Then run 368.69 installer as administrator. Restart.
Okay, just updated to 368.69 and not seeing any difference. Card is still running at the base GPU clock and the memory is running about 500 MHz under. Tried going into Nvidia Inspector and adjusting the memory clock for P2. Set it to the max (5005) and clicked Apply, but nothing happened. Am I missing something somewhere? Thanks for the advice so far.
P.S. I know the card actually boosts in P0, as I ran 3D Mark and the core clock boosted to 1947 and the memory went to its normal speed of 5005.
I seem to recall someone saying something about making changes in the P0 settings in NVI which will propagate through the performance levels. I am unable to find that particular post, though, and wonder if that is what I am missing here.
Recap: Running EVGA GTX 1080 Superclocked. Base clock 1708, boosts to 1947 at P0. Memory clock is at 5005 at P0, but only wants to run at 4513 in P2. Core clock will not boost above base at P2. I was able to increase the memory clock for P2 from Precision X OC, but am unable to do the same for core clock.
Just watched a trailer on Netflix linked from an email while I had Einstein going. Saw the core clock jump to 1961MHz. I checked and NVI reported I was still in P2. When the trailer ended, clock went back down to 1708. Tried to replicate by logging into Netflix and sampling a few shows but was unable to get the clock up again. Same goes for YouTube.
When you play video, it's a signal to the card to go into P0 mode. Once you stop video, it returns its boost levels to P2 limits. Although strange that the core clocks won't boost to thermal appropriate levels under P2, you can still have it do so manually as others have posted.
The key is to make changes BEFORE you have initiated BOINC and the card is performing computational tasks. Exit BOINC, make sure that your P0 settings for core and memory clock are what you desire, and then proceed to make the same changes for the P2 mode. Then, initiate BOINC and your card should run as you have set it to.
Also, I have found on my Maxwell series 970's, a softare such as MSI Afterburner or EVGA PrecisionX can interfere with the changes I make on NV Inspector, so be careful that you are not mixing overclocking programs together as it may cause issues.
I hope this helps you solve your issue.
P.S: Any changes made in NVI will reset to default if you log out or restart your computer. Solution is to save a profile and autorun it at startup, but I have seen that to be a little complicated to achieve.
I'm the one who has posted in this thread about overclocking the 1070 for Einstein BRP6 work.
I imagine issues are the same for the 1080.
1. stop all running BOINC on GPU jobs. I prefer to do this by suspending the complete queue of unstarted work first, then suspending specifically the currently running GPU tasks. You can leave the CPU tasks running if you like.
2. If you are using the NVI GUI, check at this point what P-state you are in. It should be P8. (and your 1080 core and memory clock rates should be some really low 3-digit numbers).
3. either use the NVI GUI or command line NVI to specify your desired delta clocks for the P0 state.
4. resume running the BOINC GPU work.
99% of all the times I have done this, I have used the command line interface. But early on I did use the GUI and that worked also.
Just to give a fully worked example, here is the exact command line I issued (with current context set to the NVI executable directory) most recently.
On my system, with BRP6 tasks running, this gave me 1961.5 core, and 2249.1 memory clock when stabilized on the GPU-Z reporting scale. The memory clock does not change with temperature, but the core clock drops a couple of times by small increments as the GPU heats up.
My system has two GPUs installed, and at the moment my 1070 is at index 1. Most likely yours is at index 0, in which case you would need to specify that in the parameter lists to the two switches. If you want to see the complete list of available command line NVI switches, with full syntax, just issue the command line with a "/?" and you'll get a pop-up.
Also, of course, your prudent initial choice of offsets would not be my current values. Not only is yours a 1080 and mine a 1070, but I don't run games, and have, perhaps, less reason to worry that this method may set the P0 state clock rates so high as to cause problems, possibly severe ones. Possibly you can use NVI's extensive profiling facility to adjust behavior in the specific games you use to your needs--I've not looked into that feature at all.
Another thing I just now thought of and am updating my original post to say: your 1080 has a different type of memory than my 1070, and it is quite possible that the whole question of memory clock rate derating, offsets, adjustment behavior, and such may differ materially.
I am not claiming this is the ONLY method that can work. I tried several that did not work before hitting on this, and found it sufficient to my purpose. I believe matters for Pascal are not exactly the same in this regard as for Maxwell2.
I just tried the last couple posts suggestions and it doesn't seem to be working for me. I did not adjust the P0 state in NVI as I don't need those clocks changed, but I did attempt to change the P2 clocks (with BOINC stopped) and this did not produce any change. However, I seem to be able to adjust the memory clock from within PrecisionX OC even in the P2 state (which was not possible with my Maxwell cards). I currently have it set at 5265MHz (+750 from P2, +250 from P0) and it seems likely I'll be able to push it further.
For some reason, the core clock seems to have adjusted itself and has been holding steady at 1961MHz. I don't know why it ticked up, but it did so when I watched a couple videos within Steam and is now staying there even without videos playing and after having restarted BOINC. I am still in P2.
Thanks for the advice and recommendations. I am currently running BRP6 x4 and am seeing about 2 hours completion time per task, most of which was with the core clock at 1708MHz.
I just tried the last couple posts suggestions and it doesn't seem to be working for me. I did not adjust the P0 state in NVI as I don't need those clocks changed
One of those sentences is not true--most likely the first one.
I just tried the last couple posts suggestions and it doesn't seem to be working for me. I did not adjust the P0 state in NVI as I don't need those clocks changed
One of those sentences is not true--most likely the first one.
I guess I'm misunderstanding something then. Is it necessary to change the P0 clocks in order to change the P2 clock? My concern is that I don't want the card to be overclocked when running in P0.
I guess I'm misunderstanding something then. Is it necessary to change the P0 clocks in order to change the P2 clock? My concern is that I don't want the card to be overclocked when running in P0.
Thanks.
I respect your concern, and before I saw your post updated my original post to mention it. But I don't know another method. And I did try to adjust the P2 state clocks directly without success. To quote myself:
Quote:
I am not claiming this is the ONLY method that can work. I tried several that did not work before hitting on this, and found it sufficient to my purpose. I believe matters for Pascal are not exactly the same in this regard as for Maxwell2.
Things may be very different for your 1080, but for my 1070 the BRP6 execution time improvement available from memory clock overclocking was much more than that available from core clock overclocking.
My current BRP6 elapsed times at 3X on the 1070 average about 1:35:00, despite the fact it is sharing the host with a 750Ti. At default conditions 3X times for this card running alone averaged 1:54:05.
RE: RE: RE: Which
)
Okay, just updated to 368.69 and not seeing any difference. Card is still running at the base GPU clock and the memory is running about 500 MHz under. Tried going into Nvidia Inspector and adjusting the memory clock for P2. Set it to the max (5005) and clicked Apply, but nothing happened. Am I missing something somewhere? Thanks for the advice so far.
P.S. I know the card actually boosts in P0, as I ran 3D Mark and the core clock boosted to 1947 and the memory went to its normal speed of 5005.
Installed PrecisionX OC and
)
Installed PrecisionX OC and was able to boost the memory clock from there while in P2, but it doesn't seem to work for the core clock.
I seem to recall someone
)
I seem to recall someone saying something about making changes in the P0 settings in NVI which will propagate through the performance levels. I am unable to find that particular post, though, and wonder if that is what I am missing here.
Recap: Running EVGA GTX 1080 Superclocked. Base clock 1708, boosts to 1947 at P0. Memory clock is at 5005 at P0, but only wants to run at 4513 in P2. Core clock will not boost above base at P2. I was able to increase the memory clock for P2 from Precision X OC, but am unable to do the same for core clock.
Just watched a trailer on
)
Just watched a trailer on Netflix linked from an email while I had Einstein going. Saw the core clock jump to 1961MHz. I checked and NVI reported I was still in P2. When the trailer ended, clock went back down to 1708. Tried to replicate by logging into Netflix and sampling a few shows but was unable to get the clock up again. Same goes for YouTube.
When you play video, it's a
)
When you play video, it's a signal to the card to go into P0 mode. Once you stop video, it returns its boost levels to P2 limits. Although strange that the core clocks won't boost to thermal appropriate levels under P2, you can still have it do so manually as others have posted.
The key is to make changes BEFORE you have initiated BOINC and the card is performing computational tasks. Exit BOINC, make sure that your P0 settings for core and memory clock are what you desire, and then proceed to make the same changes for the P2 mode. Then, initiate BOINC and your card should run as you have set it to.
Also, I have found on my Maxwell series 970's, a softare such as MSI Afterburner or EVGA PrecisionX can interfere with the changes I make on NV Inspector, so be careful that you are not mixing overclocking programs together as it may cause issues.
I hope this helps you solve your issue.
P.S: Any changes made in NVI will reset to default if you log out or restart your computer. Solution is to save a profile and autorun it at startup, but I have seen that to be a little complicated to achieve.
Matt, I'm the one who has
)
Matt,
I'm the one who has posted in this thread about overclocking the 1070 for Einstein BRP6 work.
I imagine issues are the same for the 1080.
1. stop all running BOINC on GPU jobs. I prefer to do this by suspending the complete queue of unstarted work first, then suspending specifically the currently running GPU tasks. You can leave the CPU tasks running if you like.
2. If you are using the NVI GUI, check at this point what P-state you are in. It should be P8. (and your 1080 core and memory clock rates should be some really low 3-digit numbers).
3. either use the NVI GUI or command line NVI to specify your desired delta clocks for the P0 state.
4. resume running the BOINC GPU work.
99% of all the times I have done this, I have used the command line interface. But early on I did use the GUI and that worked also.
Just to give a fully worked example, here is the exact command line I issued (with current context set to the NVI executable directory) most recently.
On my system, with BRP6 tasks running, this gave me 1961.5 core, and 2249.1 memory clock when stabilized on the GPU-Z reporting scale. The memory clock does not change with temperature, but the core clock drops a couple of times by small increments as the GPU heats up.
My system has two GPUs installed, and at the moment my 1070 is at index 1. Most likely yours is at index 0, in which case you would need to specify that in the parameter lists to the two switches. If you want to see the complete list of available command line NVI switches, with full syntax, just issue the command line with a "/?" and you'll get a pop-up.
Also, of course, your prudent initial choice of offsets would not be my current values. Not only is yours a 1080 and mine a 1070, but I don't run games, and have, perhaps, less reason to worry that this method may set the P0 state clock rates so high as to cause problems, possibly severe ones. Possibly you can use NVI's extensive profiling facility to adjust behavior in the specific games you use to your needs--I've not looked into that feature at all.
Another thing I just now thought of and am updating my original post to say: your 1080 has a different type of memory than my 1070, and it is quite possible that the whole question of memory clock rate derating, offsets, adjustment behavior, and such may differ materially.
I am not claiming this is the ONLY method that can work. I tried several that did not work before hitting on this, and found it sufficient to my purpose. I believe matters for Pascal are not exactly the same in this regard as for Maxwell2.
I just tried the last couple
)
I just tried the last couple posts suggestions and it doesn't seem to be working for me. I did not adjust the P0 state in NVI as I don't need those clocks changed, but I did attempt to change the P2 clocks (with BOINC stopped) and this did not produce any change. However, I seem to be able to adjust the memory clock from within PrecisionX OC even in the P2 state (which was not possible with my Maxwell cards). I currently have it set at 5265MHz (+750 from P2, +250 from P0) and it seems likely I'll be able to push it further.
For some reason, the core clock seems to have adjusted itself and has been holding steady at 1961MHz. I don't know why it ticked up, but it did so when I watched a couple videos within Steam and is now staying there even without videos playing and after having restarted BOINC. I am still in P2.
Thanks for the advice and recommendations. I am currently running BRP6 x4 and am seeing about 2 hours completion time per task, most of which was with the core clock at 1708MHz.
RE: I just tried the last
)
One of those sentences is not true--most likely the first one.
RE: RE: I just tried the
)
I guess I'm misunderstanding something then. Is it necessary to change the P0 clocks in order to change the P2 clock? My concern is that I don't want the card to be overclocked when running in P0.
Thanks.
RE: I guess I'm
)
I respect your concern, and before I saw your post updated my original post to mention it. But I don't know another method. And I did try to adjust the P2 state clocks directly without success. To quote myself:
Things may be very different for your 1080, but for my 1070 the BRP6 execution time improvement available from memory clock overclocking was much more than that available from core clock overclocking.
My current BRP6 elapsed times at 3X on the 1070 average about 1:35:00, despite the fact it is sharing the host with a 750Ti. At default conditions 3X times for this card running alone averaged 1:54:05.