 50 C (122 F).

I am yet to fully understand the effects of overclocking and behavior of graphics cards. Is this instability? Throttling? Is it caused by the power and voltage? What is it? What causes it? And, how can I fix or prevent it?

Can someone please provide some insight into this.

 

[Ed_1]

I've been trying to overclock my (already) factory-overclocked ASUS GTX 670 today using Evga Precision X. I've been following this guide on overclock.net, and keep having an issue where the core clocks would return to the ASUS stock speeds, while running Unigine Heaven. While running the benchmark, the screen would temporarily turn black with the monitor's icon for 'input connected' present in the top left corner. The benchmark would continue at the a lower max frequency.

Checking GPU-Z just shows that the boost clock is back to default. Precision X has no evidence of change, however changing the slider and hitting 'Apply' will work. The problem occurs while running Heaven for longer than a single benchmark, at 1140 MHz (offset +81MHz). The same problem occurs around halfway through the benchmark when at 1160MHz (offset +101MHz). The power target is at 117% and the core voltage is set to 1150mV. USing the fan curve found at the top of guide, temps have never gone over 60 C (140 F), and hang more around 50 C (122 F).

I am yet to fully understand the effects of overclocking and behavior of graphics cards. Is this instability? Throttling? Is it caused by the power and voltage? What is it? What causes it? And, how can I fix or prevent it?

Can someone please provide some insight into this.

Ok, I don't use precision X but what you describe sound like the driver reset , you normally get message down by tray when this happens .
If this happens the driver resets , so you lose any OC from software .
Your OC is probably to high for voltage .
You say voltage is set to 1150mv and is that what you see in GPU-Z sensor tab while running Heven while it is OC . It should be 1.175v or at least 1.165v . It should do that stock if load is on it , like in Heven/Valley .
Also the 1140mhz is that base clock or boost clock shown in GPU=Z and in sensor tab clocks ?
either way that would boost to over 1200, probably around 1240mhz and no way your going to do that at 1150mv .

So here is what I would do , since i don't know stock baseline yet .
try running a OC that gives 1202 mhz in GPU-Z sensor tab during a run , if it passes a few loops ok , then raise OC in 13mhz steps and then retrying test .

You should see 1.165-1.175v during test in boost stages .

 

[PersonWithTech]

Ok, I don't use precision X but what you describe sound like the driver reset , you normally get message down by tray when this happens .
If this happens the driver resets , so you lose any OC from software .
Your OC is probably to high for voltage .
You say voltage is set to 1150mv and is that what you see in GPU-Z sensor tab while running Heven while it is OC . It should be 1.175v or at least 1.165v . It should do that stock if load is on it , like in Heven/Valley .
Also the 1140mhz is that base clock or boost clock shown in GPU=Z and in sensor tab clocks ?
either way that would boost to over 1200, probably around 1240mhz and no way your going to do that at 1150mv .

So here is what I would do , since i don't know stock baseline yet .
try running a OC that gives 1202 mhz in GPU-Z sensor tab during a run , if it passes a few loops ok , then raise OC in 13mhz steps and then retrying test .

You should see 1.165-1.175v during test in boost stages .

Hi. Thanks for the reply. A driver reset sounds like the issue. I will respond to your discovery of missing details. The OC frequencies I was giving was the boost clock which is set in OC utilities. I gave the offset next to them to signal that the default factory-OC boost clock was 1059. 1150mV is the max that can be set in the Voltage Tuner, however this ramps up 1175mV during benchmarks.

SO here I am, running Heaven, watching the middle scenes of the benchmark (if it is set to 1160 MHz boost), and looking at my current/max frequency of over 1250 MHz (if I recall correctly), then suddenly the screen resets connection and continues with the default boost of 1059, and at around 1163 current frequency. I check back, and notice my low temps of under 60 C , and the max voltage of 1175mV.

What I have been doing is what the guide I linked in the 1st para of my original post has instructed, until Step 2. This is where it tells me to keep OC-ing by 20MHz until Frequency Oscillation or instability occurs. Throttling is also mentioned, and I am not sure if it is related. Now the driver reset does sound like the specific issue, but does it fall under instability and should I use the instability instructions, or can I get some alternative solutions.

I appreciate your solution, however you were lacking details from me to fully understand the problem, and it really is a more gradual version of the overclocking I had done from the guide. If all else fails, however, 13MHz does seem like the optimal increase and running it multiple times does add a useful certainty. So, as a possible solution, I should restart OCing at your specification, as I am yet to discover if the reset occurs in smaller frequencies at all.

 

[newtekie1]

You are definitely getting instability and a driver crash. When the driver recovers it resets the clocks on the card to the defaults. My GTX670s start to go unstable at 1050MHz base clock, I'm not sure what that equates to boost wise. I'm pretty sure the offset effects base clock, and the boost clock will vary based on power limit, heat, and load. So the number you should be watching is your base clock.

 

[Ed_1]

Hi. Thanks for the reply. A driver reset sounds like the issue. I will respond to your discovery of missing details. The OC frequencies I was giving was the boost clock which is set in OC utilities. I gave the offset next to them to signal that the default factory-OC boost clock was 1059. 1150mV is the max that can be set in the Voltage Tuner, however this ramps up 1175mV during benchmarks.

SO here I am, running Heaven, watching the middle scenes of the benchmark (if it is set to 1160 MHz boost), and looking at my current/max frequency of over 1250 MHz (if I recall correctly), then suddenly the screen resets connection and continues with the default boost of 1059, and at around 1163 current frequency. I check back, and notice my low temps of under 60 C , and the max voltage of 1175mV.

What I have been doing is what the guide I linked in the 1st para of my original post has instructed, until Step 2. This is where it tells me to keep OC-ing by 20MHz until Frequency Oscillation or instability occurs. Throttling is also mentioned, and I am not sure if it is related. Now the driver reset does sound like the specific issue, but does it fall under instability and should I use the instability instructions, or can I get some alternative solutions.

I appreciate your solution, however you were lacking details from me to fully understand the problem, and it really is a more gradual version of the overclocking I had done from the guide. If all else fails, however, 13MHz does seem like the optimal increase and running it multiple times does add a useful certainty. So, as a possible solution, I should restart OCing at your specification, as I am yet to discover if the reset occurs in smaller frequencies at all.

Ok, well 1254 is just not going to happen with 1.175 on latest drivers .
My guess would be somewhere in range of 1202-1241 but you need to test each step as i mentioned .
Not sure on your model but i know many Asus pre OC 6xx series have very little head room to OC as there already pushing things .

Whats the stock boost value during testing load ?
If its near 1200 then that is high already .
try1098 or 1111 boost , that should give 1215 , see if that works .

Oh, that guide is good, good info but thing is as Nvidia updated there drivers starting with 314.xx the clock head room went down , so now they don't clock as fast w/o reseting but you get new optimizations on the games to counter .

 

[Ed_1]

You are definitely getting instability and a driver crash. When the driver recovers it resets the clocks on the card to the defaults. My GTX670s start to go unstable at 1050MHz base clock, I'm not sure what that equates to boost wise. I'm pretty sure the offset effects base clock, and the boost clock will vary based on power limit, heat, and load. So the number you should be watching is your base clock.

Yes, OC affects both, base and boost values .

 

[PersonWithTech]

You are definitely getting instability and a driver crash. When the driver recovers it resets the clocks on the card to the defaults. My GTX670s start to go unstable at 1050MHz base clock, I'm not sure what that equates to boost wise. I'm pretty sure the offset effects base clock, and the boost clock will vary based on power limit, heat, and load. So the number you should be watching is your base clock.

Ok, well 1254 is just not going to happen with 1.175 on latest drivers .
My guess would be somewhere in range of 1202-1241 but you need to test each step as i mentioned .
Not sure on your model but i know many Asus pre OC 6xx series have very little head room to OC as there already pushing things .

Whats the stock boost value during testing load ?
If its near 1200 then that is high already .
try1098 or 1111 boost , that should give 1215 , see if that works .

Oh, that guide is good, good info but thing is as Nvidia updated there drivers starting with 314.xx the clock head room went down , so now they don't clock as fast w/o reseting but you get new optimizations on the games to counter .

Yes, OC affects both, base and boost values .

Thanks for the replies. I experimented with GPU-Z recently and it proved that the base value was carried with the boost value at a fixed difference (in my case 79MHz). I believe the boost clock is the clock that is changed and usually displayed in the OC utility (unlike the base clock). Now that I know this is a form of instability, should I go to the Step 3 instability fix in the guide, even though the driver crash wasn't listed?

In response to Ed_1, my graphics card is the OC edition, not the standard or TOP edition, you can see the model number in my system spec. The only difference between TOP and OC is that the clocks are lower on the OC, and the OC was supposedly targeted towards the Asian market. The OC edition was also a response to the TOP editions having quickly sold out. I do believe that the OC edition does go through the same factory overclocking tests as the TOP does, so I do believe most OC's can actually achieve the same clocks as the TOP edtion.

 

[Ralfies]

Thanks for the replies. I experimented with GPU-Z recently and it proved that the base value was carried with the boost value at a fixed difference (in my case 79MHz). I believe the boost clock is the clock that is changed and usually displayed in the OC utility (unlike the base clock). Now that I know this is a form of instability, should I go to the Step 3 instability fix in the guide, even though the driver crash wasn't listed?

In response to Ed_1, my graphics card is the OC edition, not the standard or TOP edition, you can see the model number in my system spec. The only difference between TOP and OC is that the clocks are lower on the OC, and the OC was supposedly targeted towards the Asian market. The OC edition was also a response to the TOP editions having quickly sold out. I do believe that the OC edition does go through the same factory overclocking tests as the TOP does, so I do believe most OC's can actually achieve the same clocks as the TOP edtion.

I think you're expecting a little too much. ~1200Mhz is a pretty decent overclock for a 670. I tried out two different 670 DCII Tops. The first one maxed out around 1180Mhz and the second one(which is my current card) can reach 1215Mhz. You just need to lower your overclock a bit until you can pass a few rounds of heaven.

 

[newtekie1]

The TOP cards are cherry picked cards. The testing that both go through at the factory is the binning process, if a card clocks very well it is made a TOP card, if it clocks good it is made an OC card, and if it clocks poorly then it is made a standard card. But generally the binning process is done on a batch bases. They don't test every single card, they instead pick a small sample out of a batch of cards and test those, they then rate the whole batch based on how the samples did. So it is possible to get a card that overclocks just as well as a TOP when buying the OC edition, but it is more likely that the OC card won't overclock quite as well as a TOP on average. Like for example my GTX670s, which are the standard editions, can match the TOP stock clock, they can actually do about 1075MHz, but there is no way they can do 1100MHz stable like the TOP cards can.

And yes, the Boost clock on the GTX670 is assumed to be 79MHz over the Base clock in programs like GPU-Z and eVGA Precision. However, boost clocks fluctuated based on the condition the card is in. So that 79MHz is just a rough estimation. When you are setting an offset you are really setting the Base clock, the Boost clock is controlled by the driver. And while the Boost clock is based off the Base clock, so raising the Base clock will effectively raise the Boost clock, the Boost clock is not a set number and it isn't what you are controlling with the overclocking software.

 

[Ed_1]

Your OC is to high, you need to drop it way down like I mentioned .
1202 would be a very good OC so see if that works or even try a step below (set boost to 1085) and work your way up .

 

[PersonWithTech]

I think you're expecting a little too much. ~1200Mhz is a pretty decent overclock for a 670. I tried out two different 670 DCII Tops. The first one maxed out around 1180Mhz and the second one(which is my current card) can reach 1215Mhz. You just need to lower your overclock a bit until you can pass a few rounds of heaven.

Yes. It is just that I remember having got a stable 99+ over the default boost (1158MHz), but that must have been before the newer drivers. I never id say, however, that I expected my card to reach 1200MHz boost clock. That is my max frequency, I had achieved during the driver reset/crash. So, according to what I've heard, the new drivers crash at lower clocks and prevent any major overclocking.

Oh, that guide is good, good info but thing is as Nvidia updated there drivers starting with 314.xx the clock head room went down , so now they don't clock as fast w/o reseting but you get new optimizations on the games to counter .

Also, is there any news/info I can find on this driver's side-effects, and does anywone no if nvidia is trying to improve/fix the issue?

 

[PersonWithTech]

The TOP cards are cherry picked cards. The testing that both go through at the factory is the binning process, if a card clocks very well it is made a TOP card, if it clocks good it is made an OC card, and if it clocks poorly then it is made a standard card. But generally the binning process is done on a batch bases. They don't test every single card, they instead pick a small sample out of a batch of cards and test those, they then rate the whole batch based on how the samples did. So it is possible to get a card that overclocks just as well as a TOP when buying the OC edition, but it is more likely that the OC card won't overclock quite as well as a TOP on average. Like for example my GTX670s, which are the standard editions, can match the TOP stock clock, they can actually do about 1075MHz, but there is no way they can do 1100MHz stable like the TOP cards can.

And yes, the Boost clock on the GTX670 is assumed to be 79MHz over the Base clock in programs like GPU-Z and eVGA Precision. However, boost clocks fluctuated based on the condition the card is in. So that 79MHz is just a rough estimation. When you are setting an offset you are really setting the Base clock, the Boost clock is controlled by the driver. And while the Boost clock is based off the Base clock, so raising the Base clock will effectively raise the Boost clock, the Boost clock is not a set number and it isn't what you are controlling with the overclocking software.

That's must have been the misconception that ASUS GPUTweak infected me with. When I used that, I had to move a slider which "represented" my boost clock.

 

[Ralfies]

Yes. It is just that I remember having got a stable 99+ over the default boost (1158MHz), but that must have been before the newer drivers. I never id say, however, that I expected my card to reach 1200MHz boost clock. That is my max frequency, I had achieved during the driver reset/crash. So, according to what I've heard, the new drivers crash at lower clocks and prevent any major overclocking.



Also, is there any news/info I can find on this driver's side-effects, and does anywone no if nvidia is trying to improve/fix the issue?

I've had no such problems with the latest drivers, and have heard of no such problems with them. I've had my card since not long after it came out, I update my drivers as often as they come out, and I have never had drivers affect my max clocks in any way.

 

[PersonWithTech]

I've had no such problems with the latest drivers, and have heard of no such problems with them. I've had my card since not long after it came out, I update my drivers as often as they come out, and I have never had drivers affect my max clocks in any way.

Well Ed_1 claims the newer > 314 drivers remove overclocking headroom, and I remember having achieved much higher clock rates in the past without the issue

 

[Ed_1]

That has been my experance and others to at time , it still OC but for me I needed 10mv+ added for same OC or lower by 13mhz .
This also happened with regaurd to Heven testing as they updated it and its harder more stressful than first version .
The Tessellation is more stressful in 4.0 than early version with extreme setting .

PS I don't even use heven for my testing with Tess on extreme , Valley also is good or FR3 , BF4 another good CPU/GPU game .

Note I have not tested all drivers after 320.xx but how good driver OC depends on how good driver can utilize the card, in general better it can the better chance OC won't be as high cause the card is putting out more . You should not be OC so high to be in this situation though .
So what I am saying is test to highest OC and once you find it back down a step and leave it .
You need a bit of buffer room for other app that you didn't test, that little bit is not going to do much.
Main thing is the OC is steady clock, with no drops from power or thermal throttling . That will give you smooth play .

 

[Doug2507]

Driver shouldn't be an issue, i OC'd my 670 top the other week on 331.65 and went 1400+/1900+. (custom bios obviously!)

 

[Bladedrummer]

OP, your OC is way too high for that card, you're getting a crash due to instability. Now, I will toss out the question no one has asked yet: have you run Heaven 4.0 on max settings successfully without OCing? Just run it at stock settings and see if it's stable. Some Asus factory overclocked cards have been known to be unstable already at their stock settings!

If you manage to pass the benchmark at stock settings, then try working your way up by upping your offset in 13MHz steps. I wouldn't expect to get more than +52 MHz offset, if that. I've managed to get +40 MHz on my EVGA GTX 670 FTW to run Heaven 4.0 stably. I also managed to get +575 MHz to my memory clock.
Unfortunately, upping the voltage didn't help me at all, I just had to lower the clock offset.

I'm not sure how much further we can push these factory overclocked cards, but in my experience, not that much.

 

[PersonWithTech]

OP, your OC is way too high for that card, you're getting a crash due to instability. Now, I will toss out the question no one has asked yet: have you run Heaven 4.0 on max settings successfully without OCing? Just run it at stock settings and see if it's stable. Some Asus factory overclocked cards have been known to be unstable already at their stock settings!

If you manage to pass the benchmark at stock settings, then try working your way up by upping your offset in 13MHz steps. I wouldn't expect to get more than +52 MHz offset, if that. I've managed to get +40 MHz on my EVGA GTX 670 FTW to run Heaven 4.0 stably. I also managed to get +575 MHz to my memory clock.
Unfortunately, upping the voltage didn't help me at all, I just had to lower the clock offset.

I'm not sure how much further we can push these factory overclocked cards, but in my experience, not that much.

Thanks for the reply. I've had my card since July and trust me, it runs Heaven maxed. Recently, after getting impatient with everything, I just overclocked it straight to base clock 1059MHz, or TOP edition speeds (I've got the cheaper OC edition). I got lucky, and turns out it is stable after many consecutive runs of Heaven. No over-volting was required. It is good to know that I had received a card better than it should be, however I did remember getting 99+ offset (Crazy right?) and it raised my expectations.

 

[Ed_1]

Ok, now I am confused , in your first post it was unstable at 81+ and now your saying it ok at 99+ .

Also it best to state clock speed when BM/app is running, stating offset is really meaningless as each card can have different base and boost clocks and what it boosts to in app is the important one .

Anyway anything over 1200 is good at stock voltages and anything near or over 1300 will need 1.2+ volts , thats generally the norm but YMMV .

 

[Bladedrummer]

I believe he was saying that he managed to get +99 stable with the old drivers. Personally, I've never manged to get more than +60 while gaming, less when benchmarking with Heaven. Either way, what you say seems about right, 1200 is the ball park value in my experience as well.

 

[Bones]

I think you're expecting a little too much. ~1200Mhz is a pretty decent overclock for a 670. I tried out two different 670 DCII Tops. The first one maxed out around 1180Mhz and the second one(which is my current card) can reach 1215Mhz. You just need to lower your overclock a bit until you can pass a few rounds of heaven.

This along with seeing what temps the cards are reaching during the bench.
Sometimes it's because the card simply gets too hot and this causes instability resulting in the driver reset. I've had it happen on several occasions and I noted that between temp issues and settings being too high, more often in the case of something being set at too high of an OC it's due to the card's memory clock speed rather than core speed itself. If you can, get some mini heatsinks and place them on the RAM IC's of the card, that should help them along.

You can let it run with Afterburner running alongside too and see where your temps peak while the bench is going, that will tell you what conditions it's running into and might help point you in the right direction.
It's not always about moar volts, sometimes is more about finesse when squeezing some good clocks from your stuff.