Never looked at memory clock speeds before, but after reading this thread I checked my clock speeds for my GTX 980 running GPUGRID long runs and yep, I only get 1502MHz on my memory.
This explains why GTX 980 isn't faster that 780 Ti in GPUGrid at stock speeds. I was already wondering why I wasn't beating 780 Ti...
After letting the card run at the higher memory clock of 1752 MHz, which is really the speced clock, the wu times (running two at a time) for BRP4G have gone from 3220 seconds 2850 seconds average. This is an improvement of 12%. None of them have any errors.
The clock rates can not be increased any further in this P2 mode, if you try they jump right back to the previous maximum.
Update:
It seems I was wrong about the memory OC causing my blue screens! Encouraged by your results I tried 3.5 GHz again and it's been running flawless since almost 2 days :)
So the message to get out to GM204 owners is just: Use nVidia Inspector to set 3.5 GHz in P2!
Quote:
The clock rates can not be increased any further in this P2 mode, if you try they jump right back to the previous maximum.
Same here. In fact, I can not set anything above 3.5 GHz for P0 either. It also jumps right back. Tried with Inspector, Precision and Afterburner. I'm confused, as the web is full of reports of memory overclocks.
Thanks to skgiven from GPU-Grid I can now overclock my memory! Here's how:
- the GPU must not be crunching BOINC (either pause your GPU project, or all GPUs, or suspend BOINC completely)
- in the nVidia Inspector OC tab set the overclock for P0 (because you can't go any higher than this in P2)
- now you can set up to this memory clock for P2 as well
- apply & have fun
GTX970/980 are reported to reach about 4 GHz memory clock. It should be interesting to see if this can boost Einstein performance another 10%.
Thanks to skgiven from GPU-Grid I can now overclock my memory! Here's how
Thanks for posting that.
I can affirm that I applied this process on my 970 running on a dual-core Haswell under Windows 7.
I did the before-after observations using MSI Afterburner (which has been my default fan speed controller and general GPU observer for some time).
I decided to work up by somewhat small increments.
The initial condition, while running Perseus Arm Survey, as reported by MSI Afterburner, with not changes made by me:
Core clock 1367, Memory clock 3005
After asking Inspector to raise the P2 memory clock by 95, Afterburner reports:
Core clock 1367, Memory clock 3100
I've got pretty heavy process Lasso intervention, and am only running two CPU BOINC tasks on one side of the two physical cores of this HT host, so my elapsed times are usually pretty repeatable. I should be able to tell overnight something about elapsed time improvement.
I just recently turned down the memory clock (from stock) on one of my GTX 660s hoping to reduce the rate of uncommanded downclocks. It may have helped, though the history is erratic enough that I am not confident. If it really did help, and if the 970 has a similar possibility, aside from outright malfunction, bad WU's, or blue screens, uncommanded downclocks might be another limiting consideration to 970 overclocking.
(side comment--in this setup GPU-Z reports the memory clock rate as half that reported by MSI Afterburner--to within a very small error margin)
(side comment--in this setup GPU-Z reports the memory clock rate as half that reported by MSI Afterburner--to within a very small error margin)
Yes, they use different "DDR clock speed counting". GPU-Z should show the physical clock of the memory chips, whereas twice the amount shown by Afterburner is the data rate on the external bus. Not sure what the number in between actually is.
When one changes the GPU core or memory clock rates for P0 or P2 using the nVidia Inspector process described here, how durable is the change?
1. just while nVidia inspector is still running.
2. until next system reboot?
3. until next system power down?
4. until something actively changes it or actively restores defaults?
5. something else?
To answer the question on durability of the changes: They reset after a reboot.
I increased the the gpu clock to 1392 MHz and the memory clock to 1778 MHz (3556), after that I cannot increase the memory speed , it reverts to those numbers on my card even if I try to go higher on my card after you apply the changes. This is for both gpu and memory in the P2 mode.
All tasks are error free and the time is reduced by another 90 seconds. Power consumption has increased by 2%
The changes remain "until told otherwise", which could be:
- reboot
- driver reset
- different setting from some other software
To apply the settings from nVidia Inspector upon each boot:
- in nVidia Inspector right click on "Create Clocks Shortcut" and choose "Create Clock Startup Task"
- or click "Create Clocks Shortcut" and execute the created link automatically via windows task scheduling
Never looked at memory clock
)
Never looked at memory clock speeds before, but after reading this thread I checked my clock speeds for my GTX 980 running GPUGRID long runs and yep, I only get 1502MHz on my memory.
This explains why GTX 980 isn't faster that 780 Ti in GPUGrid at stock speeds. I was already wondering why I wasn't beating 780 Ti...
I would be interested to know
)
I would be interested to know how much improvement you get on GPU-grid work units. On the Einstein wu I get a 10% improvement, but that's not a lot.
After letting the card run at
)
After letting the card run at the higher memory clock of 1752 MHz, which is really the speced clock, the wu times (running two at a time) for BRP4G have gone from 3220 seconds 2850 seconds average. This is an improvement of 12%. None of them have any errors.
The clock rates can not be increased any further in this P2 mode, if you try they jump right back to the previous maximum.
Update: It seems I was wrong
)
Update:
It seems I was wrong about the memory OC causing my blue screens! Encouraged by your results I tried 3.5 GHz again and it's been running flawless since almost 2 days :)
So the message to get out to GM204 owners is just: Use nVidia Inspector to set 3.5 GHz in P2!
Same here. In fact, I can not set anything above 3.5 GHz for P0 either. It also jumps right back. Tried with Inspector, Precision and Afterburner. I'm confused, as the web is full of reports of memory overclocks.
MrS
Scanning for our furry friends since Jan 2002
Overclocking the
)
Overclocking the memory:
Thanks to skgiven from GPU-Grid I can now overclock my memory! Here's how:
- the GPU must not be crunching BOINC (either pause your GPU project, or all GPUs, or suspend BOINC completely)
- in the nVidia Inspector OC tab set the overclock for P0 (because you can't go any higher than this in P2)
- now you can set up to this memory clock for P2 as well
- apply & have fun
GTX970/980 are reported to reach about 4 GHz memory clock. It should be interesting to see if this can boost Einstein performance another 10%.
MrS
Scanning for our furry friends since Jan 2002
MrS wrote:Thanks to skgiven
)
Thanks for posting that.
I can affirm that I applied this process on my 970 running on a dual-core Haswell under Windows 7.
I did the before-after observations using MSI Afterburner (which has been my default fan speed controller and general GPU observer for some time).
I decided to work up by somewhat small increments.
The initial condition, while running Perseus Arm Survey, as reported by MSI Afterburner, with not changes made by me:
Core clock 1367, Memory clock 3005
After asking Inspector to raise the P2 memory clock by 95, Afterburner reports:
Core clock 1367, Memory clock 3100
I've got pretty heavy process Lasso intervention, and am only running two CPU BOINC tasks on one side of the two physical cores of this HT host, so my elapsed times are usually pretty repeatable. I should be able to tell overnight something about elapsed time improvement.
I just recently turned down the memory clock (from stock) on one of my GTX 660s hoping to reduce the rate of uncommanded downclocks. It may have helped, though the history is erratic enough that I am not confident. If it really did help, and if the 970 has a similar possibility, aside from outright malfunction, bad WU's, or blue screens, uncommanded downclocks might be another limiting consideration to 970 overclocking.
(side comment--in this setup GPU-Z reports the memory clock rate as half that reported by MSI Afterburner--to within a very small error margin)
RE: (side comment--in this
)
Yes, they use different "DDR clock speed counting". GPU-Z should show the physical clock of the memory chips, whereas twice the amount shown by Afterburner is the data rate on the external bus. Not sure what the number in between actually is.
MrS
Scanning for our furry friends since Jan 2002
Question: When one changes
)
Question:
When one changes the GPU core or memory clock rates for P0 or P2 using the nVidia Inspector process described here, how durable is the change?
1. just while nVidia inspector is still running.
2. until next system reboot?
3. until next system power down?
4. until something actively changes it or actively restores defaults?
5. something else?
To answer the question on
)
To answer the question on durability of the changes: They reset after a reboot.
I increased the the gpu clock to 1392 MHz and the memory clock to 1778 MHz (3556), after that I cannot increase the memory speed , it reverts to those numbers on my card even if I try to go higher on my card after you apply the changes. This is for both gpu and memory in the P2 mode.
All tasks are error free and the time is reduced by another 90 seconds. Power consumption has increased by 2%
The changes remain "until
)
The changes remain "until told otherwise", which could be:
- reboot
- driver reset
- different setting from some other software
To apply the settings from nVidia Inspector upon each boot:
- in nVidia Inspector right click on "Create Clocks Shortcut" and choose "Create Clock Startup Task"
- or click "Create Clocks Shortcut" and execute the created link automatically via windows task scheduling
MrS
Scanning for our furry friends since Jan 2002