Is this normal for Pascal cards?

[Nocky24]

The core clock seems to be erratically fluctuating when I run Heaven (or any games for that matter) between around 1936-2040. This is a Gigabyte 1080Ti Gaming OC. My previous card was a 980Ti and the core clock only steadily decreased in steps due to temps, this one seems to be all over the place though and never settles on one clock speed for more than 2 or 3 seconds regardless of temps. The videos I've seen online of Pascal cards seem to suggest they can hold a steady clock throughout Heaven so I'm not sure why this one is any different.

Here's the output after Heaven:

Spoiler

Things I've tried:

DDU and reinstall drivers
Set card to Maximum Performance in Nvidia control panel
Set fan to 100% to discount any temp throttling
Underclocking card and adjusting power sliders

 

[Nocky24]

Had a bit more of a play and I think I may have just been bouncing off the TDP wall. Raised the power limit a bit more now.

 

[foster grant]

I get the same with my 1070 , GPU clocks at times seem to be all over the place , although it only started about 4 weeks ago, so I dunno why it started .

 

[r7slayer]

Set your Power Limit all the way to 120%. I was hitting over 110% on stock boost so your hitting the TDP easily if its still at 100%. Even at 2076 boost its going to fluctuate some what but probably just about stay @ or just over 2ghz.

 

[Rroff]

Set your Power Limit all the way to 120%. I was hitting over 110% on stock boost so your hitting the TDP easily if its still at 100%. Even at 2076 boost its going to fluctuate some what but probably just about stay @ or just over 2ghz.

For me once/if the card goes over 79C I start bouncing off the power limit regardless of any setting of temperature or power target - bit strange as at those points on stock it will drop from 1911MHz to 1860MHz and if I slightly bump the voltage drop from 2100MHz to 202xMHz. If I keep temperatures below 79C the same behaviour doesn't exhibit itself although it will always (even if its just momentarily) drop to 202xMhz at some point due to power reason eventually if pushed hard enough.

Adjusting temperature or power target seems to make zero difference on my card as to its boost behaviour.

I don't think there is one standard behaviour with Pascal cards - the way boost works seems to be much more sensitive to the characteristics of individual cores i.e. I've seen 1-2 people complaining that their cards first thermal throttle at ~32C which no Pascal card I've had access to or own myself has ever done. While others see clock drops at a couple of points around 40C and 60C as well as at the temperature target which again I've not seen on my card and only seen on one other card.

 

[r7slayer]

For me once/if the card goes over 79C I start bouncing off the power limit regardless of any setting of temperature or power target - bit strange as at those points on stock it will drop from 1911MHz to 1860MHz and if I slightly bump the voltage drop from 2100MHz to 202xMHz. If I keep temperatures below 79C the same behaviour doesn't exhibit itself although it will always (even if its just momentarily) drop to 202xMhz at some point due to power reason eventually if pushed hard enough.

Adjusting temperature or power target seems to make zero difference on my card as to its boost behaviour.

I don't think there is one standard behaviour with Pascal cards - the way boost works seems to be much more sensitive to the characteristics of individual cores i.e. I've seen 1-2 people complaining that their cards first thermal throttle at ~32C which no Pascal card I've had access to or own myself has ever done. While others see clock drops at a couple of points around 40C and 60C as well as at the temperature target which again I've not seen on my card and only seen on one other card.

Your post confused me some what. Your card's power limit is not affected by temperature. I can hit the power limit of my card 120% under 40 degrees. I did it with the stock cooler at around 78 degrees. All the power limit does is just restrict the TDP of the card. Talking in regards to say the FE card, say power target is at 100%. The TDP of the card is 250 watts. Raise it to 120% thats is a extra 50 watts right? Which brings it to its theoretical TDP of 300 watts. This makes sense in regards to specs seen as PCI_E can draw 75 watts, 6pin 75 and 8 pin 150 which adds upto 300 watts. Yes i know it can draw more than this but this is what the card is designed as is.
So nvidia designed them to stay within spec which is why 120% power target is all we get. At 2Ghz i can hit this 120% power target limit while under 40 degrees but looking at how often it does it its not that much. Increasing the core frequency upto 2050 it hits it more often and then overclocking the memory seems to make it hit the power limit slightly more. When you start hitting the power limit frequently you will see it power limit throttle. Not temp throttle. It seems to drop slightly differently to temp throttling but it changes quite quickly.

Yes my 1080ti does thermal throttle at 32 degrees i know this because on cold boot of my computer my cards idles at 24-26 degrees so when i load up a game and see it climb over 32 degrees it eventually looses 13Mhz. Not straight away but it does drop. Ive tested this the the FE card too and while i never got as low as 32 degrees with it. I just did some maths and yes its drops 13Mhz every 10 degrees increase. Actually i think it drops a lot more when it climbs past 80 degrees but i never let my FE get that high.

Thing is with how GPU boost 3.0 works with the drops with the temp throttling and then hitting the TDP wall the pascal cards can drop frequency abit. It's why i keep considering doing the shunt mod but i keep umming and arrghing about it. Would only really be worth it for benchmarking sakes. In games probably not even noticeable. Ive found with GPU boost 3.0 you have to find the sweet spot for overclocking. ive left my core at 2050 as even if i increase it to 2100 it drops all the way down to bottom of 2Ghz when it power throttles. in fact it seems to drop a lot more when its overclocked that high. 2050 does seem slightly better. Overclocking memory draws more from your power target pool too so that should be considered when overclocking.

Anyways ive rambled on too much and this is in regards to FE cards. I've seen the PCBs of other AIB partner cards and they have been designed to overcome the TDP problems of the reference design. Only problem they seem to have is still the temp throttling from 32 degrees but also the voltage wall. I'm not sure how the pascal cards react to this seen as i hit my TDP wall before voltage. Im going to assume just general instability.

 

[Rroff]

Your post confused me some what. Your card's power limit is not affected by temperature.

Yeah this is the strange thing - my card sits quite happily at 1911MHz under load until it hits 79C then it drops 13MHz (or down to 1860 depending on load) but the reason in GPU-z is the power limit one not temperature - but it never does it until 79+C.

Same with overclocking it sits at 2100MHz until it hits 79C then same thing starts dropping speed bins with power reason upto 202xMHz depending on load.

 

[r7slayer]

Yeah this is the strange thing - my card sits quite happily at 1911MHz under load until it hits 79C then it drops 13MHz (or down to 1860 depending on load) but the reason in GPU-z is the power limit one not temperature - but it never does it until 79+C.

Same with overclocking it sits at 2100MHz until it hits 79C then same thing starts dropping speed bins with power reason upto 202xMHz depending on load.

That is odd mate to say the least lol. Well if that is true then keep it under 79 degrees and no throttling for you haha

 

[Goat-]

Yes that's normal, GPU boost 3.0 fluctuates the core clock constantly in response to what the temps are.

 

[Rroff]

That is odd mate to say the least lol. Well if that is true then keep it under 79 degrees and no throttling for you haha

Just testing it today it dropped a bin at 61C reason VRel - never did that in the past - same thing at 79C though drops because reason power (and this is where it starts changing clocks proper) not temperature - started it at just below 20C and no drop at 32C.

In most stuff it doesn't go over 71C anyhow - doesn't make much odds to the overclocking as before I still max out at 2100MHz and always at some point drop to 202xMHz due to power at some point regardless of temp just not so often if its sub 79C.

 

[r7slayer]

Weird to say the least. Tbh i think its stupid it dropping clocks from 32 degrees. I mean im watercooling so it does not need to drop any clocks to keep cool but does at 32 degrees. Totally bonkers. Id understand if it started at 62 degrees. Ask kaap hes seen it drop from 32 degrees i beleive. Thing that annoys me is how much it throttles from power limit. Tbh had some gaming in and it doesn't throttle as much as it does in benchmarking.
Ive not seen my card drop from vrel or voltage related? I can load upt gpu z and use the pcie tester render and it will say vrel with that. Even in game menus.

 

[Rroff]

Not saying some don't drop from 32C - just I've seen different behaviour from different cards.

Pascal is pushing the edge of 16nm FF+ to be honest so aggressive power/thermal/voltage management is kind of needed otherwise they'd be shipping cards doing like 1.86GHz max out the box and people would be celebrating 2.1GHz overclocks lol.

 

[r7slayer]

Not saying some don't drop from 32C - just I've seen different behaviour from different cards.

Pascal is pushing the edge of 16nm FF+ to be honest so aggressive power/thermal/voltage management is kind of needed otherwise they'd be shipping cards doing like 1.86GHz max out the box and people would be celebrating 2.1GHz overclocks lol.

Thats whats bugging me. I think i could get 2100mhz stable. Just the power limit that restricts me. Throttles hard when it gets a high gpu utlisation. I like the idea of gpu boost 3.0 and it works well. Just needs refining. Cards do not need to throttle at low temps at all. And i don't know why they dont allow use of full tdp of the card. Why you have to set 120% is daft.

 

[Gripen90]

Have you adjusted the temperature limit to 85°c to allow for a more stable boost clock ?