So I've notice something funny with my 1080

Normally it runs about 1.89GHz with it's own boost and no need adjust anything.

But recently I've noticed the speed has dropped to 1.65GHz

So to remedy this, I've had to adjust the clock offset on the Precision X to push the speed back to to 1.9GHz

It's holding for now, but I wanted to see if anyone else has seen this.

GPU temp is holding at 41C ( I hybrid this GPU a while back)

On Beta my GTX 1070 was dropping it's clock speed down due to all the noisy WUs going through. Most of them were finishing before they had a chance to even start on the GPU; every now & then one would last a bit longer & start on the GPU- the clock speed would pick up again, then the WU would finish a few seconds later & the clock would drop down again.
Things seem to have settled down there, for now.


EDIT- while running GPU-Z it's PerfCap Reason would change from VRel (Voltage Reliability) to Util (Limited by GPU utilisation).

---

Grant
Darwin NT

This GPU is over here on Main, and the work units appear to be normal.

So I can't figure out why it's doing that.

Work is validating but just running slower than normal

Work is validating but just running slower than normal

Does GPU-Z give any indication?

You've tried the usual fix (reboot?)

---

Grant
Darwin NT

Yes I reboot it and it was still low.

Going to download GPU-Z and see what it says.

I tend to run SIV64X

edit..

Downloaded and running but not really seeing anything unusual, other than the speed.

The speed is up with my adjustment to the clock offset.

I have another downlock observation with GTX1080; memory clock goes down around 10% as soon as SETI or any other pure GPU compute starts, card is pushed into P2 mode instead of P0 (max performance mode). Win10x64 and 368.81 drivers. Why nVidia is doing this, it is unknown...

No any issues with games, all are running perfectly on P0, is max. performance mode and normal memory and boost GPU clocks.

Seems like nVidia driver as soon pure compute load is detected (no graphics output), cards are pushed to P2 mode.

Some others also found this, and happens also with Maxwell and recent drivers as well...
https://devtalk.nvidia.com/default/topic/940304/cuda-programming-and-performance/grim-memory-bandwidth-gtx-1080/1

There is some workaround but it doesn't work on Pascal, only on Maxwell....
https://devtalk.nvidia.com/default/topic/892842/one-weird-trick-to-get-a-maxwell-v2-gpu-to-reach-its-max-memory-clock-/

---

Downloaded and running but not really seeing anything unusual, other than the speed.

The speed is up with my adjustment to the clock offset.

If you cancel the overclock & reset everything back to the defaults, does the GPU still drop back down to the lower speed?
If so, what does GPU-Z display for PerfCap Reason?

---

Grant
Darwin NT

I reset the GPU, but it did not go back down to the lower speed.

It did drop down to 1.88Ghz, which is what it was before all this.

This is the second time I've seen it drop down to the 1.65.

It had done that last week as well but then I turned off that machine for a week before restarting it Friday.

The PerfCap Reason onlys says VRel (reliable voltage)and a solid blue bar but it's speed is back to what it used to be.

I'll keep an eye on it and if it drops again, I'll check the PerfCap Reason again. I didn't know that was on there, nice thing to have so I can check it next time

Thanks Grant

I have another downlock observation with GTX1080; memory clock goes down around 10% as soon as SETI or any other pure GPU compute starts, card is pushed into P2 mode instead of P0 (max performance mode). Win10x64 and 368.81 drivers. Why nVidia is doing this, it is unknown...

No any issues with games, all are running perfectly on P0, is max. performance mode and normal memory and boost GPU clocks.

Seems like nVidia driver as soon pure compute load is detected (no graphics output), cards are pushed to P2 mode.

Some others also found this, and happens also with Maxwell and recent drivers as well...
https://devtalk.nvidia.com/default/topic/940304/cuda-programming-and-performance/grim-memory-bandwidth-gtx-1080/1

There is some workaround but it doesn't work on Pascal, only on Maxwell....
https://devtalk.nvidia.com/default/topic/892842/one-weird-trick-to-get-a-maxwell-v2-gpu-to-reach-its-max-memory-clock-/

M_M

This is interesting reading. Only been thru about the first 2 pages but it does seem to point to an issue, don't know if it is directly related to mine yet but good knowledge to have, thanks

As far as P2 states, that is built in by Nvidia to slow down the cards for Scientific calculations.

Last year they commented and said if the cards was run in a P0 state, that there might be errors in calculating, so that is why they lowered it to P2 states.

Edit...

Hmm, finished reading all the threads.. You might be onto some there there M_M

I'll have to keep an eye on that card.

Thanks

The PerfCap Reason onlys says VRel (reliable voltage)and a solid blue bar but it's speed is back to what it used to be.

I'll keep an eye on it and if it drops again, I'll check the PerfCap Reason again.

If it remains VRel when the speed drops again, then i'd say it's related to what M_M posted about- deliberate throttling by the driver.
A slight possibly if it becomes VOp (Voltage Output), the driver is limiting the voltage, so the clock speed is dropping. Or issues with the onboard regulators.


Would be worth noting what the VDDC voltage is while it's running OK. When my GPU loads drops off & the clock speed goes down with it, so does the VDDC voltage.

---

Grant
Darwin NT

Hi,
I'm running on P2 state but with my driver version from June the settings for P3 are copied to P2. So +200 for gpu and memclock to 11000.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

my only observation based on a GTX 1070 FE that stays at 1822-1867MHz on the GPU while doing GPU work units all day (I haven't done CPU work units) is your best off assigning plenty of access to the CPU for cuda42/50 and SoG/sah tasks.

i.e. I use <cpu_usage>1</cpu_usage> as even though a single cuda42/50 and SoG/sah task all peak out at ~11% CPU load they still need access to 4 or more CPU cores to run efficiently, so if your avg_ncpus is low like 0.04, 0.1 or even 0.5 the GPU MHz will drop.

---

@memory downclock issue: I think nVidia should clearly state a downclock reason and provide option so users can decide, i.e. for example, in control panel to give us the option would we use P2 for number crunching (could be a default setting) or P0 (with some exclamation that calculation accuracy could not be guaranteed, if this is the reason).

As I can see, for SETI calculation, memory controller load is between 50-70%, so I suspect memory clock matters and makes a difference in calculation efficiency.

---

Grant,

it did it again.

Took the speed down to 709MHz

PerfCap Reason said Pwr (power) solid green line

VDDC said 0.7930 V

I had to use precision X to raise it back up to 1974 MHz
PerfCap Reason now says Reliability Voltate Blue bar
VDDC says 1.062 V


I can't understand why it would say power. That machine has a EVGA 1600W in there,

There are 3 980TI hybrids and 1 1080 Hybrid.

I got the requirements for that system from EVGA's own powermeter, so it's supposed to be more than enough, and given that the 1080 uses less than the 980Ti it replaced, it should be even less now.


Looking at SIV64X

it say PSU +5.04 +12.10 +3.26 +3.41 VBAT +3.33

It's hooked into a UPS rated for 2000 Watts

Anyone with any ideas?

"power" in this context is the on-board power management not the available power.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Another to consider is, "what else is running on this electrical circuit?", as I all too well know that certain other electrical equipment can cause very bad effects to happen on it.

Cheers.

Hey Wiggo.

That was a concern of mine as well but I have this out in a separate building behind my house with only the AC on in there.

So there really shouldn't be anything on that circuit besides the computer and UPS.

I checked all the other GPUs in that machine and they are running as they are supposed to, no downclocking.

I'm with Rob.
The power limit is the board's power limit, not the computer power supply to the card.

Why would it think that it's hit that limit?
It could relate to the issue M_M mentioned, what is odd is that it doesn't boost up again after the GPU load has dropped off then risen again, or even after a re-boot.


Might be worth getting a copy of NVidia Inspector, it will allow you to see what P state (Power state) the card is in.
Nvidia P states.
•P0/P1 - Maximum 3D performance
•P2/P3 - Balanced 3D performance-power
•P8 - Basic HD video playback
•P10 - DVD playback
•P12 - Minimum idle power consumption

On older hardware there have been times the driver stopped a card from boosting. But haven't seen one where it can boost (sometimes).


Only other thought that comes to mind is Nvidia control panel, Manage 3D settings, Power Management mode & set it to "Prefer Maximum performance" (from Adaptive).

---

Grant
Darwin NT

I posted this early last month, could this be the problem you are experiencing.

GPU FLOPS: Theory vs Reality - msg 1800841

I posted this early last month, could this be the problem you are experiencing.

GPU FLOPS: Theory vs Reality - msg 1800841

Maybe, but if it's anything like my card then even with 3WUs at a time it's power consumption rarely gets up to 65% of it's maximum, let alone above it.

---

Grant
Darwin NT