Hitting Power Limit Early?

[Mike Soda]

While attempting a +160 on the core of my 3GB MSI GTX 1060 Gaming X; even though power & voltage limit were set to max, it still somehow hit a power limit of 1 & throttled down to 1800 something after a few hours of Heaven. It was only running at 2060 something & fans maxed at about 60, same with temp. I have Nvidia control panel set to prefer maximum performance & Windows 10 set to similar too. Is there something faulty with my card or did I just lose the silicon lottery really bad?

[LooneyJuice]

Just sounds like normal GPU Boost 3.0 behavior. Core clock will rarely stick to whatever it starts running at. As a side note though, have you maxed your temp target? That can usually flatten the curve out quite a bit, especially if the cooler keeps the card below the temp threshold anyway. Also just as a precaution, make sure it's not running too toasty.

[ARikozuM]

I hate when that happens. You're playing, all is well and good, but then you hit your power limit... 

 

I would check if the memory clock is able to keep up. What are your temps?

[Mike Soda]

8 minutes ago, LooneyJuice said:

Just sounds like normal GPU Boost 3.0 behavior. Core clock will rarely stick to whatever it starts running at. As a side note though, have you maxed your temp target? That can usually flatten the curve out quite a bit, especially if the cooler keeps the card below the temp threshold anyway. Also just as a precaution, make sure it's not running too toasty.

Temperature limit I set to 88 but the card only peaked at 62 briefly, usually stayed between 58 & 60. I was under the impression though that almost any Pascal GPU could easily hit 2100Mhz but I'm running into throttling before that =\.

[Mike Soda]

1 minute ago, ARikozuM said:

I hate when that happens. You're playing, all is well and good, but then you hit your power limit... 

 

I would check if the memory clock is able to keep up. What are your temps?

Between 58 & 60C, peaked at 62 & I haven't overclocked memory at all yet. Was going to do Core first, find the stable of that then do memory.

Edited June 9, 2017 by Mike Soda
capitalized my beginning B

[ARikozuM]

3 minutes ago, Mike Soda said:

Between 58 & 60C, peaked at 62 & I haven't overclocked memory at all yet. Was going to do Core first, find the stable of that then do memory.

You had better capitalize that "b", sir!

 

I would see if you can raise the power limit further. If you're at max, I'd back off the core and test again.

[Mike Soda]

Just now, ARikozuM said:

You had better capitalize that "b", sir!

 

I would see if you can raise the power limit further. If you're at max, I'd back off the core and test again.

Power limit was set to 108% & voltage to 100%, both as high as they will go. Since it took about 3-4 hours for that to even show up I'll have to try decreasing it by 20 tomorrow & let ya know how it goes. Do you know if what I'm experiencing is common for this card in particular or is it like I said before in that I just lost the lottery?

[ARikozuM]

I don't know to be honest. I don't know much about the 1060 cards. If the voltage went as high as it could, but could only keep 1800MHz stable then that might be GPU Boost holding it back.

[LooneyJuice]

26 minutes ago, Mike Soda said:

Temperature limit I set to 88 but the card only peaked at 62 briefly, usually stayed between 58 & 60. I was under the impression though that almost any Pascal GPU could easily hit 2100Mhz but I'm running into throttling before that =\.

Well, contrary to what a lot of the promo material says, that 2100mhz still is largely silicon lottery (as you cannot technically overvolt any Pascal card). Some do that, very few do slightly over, and lots do quite a bit under. Additionally, I have heard some weird instances of maybe somewhat unstable overclocks fluctuating madly rather than flat out crashing. Furthermore, all those lovely 2000mhz+ numbers we see everywhere are more prevalent during the first 5 mins or so of any benchmark, and then they drop substantially. Regardless of all of the above, the gains - or losses for that matter - from these sorts of marginal overclocks are pretty much imperceptible, so I wouldn't worry much if I were you. Unless you encounter a problem, like a hardware malfunction of sorts, it's all business as usual.

[Mike Soda]

32 minutes ago, ARikozuM said:

I don't know to be honest. I don't know much about the 1060 cards. If the voltage went as high as it could, but could only keep 1800MHz stable then that might be GPU Boost holding it back.

The weird part is even though the power limit hit 1, my power percentage used never got above 80ish percent.

21 minutes ago, LooneyJuice said:

Well, contrary to what a lot of the promo material says, that 2100mhz still is largely silicon lottery (as you cannot technically overvolt any Pascal card). Some do that, very few do slightly over, and lots do quite a bit under. Additionally, I have heard some weird instances of maybe somewhat unstable overclocks fluctuating madly rather than flat out crashing. Furthermore, all those lovely 2000mhz+ numbers we see everywhere are more prevalent during the first 5 mins or so of any benchmark, and then they drop substantially. Regardless of all of the above, the gains - or losses for that matter - from these sorts of marginal overclocks are pretty much imperceptible, so I wouldn't worry much if I were you. Unless you encounter a problem, like a hardware malfunction of sorts, it's all business as usual.

Thank you both & I'll report back later tomorrow with whatever Core OC I'm able to get without hitting that limit.

[xAcid9]

Try undervolting the card in MSI AB frequency/voltage curve.

[Mike Soda]

11 hours ago, xAcid9 said:

Try undervolting the card in MSI AB frequency/voltage curve.

I've been using Afterburner for weeks & it's taken me until now to even find that menu because the graph icon was so tiny lol. I'm not sure how to undervolt it though & how would that help? Here's what it looks like & I'll be trying a +140 on the core for the next 5 hours now so will check back when it's done or if power limit is hit again before. Oh & also I changed the setting next to Unlock Voltage Control from Standard MSI to Extended MSI, not sure if that'll help but we'll see.

[LooneyJuice]

6 minutes ago, Mike Soda said:

I've been using Afterburner for weeks & it's taken me until now to even find that menu because the graph icon was so tiny lol. I'm not sure how to undervolt it though & how would that help? Here's what it looks like & I'll be trying a +140 on the core for the next 5 hours now so will check back when it's done or if power limit is hit again before. Oh & also I changed the setting next to Unlock Voltage Control from Standard MSI to Extended MSI, not sure if that'll help but we'll see.

 

 

The extended voltage control shouldn't do anything. That's for card variants such as the MSI lightning edition cards which at least usually had more voltage to play with. In this case, whether you select it or not, you'll just have the standard Pascal voltage control. I'm not joking when I'm saying you really have no proper control over voltage on Pascal. In fact, all the voltage slider does for the most part is that it makes the card use the predetermined max voltage sooner. So don't fret over voltage, just max the slider (it's more a placebo anyway) and tweak around with your curve.

[xAcid9]

19 minutes ago, Mike Soda said:

I've been using Afterburner for weeks & it's taken me until now to even find that menu because the graph icon was so tiny lol. I'm not sure how to undervolt it though & how would that help? Here's what it looks like & I'll be trying a +140 on the core for the next 5 hours now so will check back when it's done or if power limit is hit again before. Oh & also I changed the setting next to Unlock Voltage Control from Standard MSI to Extended MSI, not sure if that'll help but we'll see.

-snip-

Uhh.. lover voltage mean it use less power and probably not hitting power limit sooner or at all? Not sure how effective this is since i don't have Pascal card to play with.

 

[LooneyJuice]

1 hour ago, xAcid9 said:

Uhh.. lover voltage mean it use less power and probably not hitting power limit sooner or at all? Not sure how effective this is since i don't have Pascal card to play with.

-snip-

The thing is that Buildzoid did it with a heavily modded card and manual voltage control. I watched his crazy 1070 livestream a few days ago, and because he broke the stock voltage regulation and such, he found that "his" flat 2000mhz ran way faster than what GPU Boost said was 2000mhz. He also touched on the update rate of the core clock and power regulation being way faster than the polling rate on any software monitoring tool. He made the hypothesis that under GPU Boost 3.0, the numbers were just eye-candy, and that the card underclocked itself as soon as it had to do any work (but very very fast) and overclocked itself again during small intermissions like texture loading and whatnot.

 

This isn't to say that undervolting and such may not provide any benefits for stock cards, as the OP mentions. But, I do think it's more a matter of finding your max clock based on actual bench numbers rather than high frequency this time around. Reason being, core clocks were a big thing during the Pascal marketing campaign, and GPU boost may be trying to maintain an apparent high clock (at a fast enough update rate to not be seen fluctuating) rather than drop to where the card should be running. Also touched on it prior.

13 hours ago, LooneyJuice said:

 Additionally, I have heard some weird instances of maybe somewhat unstable overclocks fluctuating madly rather than flat out crashing.

So, it's a pain in the butt, but it looks like the proper way to do it is by pure bench performance rather than clocks, sadly.

[Mike Soda]

6 hours ago, LooneyJuice said:

 

The extended voltage control shouldn't do anything. That's for card variants such as the MSI lightning edition cards which at least usually had more voltage to play with. In this case, whether you select it or not, you'll just have the standard Pascal voltage control. I'm not joking when I'm saying you really have no proper control over voltage on Pascal. In fact, all the voltage slider does for the most part is that it makes the card use the predetermined max voltage sooner. So don't fret over voltage, just max the slider (it's more a placebo anyway) and tweak around with your curve.

I'll put it back to standard then & I have some good news. Just tested +140 on the Core for 5 hours in Heaven, started out at 2076, maxed at 2088 but settled at 2050 for the duration of the test never hitting power limit. Power percent used never went above 83% & voltage peaked at 1.093. Memory was at it's default 4007 (8040) while both temp & fan stayed between 58 & 62.

6 hours ago, xAcid9 said:

Uhh.. lover voltage mean it use less power and probably not hitting power limit sooner or at all? Not sure how effective this is since i don't have Pascal card to play with.

 

My bad I didn't even think of it that way but good idea, maybe the card is accidentally over-volting itself when it doesn't need to thus hitting that power limit early. I'll just try lowering voltage limit to 75% when I have the time to test more next week. A couple things I gotta ask both of you though. I'm currently using RivaTuner 6.5.0.8709 but there's a 6.6 & even 7.0 beta versions available from Guru3d. Should I get one of those & which? I tried installing 6.6 but Windows claimed unknown publisher so I canceled that until I know for sure it's just a bug & not anything malicious. Also my temp limit is currently set to 88, should I lower it back to 83, keep at 88, max out or does it not matter since I've been in low 60's?

Edited June 9, 2017 by Mike Soda
changed one to a couple :S

[Mike Soda]

On 6/9/2017 at 10:27 AM, xAcid9 said:

Uhh.. lover voltage mean it use less power and probably not hitting power limit sooner or at all? Not sure how effective this is since i don't have Pascal card to play with.

 

 

On 6/9/2017 at 10:16 AM, LooneyJuice said:

 

The extended voltage control shouldn't do anything. That's for card variants such as the MSI lightning edition cards which at least usually had more voltage to play with. In this case, whether you select it or not, you'll just have the standard Pascal voltage control. I'm not joking when I'm saying you really have no proper control over voltage on Pascal. In fact, all the voltage slider does for the most part is that it makes the card use the predetermined max voltage sooner. So don't fret over voltage, just max the slider (it's more a placebo anyway) and tweak around with your curve.

Okay for both of you, I just ran a test for several hours with +155 on the core & Voltage limit set to max as it wouldn't keep it the 2063 clock without it. It ran fine for 4 hours until suddenly dropping voltage even though it did not hit power limit. Does this mean my card's voltage limiter is faulty or could it be because Windows Defender decided to run a quick scan sometime today?

[LooneyJuice]

1 minute ago, Mike Soda said:

 

Okay for both of you, I just ran a test for several hours with +155 on the core & Voltage limit set to max as it wouldn't keep it the 2063 clock without it. It ran fine for 4 hours until suddenly dropping voltage even though it did not hit power limit. Does this mean my card's voltage limiter is faulty or could it be because Windows Defender decided to run a quick scan sometime today? -snip-

Impeccable timing as I was passing through the forum, lol. Well, look at what your GPU usage did during the dip. It dropped, and so did the voltage as the card momentarily wasn't doing anything. Yes, probably a hitch or background application such as Defender doing something 'cause then your voltage/usage maintained its previous state right after. There's really nothing to worry about unless you encounter massive performance drops due to an unstable OC or something obviously wrong with the card (like running below posted base/boost clocks or something).

[Mike Soda]

1 minute ago, LooneyJuice said:

Impeccable timing as I was passing through the forum, lol. Well, look at what your GPU usage did during the dip. It dropped, and so did the voltage as the card momentarily wasn't doing anything. Yes, probably a hitch or background application such as Defender doing something 'cause then your voltage/usage maintained its previous state right after. There's really nothing to worry about unless you encounter massive performance drops due to an unstable OC or something obviously wrong with the card (like running below posted base/boost clocks or something).

Phew okay thank you, should I try running it again tomorrow? Is there a way I can prevent Defender from doing that again? How long should I test for at max?

 

Sorry so many questions

[LooneyJuice]

7 hours ago, Mike Soda said:

Phew okay thank you, should I try running it again tomorrow? Is there a way I can prevent Defender from doing that again? How long should I test for at max?

 

Sorry so many questions

I just don't get the concern is all. If it benches on par (within a few points) with similar GPUs, if you don't encounter any massive hitching, or you're getting performance comparable to game bench framerates, there's nothing wrong with the card. Core frequency is never guaranteed, varies with every individual GPU and too high overclocks can result in performance instability with Pascal, not just crashes. Not to mention that you can monitor all your telemetry with an On Screen Display in real time with Afterburner, from voltage to power overall power usage. So if ever you encounter issues, you can troubleshoot like that. There's just nothing wrong with the card, you don't need to test any more. Just don't be overzealous with frequencies. Review samples are a very small percentage of the number of GPUs on the market, so don't always aim for those numbers.

[done12many2]

7 hours ago, Mike Soda said:

Phew okay thank you, should I try running it again tomorrow? Is there a way I can prevent Defender from doing that again? How long should I test for at max?

 

Sorry so many questions

 

I have seen you mention that you were hitting power limits a few times in this thread.  Can you show us what makes you believe that you are power limiting?

 

Getting clock speeds to hold with Pascal are just about temps.  That's pretty much it until you get into the bigger cards like the Titan or Ti series.  

 

Keep a Pascal card cool and it will hold clocks.  There are several temperature throttle points along the way within Pascals safe temp operating range.  Hit one and it will automatically scale down one or more bins.  There's nothing that you can do about that other than cool the card better.

 

You may be throttling, but I suspect it's temperature related and not power limited.  

[Mike Soda]

5 hours ago, LooneyJuice said:

I just don't get the concern is all. If it benches on par (within a few points) with similar GPUs, if you don't encounter any massive hitching, or you're getting performance comparable to game bench framerates, there's nothing wrong with the card. Core frequency is never guaranteed, varies with every individual GPU and too high overclocks can result in performance instability with Pascal, not just crashes. Not to mention that you can monitor all your telemetry with an On Screen Display in real time with Afterburner, from voltage to power overall power usage. So if ever you encounter issues, you can troubleshoot like that. There's just nothing wrong with the card, you don't need to test any more. Just don't be overzealous with frequencies. Review samples are a very small percentage of the number of GPUs on the market, so don't always aim for those numbers.

Thank you so much again, I'll stop worrying about it.

5 hours ago, done12many2 said:

 

I have seen you mention that you were hitting power limits a few times in this thread.  Can you show us what makes you believe that you are power limiting?

 

Getting clock speeds to hold with Pascal are just about temps.  That's pretty much it until you get into the bigger cards like the Titan or Ti series.  

 

Keep a Pascal card cool and it will hold clocks.  There are several temperature throttle points along the way within Pascals safe temp operating range.  Hit one and it will automatically scale down one or more bins.  There's nothing that you can do about that other than cool the card better.

 

You may be throttling, but I suspect it's temperature related and not power limited.  

The times when I did hit power limit in the past was when the power limit graph which has a value of 0 or 1, hit 1 & frequency throttled down to almost stock. It wasn't temperature because that never got above 62 with fan always matching. Although yesterday when I retried that same +160 with voltage & power percentages maxed out it acted differently. Rather than trying to hold the 2088 clock speed & hit power limit after a few hours like prior, it just downclocked to 2063. So that's why I decided on testing +155 for the few hours I did as 2063 seems to be the absolute max. If you're interested though I can rerun that +160 to see if I can replicate that hitting power limit again as I'll be gone all day tomorrow like today so time ain't an issue.