GPU OC Stability...

[JaZoN_XD]

So I'm here trying to overclock my new 970 and I'm trying at 1550MHZ core and 1880MHZ memory with NO extra voltage. I try Firestrike and it passes with a score of 10200 something. Then I try Heaven and Valley and they are all good. Then next day, I try Firestrike again and it crashes several times. Even if I increase the voltage to maintain the 1550MHZ, it still crashes. Reboot everytime it crashes... I try Valley with the OC and NO extra voltage, crash almost right away. I reboot and add extra voltage and it runs just fine... what the hell is going on can anyone explain? I am on the latest drivers. Thanks.

 

EDIT: Now I got this from Firestrike before the SystemInfo even finished running:

Unexpected error running tests.
Workload Single init returned error message: DXGI call IDXGISwapChain::SetFullscreenState failed [-2005270494]:

The requested functionality is not supported by the device or the driver.

DXGI_ERROR_NOT_CURRENTLY_AVAILABLE

[Bananasplit_00]

it was a one off OC, you got lycky first time and now something is slightly difirent and it dosent work, also i got 1603MHz core with +50mV on my 970  and that memory clock seems wierd seing as im running 3605MHz on my memory lol

[JaZoN_XD]

17 minutes ago, Bananasplit_00 said:

it was a one off OC, you got lycky first time and now something is slightly difirent and it dosent work, also i got 1603MHz core with +50mV on my 970  and that memory clock seems wierd seing as im running 3605MHz on my memory lol

take your memory clock and divide it by 2, that's the actually memory clock. Otherwise, multiply by 4 to get effective data clock

[Bananasplit_00]

1 minute ago, JaZoN_XD said:

take your memory clock and divide it by 2, that's the actually memory clock. Otherwise, multiply by 4 to get effective data clock

really? never heard that before but ok, in that case its the memeory more then likely causing the instability

[JaZoN_XD]

15 minutes ago, Bananasplit_00 said:

really? never heard that before but ok, in that case its the memeory more then likely causing the instability

Naw it isn't... i brought it all the way back to stock and put JUST the core clock OC and it still crashed. And now, I JUST benched it again with max OC (1550 core and 3760 or 1880 mem) and it worked fine on valley even WITHOUT any voltage increase (stock voltage at 1.2v)

[Bananasplit_00]

On 2016-02-03 at 0:42 AM, JaZoN_XD said:

Naw it isn't... i brought it all the way back to stock and put JUST the core clock OC and it still crashed. And now, I JUST benched it again with max OC (1550 core and 3760 or 1880 mem) and it worked fine on valley even WITHOUT any voltage increase (stock voltage at 1.2v)

well thats nice then

[JaZoN_XD]

1 hour ago, Bananasplit_00 said:

well thats nice then

yeah which tears me apart... I put just in case 1.225V to keep it stable for all games (GTA V for some reason doesn't like the 1550 when it's at stock voltage) and now every single benchmark runs stably at that overclock EXCEPT firestrike... it's not even that I'm crashing during the test and it just crashes to the desktop, it crashes whenever it loads in a new scene. Right after the loading bar finishes, it goes to a blank screen of a weird colour, similar to grey and blue mixed together. It can be different colours but it's just a really sad colour. I can't alt tab out of it or control delete or anything, I have to hard restart everytime this happens. I don't understand!

[Bananasplit_00]

23 hours ago, JaZoN_XD said:

yeah which tears me apart... I put just in case 1.225V to keep it stable for all games (GTA V for some reason doesn't like the 1550 when it's at stock voltage) and now every single benchmark runs stably at that overclock EXCEPT firestrike... it's not even that I'm crashing during the test and it just crashes to the desktop, it crashes whenever it loads in a new scene. Right after the loading bar finishes, it goes to a blank screen of a weird colour, similar to grey and blue mixed together. It can be different colours but it's just a really sad colour. I can't alt tab out of it or control delete or anything, I have to hard restart everytime this happens. I don't understand!

i used to get that when i had my processor OCd to 4.6GHz and my graphics card at what it is now, removed the CPU OC and it was still doing it. turned out it was the graphics card running to high memory clock for me, turned it down about 30MHz and it was fine and i have still not put back my CPU overclock, not that i need it but its just because i can

[JaZoN_XD]

7 hours ago, Bananasplit_00 said:

i used to get that when i had my processor OCd to 4.6GHz and my graphics card at what it is now, removed the CPU OC and it was still doing it. turned out it was the graphics card running to high memory clock for me, turned it down about 30MHz and it was fine and i have still not put back my CPU overclock, not that i need it but its just because i can

haha i guess you're lucky then because I tried to put on JUST my core clock OC and turned the memory to stock speeds and even then firestrike will crash. CPU for meis obviously no OC since it's locked.

[Monarch]

On 2/5/2016 at 5:32 PM, JaZoN_XD said:

yeah which tears me apart... I put just in case 1.225V to keep it stable for all games (GTA V for some reason doesn't like the 1550 when it's at stock voltage) and now every single benchmark runs stably at that overclock EXCEPT firestrike... it's not even that I'm crashing during the test and it just crashes to the desktop, it crashes whenever it loads in a new scene. Right after the loading bar finishes, it goes to a blank screen of a weird colour, similar to grey and blue mixed together. It can be different colours but it's just a really sad colour. I can't alt tab out of it or control delete or anything, I have to hard restart everytime this happens. I don't understand!

You've been caught in GPU boost 2.0's trap.

 

Long story:

GPU boost operates on the concept of steps. Each step is 13 MHz apart, and each step is basically a clock speed that has voltage range assigned to it.

When you OC the GPU, the whole boost table shifts. You don't just change the highest clock speed. So for example, say your card boosts to CLK 63 by default. Say that's 1380 MHz. When you add +150 MHz with Afterburner, for example, the whole table shifts so that CLK 63 becomes 1380+150=1530 (highest boost CLK your card will boost to). Even if you OC trough Bios modding, the only way to do it is to shift the whole boost table. But then, for example, CLK 47 is no longer 1177 MHz, but 1329, and yet the voltage ranges assigned to each CLK remain the same. This will obviously cause instability. So you'd need to shift the whole boost CLK voltage table as well when overclocking, but sadly this can only be done via a bios mod.

 

TL;DR:

When you OC the GPU, the whole boost table shifts, and then voltage ranges assigned to each clock speed are too low, as they are meant for lower clocks. The only way to fix it is bios modding.

[JaZoN_XD]

24 minutes ago, Monarch said:

You've been caught in GPU boost 2.0's trap.

 

Long story:

GPU boost operates on the concept of steps. Each step is 13 MHz apart, and each step is basically a clock speed that has voltage range assigned to it.

When you OC the GPU, the whole boost table shifts. You don't just change the highest clock speed. So for example, say your card boosts to CLK 63 by default. Say that's 1380 MHz. When you add +150 MHz with Afterburner, for example, the whole table shifts so that CLK 63 becomes 1380+150=1530 (highest boost CLK your card will boost to). Even if you OC trough Bios modding, the only way to do it is to shift the whole boost table. But then, for example, CLK 47 is no longer 1177 MHz, but 1329, and yet the voltage ranges assigned to each CLK remain the same. This will obviously cause instability. So you'd need to shift the whole boost CLK voltage table as well when overclocking, but sadly this can only be done via a bios mod.

 

TL;DR:

When you OC the GPU, the whole boost table shifts, and then voltage ranges assigned to each clock speed are too low, as they are meant for lower clocks. The only way to fix it is bios modding.

Ahh, I was thinking about that too. I've noticed this shift as well. When the GPU is under 60C, even with just a +25mV voltage, the voltage that's reported goes to 1.25V. That causes the core clock to jump to 1563 instead of 1550. When the GPU goes above 60C, the voltage goes back down to 1.225V which corresponds to the +25mV that I added. However, between 60C and 70C, if I push +50mV on the slider, the card will force 1.25V. But if I go past 70C, the maximum voltage it seems to allow me to run is 1.225V. Even if I push +87mV or something, the voltage doesn't move and stays at the 1.225V if the card is above 70C. I know at stock, my card boosts to CLK 64 (1405) but for a short moment when the card is under 60C, the voltage will be upped to 1.225V (this is when it is +0mV) and boost to CLK 65 which is 1418. After that temp, it will go back down to 1.2V and remain constant at CLK 64. So I guess the way to mitigate this might be turning off boost?

 

EDIT: But despite that you said the overclock shifts the whole boost table, the NON 3D clock speed which is CLK 47 at 1190 still remain constant even though I have a +132 on afterburner. Weird.

[Monarch]

40 minutes ago, JaZoN_XD said:

Ahh, I was thinking about that too. I've noticed this shift as well. When the GPU is under 60C, even with just a +25mV voltage, the voltage that's reported goes to 1.25V. That causes the core clock to jump to 1563 instead of 1550. When the GPU goes above 60C, the voltage goes back down to 1.225V which corresponds to the +25mV that I added. However, between 60C and 70C, if I push +50mV on the slider, the card will force 1.25V. But if I go past 70C, the maximum voltage it seems to allow me to run is 1.225V. Even if I push +87mV or something, the voltage doesn't move and stays at the 1.225V if the card is above 70C. I know at stock, my card boosts to CLK 64 (1405) but for a short moment when the card is under 60C, the voltage will be upped to 1.225V (this is when it is +0mV) and boost to CLK 65 which is 1418. After that temp, it will go back down to 1.2V and remain constant at CLK 64. So I guess the way to mitigate this might be turning off boost?

Yeah, the CLK your GPU will boost to depends on things like ASIC quality, voltage, temperature. Once you reach a certain temp, it downclocks and downvolts the GPU. Boost 2.0 is actually a good thing, so you wouldn't want to disable it, but you'd need to correct it in the bios so that the voltage ranges match the boost table and its clocks, and also disable the downvolting and downclocking.

But it's a bit risky and voids warranty. And in addition to that, no one knows exactly how the bios and boost 2.0 work. You can find guides on overclock.net on how to mod bios and fix the issues you're having, but a lot of info's missing and I've found some things they say to be inaccurate.  

I think attention should be brought to this issue, it's pretty serious. People say Maxwell GPUs are good at overclocking, but they are unaware of the fact that GPU boost 2.0 kills overclocks.

 

Quote

EDIT: But despite that you said the overclock shifts the whole boost table, the NON 3D clock speed which is CLK 47 at 1190 still remain constant even though I have a +132 on afterburner. Weird.

 

Are you sure? Cause afaik that's not possible. CLKs 35-74 shift when you OC, so 47 has to shift as well.

 

 

[JaZoN_XD]

3 hours ago, Monarch said:

Yeah, the CLK your GPU will boost to depends on things like ASIC quality, voltage, temperature. Once you reach a certain temp, it downclocks and downvolts the GPU. Boost 2.0 is actually a good thing, so you wouldn't want to disable it, but you'd need to correct it in the bios so that the voltage ranges match the boost table and its clocks, and also disable the downvolting and downclocking.

But it's a bit risky and voids warranty. And in addition to that, no one knows exactly how the bios and boost 2.0 work. You can find guides on overclock.net on how to mod bios and fix the issues you're having, but a lot of info's missing and I've found some things they say to be inaccurate.  

I think attention should be brought to this issue, it's pretty serious. People say Maxwell GPUs are good at overclocking, but they are unaware of the fact that GPU boost 2.0 kills overclocks.

Yeah, I think the voltage ranges are still fine... the card is running fine on the overclock even with the default voltage ranges. All I need to do is to find a way to disable the downvolting and I am going to be good. That's the one specific thing I need to find out how to do. You mentioned I can find this information on overclock.net? I'll trust you with that  

 

3 hours ago, Monarch said:

Are you sure? Cause afaik that's not possible. CLKs 35-74 shift when you OC, so 47 has to shift as well.

Yeah it should be all the CLKs that are highlighted in yellow correct? For some reason, CLK 47 doesn't shift. In the attached file, the top graph is the voltage and the bottom is the clock speed. I had 3DMark opened, which induces the card to push more voltage and power and ups the clock speed to the Non-3D speeds.

[Monarch]

10 hours ago, JaZoN_XD said:

Yeah, I think the voltage ranges are still fine... the card is running fine on the overclock even with the default voltage ranges. All I need to do is to find a way to disable the downvolting and I am going to be good. That's the one specific thing I need to find out how to do. You mentioned I can find this information on overclock.net? I'll trust you with that  

 

Yeah it should be all the CLKs that are highlighted in yellow correct? For some reason, CLK 47 doesn't shift. In the attached file, the top graph is the voltage and the bottom is the clock speed. I had 3DMark opened, which induces the card to push more voltage and power and ups the clock speed to the Non-3D speeds.

 

I think I have an idea why that is. In my case, my clock always jumps to 1177. That's because if I go to the common tab in bios tweaker, it says that my card's base clock is 1177, which means it starts from there, no matter what CLK 1177 MHz is under.

 

Now, the reason why you crash on loading screens is because right before the loading screen goes away, your clock jumps from 1190 to a couple other CLKs until it reaches the final CLK, which is your max clock speed. The problem is, those clocks it jumps across to reach the final CLK have too low voltage ranges. I had the same exact issue. So you have to shift the voltage range tables to match your clock speeds.

Can you upload your bios so i can see what's going on?

 

Oh and here are the threads that helped me:

http://www.overclock.net/t/1522651/disable-boost-and-bake-in-max-game-stable-clocks-for-maxwell

http://www.overclock.net/t/1467851/nvidia-maxwell-kepler-bios-editing-thread-gtx-2xx-to-9xx-now-supported

You can also find discussions about this in the gtx 970 thread on OCN, but there are a lot of pages so you have to search. part of the discussion is located around here: http://www.overclock.net/t/1514085/official-nvidia-gtx-970-owners-club/17100#post_24041345

I suggest you consider the findings of a user named flexy123.

 

We also have 970 owners club here, so you can check it out and find some advice there as well: 

 

You shouldn't disable boost imo, but rather configure it so suit your OC. The first thread I linked can help you explain what's what and how some things work, but I suggest you don't follow a single guide. Do a research, consider what people found to be working properly and form your own opinion on what's the best way to mod the bios. Like I said no one knows exactly how the bios works, and I personally found some info to be inaccurate. 

I can tell you what worked for me, but I don't know for sure if it should be done the way I did it.

[JaZoN_XD]

1 hour ago, Monarch said:

~~~~~SNIP~~~~~

Thank you so much man! I'm going to look into those links. Now that I've also taken a look at the 970 owner's club and some of the BIOS that he provided, I think I'm going to do the same - I'm going to shift the boost table so that where CLK 64 was originally will now be the last CLK which is 74 or something. Also, I'll do like you said and increase the voltage ranges on the different CLKs, but I don't know why they should be a problem since the voltage on them are from x to 1.283 which is obviously plenty enough to run any speeds. Thanks a lot again, I'll let you know some results once I have the time to test them.

[Monarch]

1 hour ago, JaZoN_XD said:

Thank you so much man! I'm going to look into those links. Now that I've also taken a look at the 970 owner's club and some of the BIOS that he provided, I think I'm going to do the same - I'm going to shift the boost table so that where CLK 64 was originally will now be the last CLK which is 74 or something. Also, I'll do like you said and increase the voltage ranges on the different CLKs, but I don't know why they should be a problem since the voltage on them are from x to 1.283 which is obviously plenty enough to run any speeds. Thanks a lot again, I'll let you know some results once I have the time to test them.

Well on stock BIOS they're not 1.283. But if you take settings from a custom bios that already has voltage ranges increased then they shouldn't be a problem. 

[JaZoN_XD]

1 hour ago, Monarch said:

Well on stock BIOS they're not 1.283. But if you take settings from a custom bios that already has voltage ranges increased then they shouldn't be a problem. 

Stock bios: everything that's CLK 60 and above is 1281.3mV

 

EDIT: I just shifted all of the boost clocks over so that my current boost clock of CLK 64 @1405 is corresponding with the last boost clock which is CLK 74. Now I am using a BIOS that has CLK 74 as 1405. However, the card still boosts up to only CLK 64 which is 1278, not up to CLK 74 which is 1405. I guess I have to start messing with the voltage instead of the boost table now...

[Monarch]

1 hour ago, JaZoN_XD said:

Stock bios: everything that's CLK 60 and above is 1281.3mV

 

 

EDIT: I just shifted all of the boost clocks over so that my current boost clock of CLK 64 @1405 is corresponding with the last boost clock which is CLK 74. Now I am using a BIOS that has CLK 74 as 1405. However, the card still boosts up to only CLK 64 which is 1278, not up to CLK 74 which is 1405. I guess I have to start messing with the voltage instead of the boost table now...

I just checked my stock bios, and yes, CLKs 60-74 are set to use up to 1281.3 mV. But those clocks are not the issue. The problem is the instability at lower clocks, which have a limit much lower limit than 1281.3mV. That's why you're crashing at loading screens.

 

As for your card boosting up to only CLK 64, it seems that it depends on ASIC quality and voltage. My card boosts to CLK 68 at best, for example, with ASIC quality of 77.5%, but only if I add more voltage. Otherwise it boosts to CLK 67. You said in one of your earlier posts that sometimes your GPU boosts to CLK 65, but then throttles to CLK 64? This may indicate that it's in fact CLK 65 that your GPU will boost to at best, but if it has enough voltage.

 

 

So what you need to do is shift the voltage tables. If you shift the whole boost table by 13 MHz, then CLK 50 for example will become clock speed that used to be CLK 51. So you need to take the voltage range that's meant for CLK 51, and apply it to CLK 50. Otherwise you're running CLK 50 that's meant to run at voltage assigned to CLK 51.

[JaZoN_XD]

20 minutes ago, Monarch said:

I just checked my stock bios, and yes, CLKs 60-74 are set to use up to 1281.3 mV. But those clocks are not the issue. The problem is the instability at lower clocks, which have a limit much lower limit than 1281.3mV. That's why you're crashing at loading screens.

 

As for your card boosting up to only CLK 64, it seems that it depends on ASIC quality and voltage. My card boosts to CLK 68 at best, for example, with ASIC quality of 77.5%, but only if I add more voltage. Otherwise it boosts to CLK 67. You said in one of your earlier posts that sometimes your GPU boosts to CLK 65, but then throttles to CLK 64? This may indicate that it's in fact CLK 65 that your GPU will boost to at best, but if it has enough voltage.

 

 

So what you need to do is shift the voltage tables. If you shift the whole boost table by 13 MHz, then CLK 50 for example will become clock speed that used to be CLK 51. So you need to take the voltage range that's meant for CLK 51, and apply it to CLK 50. Otherwise you're running CLK 50 that's meant to run at voltage assigned to CLK 51.

Understand. So in theory, if I shift all of the voltage of the lower clock states to like 1.225, then they will all be at 1.225mV when I'm at the loading screen? Thing is, I have a +132 overclock and even if I take CLK 59 (1342 MHZ) as seen above and add 132 to it, the overclocked version of CLK 59 (1474MHZ) is still well within the voltage range set in the voltage table for CLK 59.

 

Anyway, I'll give it a try to shift the whole voltage table later today when I have the time. I was experimenting with the BIOS just now and I've found a way to prevent the card from throttling and making it constant at 1.25V. However, that's when I use power at 115%. If I drop the power, the voltage will also decrease. But even at 1.25V, firestrike still crashes at loading screen.

[Coaxialgamer]

A used card. 

[Monarch]

1 hour ago, JaZoN_XD said:

Understand. So in theory, if I shift all of the voltage of the lower clock states to like 1.225, then they will all be at 1.225mV when I'm at the loading screen? Thing is, I have a +132 overclock and even if I take CLK 59 (1342 MHZ) as seen above and add 132 to it, the overclocked version of CLK 59 (1474MHZ) is still well within the voltage range set in the voltage table for CLK 59.

 

Anyway, I'll give it a try to shift the whole voltage table later today when I have the time. I was experimenting with the BIOS just now and I've found a way to prevent the card from throttling and making it constant at 1.25V. However, that's when I use power at 115%. If I drop the power, the voltage will also decrease. But even at 1.25V, firestrike still crashes at loading screen.

I see what you mean. The thing is, the GPU doesn't take advantage of the full voltage range. In fact, from what I've observed, it uses the minimum value, or slightly more voltage than the minimum value. I should have been more clear on how the voltage range works. I made it sound like it's just the voltage range that matters, while it's actually the minimum values that are important.

[JaZoN_XD]

47 minutes ago, Monarch said:

I see what you mean. The thing is, the GPU doesn't take advantage of the full voltage range. In fact, from what I've observed, it uses the minimum value, or slightly more voltage than the minimum value. I should have been more clear on how the voltage range works. I made it sound like it's just the voltage range that matters, while it's actually the minimum values that are important.

Yeah, I've noticed that too. What do you recon some shifted voltages should be? I'm not at my computer right now I can't really test, but this shifting of voltage seems hopeful at the moment

[Monarch]

26 minutes ago, JaZoN_XD said:

Yeah, I've noticed that too. What do you recon some shifted voltages should be? I'm not at my computer right now I can't really test, but this shifting of voltage seems hopeful at the moment

No need to randomly increase voltage values. Voltages should progressively increase from CLK 35 to CLK 74, in the same way they do in the stock BIOS. Like I said, if you add +13 MHz, CLK 50 becomes 51, then what you do is take the voltage for CLK 51 and apply it to CLK 50. And you do this for all CLKs. In this example the boost table is shifted by 1 step (13 MHz) forward, which is why voltages for CLKs in the voltage table also need to be shifted by 1 step, but backwards. Voltage assigned to CLK 51 -> voltage assigned to CLK 50.

 

Again this may not be the best way, this may not be the right way, but it's what makes sense imo and it's what works for me.

[JaZoN_XD]

3 hours ago, Monarch said:

No need to randomly increase voltage values. Voltages should progressively increase from CLK 35 to CLK 74, in the same way they do in the stock BIOS. Like I said, if you add +13 MHz, CLK 50 becomes 51, then what you do is take the voltage for CLK 51 and apply it to CLK 50. And you do this for all CLKs. In this example the boost table is shifted by 1 step (13 MHz) forward, which is why all CLKs in the voltage table also need to be shifted by 1 step, but backwards. Voltage assigned to CLK 51 -> voltage assigned to CLK 50.

 

Again this may not be the best way, this may not be the right way, but it's what makes sense imo and it's what works for me.

Alright thank you very much man. I'll work on that

 

EDIT: Interestingly though, although you say that the lower limit is the actual voltage used for that clock state, CLK 64 which is my boost clock will use 1.200V which is obviously not the lower limit of the BIOS for CLK 64.

[JaZoN_XD]

On 2/7/2016 at 3:52 PM, Monarch said:

~Snip~

Well, I've moved all the voltage tables so that starting from CLK 35, it's using the voltage range that's roughly 10 clock states above it because that's about what my +132 MHZ core clock corresponds to and I finally ran Firestrike with that OC properly! I am going to test it again because I am still kinda skeptical with the stability and also, I did recieve a really weird result for the score, I got a really low score of only 13070 graphics score when I normally should get 13900+. Also, the driver was not approved because I am using the new drivers for the "eject Nvidia GPU" hotfix which is 361.82.

 

EDIT: I was being a derp and I still had shadowplay on which decreased the performance by a little bit. A BIG THANK YOU TO YOU!