Kepler/NVidia: Power Control/Dynamic OC etc. is extremely annoying

[flexy]

After many years having gone through a ton of AT/Nvidia cards as so many others I finally got a used GTX 660 TI.

I already read up plenty on Bios tweaking, overclocking etc. and I am not really a dummy when it comes to that.

The way the "newer" cards overclock I can only call as entirely annoying.

I remember a time where all we had was two sliders, core/memory, you did some testing with various benchmark programs, found a sweet spot for overclocking and THIS WAS ALL there was to it. You KNEW your card ran at X core and Y memory, regardless what you threw at it.

Overclocking today is nothing but a pain in the a$$ with the power control setting and then dynamic "keep the card within TDP" what I can only call pseudo overclocking.

The last few days I did a lot of testing and benchmarking and the TDP is constantly fluctuating, so I thought if I set in Nvidia CP to "prefer performance" I can prevent that - but NO.

One benchmark the effing card is running at 1202 or better, the next it is only at 1032.

With a benchmark where you would expect maximum TDP (say, furmark or OCCT)....it runs entirely contradictionary and only uses 80% of power and lower core.

I also observed in Tomb Raider that it sometimes runs as it should as 1202 core and 100% ish TDP, but then all of a sudden for some reason goes down to 80% TDP and clocks down. (heat????)

(Mind you this is with an unlocked BIOS where I expanded power control and wattage tolerance, the card should NOT throttle).

For me, the constant not knowing at what TDP and actual clock the card is running is VERY annoying and I am (almost) considering re-selling the card and getting an ATI card instead. I did not get a "new" card just to see that it randomly changes core/TDP...as in Tomb Raider, especially if you see it slightly stuttering in some cut sequences (2xSSAA, TressFX on etc.) and I would need every little ounce of performance the card has.

I seriously, seriously would just want to get an ATI card again BECAUSE of that, at least I would know that my card runs stable at a given clock and then never ever having to bother about the stupid way Nvidia did their pseudo overclocking now. (We all know it's only a protection that people cannot overclock to reach the next best card's performance, it's NOT something to the benefit of the enthusiast user in the slightest).

The argument that throttling/power control etc. is implemented to "save power" doesn't fly for me either.

 

[BrightCandle]

In some ways its a decent advance that now the cards will defend themselves from thermal and wattage overload so the cards don't get damaged after being pushed beyond design. Thermal and electrical limits are now more regularly seen, allowing a card to get closer to its maximum rather than being artificially limited in all scenarios or potentially damaged in others.

Ultimately CPUs do the same thing and while it complicates overclocking it is neitherless beneficial. It definitely saves power, considerable power over the population and it also allows the system to get to more performance when parallel usage is lower.

That isn't to say that the current form of what NVidia has is necessarily good or predictable, but lets not forget that AMD also has a version of this which is implemented in software. The AMD implementation has had a lot of bugs over the last year, many of which have meant games only see 300Mhz downclocked cores instead of the full speed. The overclocking grass is a little greener on the AMD side of the fence but the dynamic clocking is also less sophisticated.

No one has a perfect power saving solution today, Intel's CPUs loose performance from the dynamic clocking as they too aggressively return to sleep when under use and AMDs cards dont ramp up enough under certain circumstances. Power saving techniques have been a problem ever since they were introduced. If you overclock an Intel CPU today with offset voltage you don't just care about all cores loaded at top clocks, you also care about 3 cores, and 2 cores and 1 core and all the intervening clock speeds from idle up to the maximum clocks, they all have to stable.

 

[bystander36]

Btw, you can adjust how much over the TDW it can go with MSI afterburner or Precession X. It won't fix everything, but it will allow you to draw more power before it down clocks. It is a percent slider.

Also, some of the down clocking is due to heat. A more aggressive fan profile might help you.

And as mentioned above, AMD also does these things, so you can't avoid it completely, unless there is a BIOS flash that removes it. It's a new style that seems to be here to stay.

 

[willomz]

Who cares what clocks it is running at as long as your frame rates are OK?

Sometimes it downclocks because there is a bottleneck elsewhere, memory perhaps.

If you find the fluctuations annoying just don't look!

 

[SirPauly]

Consider GPU boost efficient and smart, which allows more performance in thermal and power envelopes.

 

[wand3r3r]

The way the "newer" cards overclock I can only call as entirely annoying.

I remember a time where all we had was two sliders, core/memory, you did some testing with various benchmark programs, found a sweet spot for overclocking and THIS WAS ALL there was to it. You KNEW your card ran at X core and Y memory, regardless what you threw at it.

Overclocking today is nothing but a pain in the a$$ with the power control setting and then dynamic "keep the card within TDP" what I can only call pseudo overclocking.

I also observed in Tomb Raider that it sometimes runs as it should as 1202 core and 100% ish TDP, but then all of a sudden for some reason goes down to 80% TDP and clocks down. (heat????)

For me, the constant not knowing at what TDP and actual clock the card is running is VERY annoying and I am (almost) considering re-selling the card and getting an ATI card instead. I did not get a "new" card just to see that it randomly changes core/TDP...as in Tomb Raider, especially if you see it slightly stuttering in some cut sequences (2xSSAA, TressFX on etc.) and I would need every little ounce of performance the card has.

I seriously, seriously would just want to get an ATI card again BECAUSE of that, at least I would know that my card runs stable at a given clock and then never ever having to bother about the stupid way Nvidia did their pseudo overclocking now. (We all know it's only a protection that people cannot overclock to reach the next best card's performance, it's NOT something to the benefit of the enthusiast user in the slightest).

The argument that throttling/power control etc. is implemented to "save power" doesn't fly for me either.

Yeah they came up with the new scheme to tie it to TDP and temperature, but in the end the NV cards are simply neutered as you can't touch the voltage to have the final control over the end result. (So are some non-reference AMD models so you need to research them too )

If you don't OC or OV they probably don't matter as much but even still as the titan showed they are tricky and start downclocking after warming up. (I don't think this has even been fixed by NV yet)

I agree the card should not be downclocking on it's own during intense scenes. It's actually ridiculous.

As for the arguments about protecting the card, yeah that's probably what it is. They don't want to have any RMAs so they neuter them. They could just build them tougher especially since they have raised the price bracket by nearly double this generation.

 

[Nintendesert]

Yeah they came up with the new scheme to tie it to TDP and temperature, but in the end the NV cards are simply neutered as you can't touch the voltage to have the final control over the end result. (So are some non-reference AMD models so you need to research them too )

If you don't OC or OV they probably don't matter as much but even still as the titan showed they are tricky and start downclocking after warming up. (I don't think this has even been fixed by NV yet)

I agree the card should not be downclocking on it's own during intense scenes. It's actually ridiculous.

As for the arguments about protecting the card, yeah that's probably what it is. They don't want to have any RMAs so they neuter them. They could just build them tougher especially since they have raised the price bracket by nearly double this generation.

You can thank all the dishonest customers that overclock their cards, burn them up and then RMA them. There are tons of threads, even on this site which denounces such fraud, where people recommend or state they are going to RMA after a bad overclock.

There's no way to build a "tougher" card for some of the stupidity that goes on with noob overclockers. Read the CPU forums sometime and see how much voltage people run through their chips.

 

[Lepton87]

Let's not forget the titan with its GPU boost 2.0 which is different then other kepler cards.
GPU clock is 837MHz boost is 876 and the card very often reaches over 1GHz without overclocking. Overclocking this card is extremely annoying and unpredictable. It boosts very well when it does not need to and clocks down when performance is needed. What a mess. And you can only increase the tdp by 6%. Pathetic for a card that cost as much as an entire reasonable computer.

 

[BrightCandle]

The amount of over clocking is not a factor of what we want. Its a factor based on what the final silicon can do. If nvidia could build these cards without dynamic clocks I have no doubt they would, but they are up against thermal and electrical limitations, the laws of physics and the behavior of their customers. You can always physically mod your card to push unsafe voltages through it, that is always an option.

Titan would clock higher if it could, but its a big powerful die and it gets hot, so rather than burn up it decreases its performance. Its annoying for the user but much less annoying than the card exploding.

 

[Xarick]

I like the downclocking because it saves power and heat. But don't like dynamic boost because I have had two 670s and one pushed 1175 and one pushed 1136. 40mhz difference on the same brand card.

 

[flexy]

They are "up against thermal and electric limitations" for DECADES already, ever since the early days when I got my Riva 128. This is what OCing is, the "enthusiast" and his battle with thermal/electrical limitations.

Nothing has changed there.

This overclocking charade-farce/thing is NOTHING but a economical measure for the benefit of NV ONLY.

I like gaming, but here is the deal:

90% of the time i am NOT gaming since I am also using my PC for other stuff. The power saving can happen when idle but not fricking WHILE YOU PLAY.

Right now I am playing around with this card and I am seeing the card ITSELF clocks itself down (without any valid reason on occasion), I see 1202 on the core in one run of OCCT/Kombustor and the next time the card MAGICALLY clocked down to 1032 which is a whopping 17% on the core which NV literally takes away from me due to their stupid throttling.

Then it stays at 1032, even if temps drop again. Someone explain that!

NV is taking away all the control from the user/enthusiast and basically tells them to ##### ###. This is how it is.

I was very hyped getting that card but now I am about ready to throw it out due to that overclocking BS , a card which decides on its own at what performance it runs.

 

[Arzachel]

Hey, the benches look better so all is fine. It's not like you needed the performance for longer than it takes to run a bench after a cold boot, who the hell actually plays games anyway?

I didn't mind that much, when "reviewers specials" only affected power and noise metrics a la the factory overclocked GTX 460 that ran quite a bit cooler in the review than for people actually owning the thing, but when there is tangible performance difference not only from card to card but for how long the card's been under load, something is not right.

I mean, if you don't mix stock and overclocked benchmarks, why bench on anything higher than the official approved minimums? It's manufacturer approved overclocking with all the limitations of regular overclocking and then some, why make an exception?

 

[SirPauly]

They are "up against thermal and electric limitations" for DECADES already, ever since the early days when I got my Riva 128. This is what OCing is, the "enthusiast" and his battle with thermal/electrical limitations.

Nothing has changed there.

This overclocking charade-farce/thing is NOTHING but a economical measure for the benefit of NV ONLY.

I like gaming, but here is the deal:

90% of the time i am NOT gaming since I am also using my PC for other stuff. The power saving can happen when idle but not fricking WHILE YOU PLAY.

Right now I am playing around with this card and I am seeing the card ITSELF clocks itself down (without any valid reason on occasion), I see 1202 on the core in one run of OCCT/Kombustor and the next time the card MAGICALLY clocked down to 1032 which is a whopping 17% on the core which NV literally takes away from me due to their stupid throttling.

Then it stays at 1032, even if temps drop again. Someone explain that!

NV is taking away all the control from the user/enthusiast and basically tells them to ##### ###. This is how it is.

I was very hyped getting that card but now I am about ready to throw it out due to that overclocking BS , a card which decides on its own at what performance it runs.

Can you offer some saved examples of the throttling you don't like?

 

[notty22]

^^^
Can't make you like the software. I do. Also, those synthetics, occp and kornbuster trigger throttling. It's design. I have run them, and also games. I suggest Valley or Heaven, which loops continuous to watch the clocks/ TDP % / gpu %. I get correct performance.

 

[Wall Street]

If you don't like the current overclocking methods you can thank all of those who overvolted, overclocked, ran Furmark and then RMAed their cards.

As a rule of thumb, don't worry about Furmark/OCCT/Kombustor, they are not real apps.

 

[MrK6]

Welcome to the dumbing down of technology for the common idiot. It happened with operating systems, then video games, now it's video cards. Nvidia is especially good at it. I'd say vote with your wallet and don't buy the crap; at least the OP bought his card used.

 

[ICDP]

I have owned or tested 4 GTX680s, 3 of which were reference designs with no overclock advertised. All 3 reference cards had very different boost clocks that were the default 3D clocks.

EVGA 2GB
Boost clock speed. 1120

Gainward 2GB
Boost clock speed: 1136

KFA 2GB
Boost clock speed: 1086

The Gainward GTX 680 was around 4% faster than the KFA one. When released very few reference GTX680 reviews showed the actual boost clocks the cards were running at. Of course this meant that the results you get on your new GTX680 could be out by a decent margin on ones from a review sample. No doubt we will hear how all of these boost clocks were higher than the minimum Nvidia advertise but that is not the point of my argument.

Putting it very simply. The performance 2x identical GTX680s could end up very different, by as much as 5% or slightly more in some cases.

 

[Smoblikat]

Thats why I will never buy an Nvidia card, all that stupid throttling and the inability to overclock your cards is just annoying. Plus, all the 6XX cards are midrange cards at a high end price.

AMD on the other hand, has their cards priced exactly where they should and they still allow you to overclock them.

 

[railven]

In some ways its a decent advance that now the cards will defend themselves from thermal and wattage overload so the cards don't get damaged after being pushed beyond design. Thermal and electrical limits are now more regularly seen, allowing a card to get closer to its maximum rather than being artificially limited in all scenarios or potentially damaged in others.

Ultimately CPUs do the same thing and while it complicates overclocking it is neitherless beneficial. It definitely saves power, considerable power over the population and it also allows the system to get to more performance when parallel usage is lower.

That isn't to say that the current form of what NVidia has is necessarily good or predictable, but lets not forget that AMD also has a version of this which is implemented in software. The AMD implementation has had a lot of bugs over the last year, many of which have meant games only see 300Mhz downclocked cores instead of the full speed. The overclocking grass is a little greener on the AMD side of the fence but the dynamic clocking is also less sophisticated.

No one has a perfect power saving solution today, Intel's CPUs loose performance from the dynamic clocking as they too aggressively return to sleep when under use and AMDs cards dont ramp up enough under certain circumstances. Power saving techniques have been a problem ever since they were introduced. If you overclock an Intel CPU today with offset voltage you don't just care about all cores loaded at top clocks, you also care about 3 cores, and 2 cores and 1 core and all the intervening clock speeds from idle up to the maximum clocks, they all have to stable.

I bought a Sapphire HD 7970 DualX advertised with Boost and it took me a second to realize what was happening. I normally ran my Ref card (without boost) @ 1125/1575 @ 100% power tune, the second card would throttle a bit, never reaching those clock. Turns out turning powertune to 120% fixed that.

Not sure if my DualX needed more juice or Boost is some kind of artificial limiter, but my volts never seemed to climb over 1.175v on either card.

Now I can run both @ 1100/1550 @ 1.125v @ 100% powertune and not have issues. Weird.

 

[SirPauly]

I would say there are so many settings for the enthusiasts, specifically with features and enhancements to improve immersion and gaming experience potential using dumb down is ridiculous.

 

[railven]

I would say there are so many settings for the enthusiasts, specifically with features and enhancements to improve immersion and gaming experience potential using dumb down is ridiculous.

Power regulation did get "dumb downed" at least on nVidia hardware and now this weird Boost thing on AMD.

Someone said it above, this thin line of balance between stable/not-stable has always existed. We enthusiast walked it. Now I really don't bother trying to find the sweet spot on our GTX 680 (or my new GTX 660 Ti), just crank the Offset to something reasonable and let the card pick a number between stock and it's offset to settle.

I guess now I can spend more time on IQ settings haha.

 

[SirPauly]

imho,

It wasn't dumb downed; one has thermal envelopes to consider; one has TDP envelopes to consider; if anything it is much more complex than just setting a clock like the past.

One may make the point about reservations about flexibility of volts and a desire to have more from OC enthusiasts. However, with these boost features, it can get a bit complicated and a learning curve may be needed compared to the past.

I think GPU boost is efficient and smart and may be dead wrong but personally allow the market to decide. Desire this to improve, evolve and mature and would like to see more of this from nVidia and AMD.

 

[railven]

imho,

It wasn't dumb downed; one has thermal envelopes to consider; one has TDP envelopes to consider; if anything it is much more complex than just setting a clock like the past.

One may make the point about reservations about flexibility of volts and a desire to have more from OC enthusiasts. However, with these boost features, it can get a bit complicated and a learning curve may be needed compared to the past.

You don't have to consider any of that, the system does. I guess what you are interpreting as "dumb downed" is most likely different how I used it. The system itself is more robust, so robust it get's stupid in it's calculations and throttles you without real indication.

I know on my Kepler card's thermal's are the reason they throttle. If it is power draw, well I don't get how giving the card more power through Power Control causes it to throttle more. To the end user: the system is less sophisticated since we don't have direct control on the outcome, ie it's dumb downed.

Like I said, in the end I just give it a modest Offset and let the card decide what to use - since I really have no control on that anymore. I can watch the card's clock fluxuate up and down all day, it would drive me crazy - glad I don't use that card else I'd invest more time in trying to figure out why haha.

 

[SirPauly]

Sure you do if one desires to over-clock their GPU.

 

[railven]

Sure you do if one desires to over-clock their GPU.

Sure you do what?