Since recently upgrading to a GTX 780 (my first card which "boosts") I've been having a play with MSI Afterburner and I haven't quite got my head around it yet. At stock clocks, when I increase the voltage and test the card in Valley for example, the boost clock increases slightly (from 1019 to 1030), suggesting that it was voltage limited. However, when I return the voltage to stock and increase the GPU clock to +88 and run Valley again, it boosts to around 1100, suggesting voltage isn't a problem. Woss-a-goin-on?
i can't recall the range exactly, as i don't bother overclocking GPUs anymore, but relying solely on core voltage &/or boost is very limited. The general methodology is to start by increasing the power limit - if your 780 is limited to much less than ~120-125% in Afterburner then there is the option of flashing the bios http://forums.anandtech.com/showthread.php?t=2341641... ...i've flashed bioses on other cards before without issue, but YMMV of course. ***always backup your stock bios just in case a replacement is unstable*** & then increasing the core clock gradually - testing under high load for artefacts & keeping an eye on core temps/adjusting fan speeds as necessary. Then, after that, you can look at increasing the voltage by a small amount to see if that will give you a little more speed - again testing & looking at temps. & finally it's increasing the memory speed - again testing.
Cheers, I may have a look into BIOSes at some point but at the moment I'm just trying to understand how the boosting works. I've increased the power limit to 106% but the same thing happens. Stock volts & stock clocks = 1019 boost. Raised GPU volts & stock clocks = ~1030 boost (I can't remember exactly but certainly higher). Stock volts & +88MHz GPU = 1097 boost. If it can boost to 1097MHz at stock volts when I overclock the GPU, why can't it boost to 1030 at stock volts & stock clocks?
IIRC there are different boosts on Geforce cards. There is the stock speed, and the boost speed. However, there is a 3D clock? boost speed also. The only way to really tell what you are boosting to is to run an OSD (I use Precision X) and then you will be able to see the exact clocks the card is boosting to when gaming
I use Afterburner's graphs to monitor the GPU speed. After running Valley, I open the graphs and have a look how the GPU speed, GPU volts, temps etc behaved during the run. I don't know about the 3dclock speed but I have noticed Valley's GPU monitor always reports a much higher GPU speed than Afterburner. Could that be it? Interestingly, I've also noticed that according to Afterburner's graphs, the GPU voltage doesn't change at all until I add +25, when it jumps to 1.187v. If I add any less voltage (like +20 for example) then according to the graph, the voltage remains at stock (1.162v).
i think you've misunderstood. What i was saying was to (first) increase the power limit & then solely increase the GPU clock to a point just below where you're getting artefacts when testing... ...not to increase the power limit (or voltage) & then expect a significant difference to the boost clock without also altering the GPU clock speed. What you're naturally aiming for is the highest possible GPU clock speed (which the boost speed is then derived from) without issues - & increasing the power limit simply gives more flexibility to do so. So, imagining that +88Mhz was the max before artefacts under load with the power limit at 100%, you should be able to hit >+88Mhz with it at 108%... ...& higher still at 120-125%; which is the sensible max. (again, the voltage setting is just for a final bit of tweaking at the end if you can be bothered)
I get what you're saying but I'm not aiming for maximum overclock at the moment, and I certainly wouldn't expect to reach it by increasing the power limit or voltage alone. What I'm doing at the moment is experimenting to understand how it works, and I'm not quite there yet (obviously).
Unigene is always wrong. It thinks your card is running 200mhz or so faster than it really is. Google "Afterburner OSD" and enable the clock speed. Then run Firestrike or something similar to get your actual clock speeds.
Yes, I thought there was something wrong with the speeds reported by Unigene although it had me a bit worried at first. What is the advantage of using Afterburner's OSD over looking back over the graphs? Is it likely to give a different reading?
Heh, I still don't fully understand what's going on - for example if I trigger the power limit flag with a certain config, why am I then able to increase the clockspeeds even further by adding to the GPU clock offset? Anyway, all that aside, I decided to enjoy my ignorance and have a little play anyway. I've now switched from Afterburner to Precision X and I opted for a middle-of-the-road +25mV voltage increase to see how it gets on (actual volts = 1.187v). Incidentally, Precision X only offers the option to increase voltage in 3 increments (+13mV, +25mV, +37mV) which might explain why most of the positions on Afterburner's voltage slider did nothing. With 106% power limit and +170MHz on the GPU I ended up with a core speed of 1030MHz, boosting to 1202MHz under load at 64C, as can be seen by the OSD info in the screenshot below. Not too shabby!
Just as a rough comparison, from a very cursory check there were certainly air cooled 980 models which came with a 1216MHz Core // Boost Clock:1317MHz starting point; before playing with Afterburner - https://uk.msi.com/Graphics-card/GTX-980-GAMING-4G.html#hero-overview &, ttbomk, they would typically be stable with around a 15% increase above those levels with a ~120-125% & +180-200MHz. Now obviously you're limited with the current f/w on your card & may have temp issues before hitting that point d.t. a lesser fan setup - but it's gives an idea of what you could be aiming towards.
This is a reference 780 (Kepler), not a 980 (Maxwell). I've put it back to default volts and boosting to 1097 again. It's fast enough for me so I'll give the card a break since it's a few years old and I've only had it a couple of days. Nice to know it'll do at least 1200MHz though (I only stopped there because it was a nice round number. It might go further). I'm still not clear on exactly how the boost works but I'll live with that seeing as no one else seems to know either.
IIRC, the way Nvidia Boost works is that, unless there is a clock limit in effect (e.g. when on the desktop the card will throttle down to low clocks, barring the multi-monitor issue) the card will attempt to boost as high as possible, limited by one of several limits: Power limit - Will stop boosting when the power draw reaches the maximum set Voltage limit - Will stop boosting when the clock/voltage curve would exceed the voltage limit. For Maxwell this was a static curve, and 'changing the voltage' adds an offset to this curve (raising the max value, raises all lower values) for Pascal you can tweak it per-clock and per-voltage. Temperature limit - Will stop boosting if max core temperature exceeded. Afterburner will show when each of these limits is active, so you can see what (if any) headroom is available, and what needs to be tweaked. All three are interrelated (raising voltage dumps more 'waste' power into the die which brings the effective power limit down, and more power in the die raises temperatures) so optimum tweaking for the maximum possible boost clock is more complex than just turning everything up to 11.
That's exactly how I understand it too, in theory. In practice however, something else is happening, as I explained earlier: "Stock volts & stock clocks = 1019 boost. Raised GPU volts & stock clocks = ~1030 boost (I can't remember exactly but certainly higher). Stock volts & +88MHz GPU = 1097 boost. If it can boost to 1097MHz at stock volts when I overclock the GPU, why can't it boost to 1030 at stock volts & stock clocks?" PS my card is a Kepler 780. I don't know why 980/Maxwell keeps being mentioned.
To know that, check Afterburner's Limit graphs. Those will show you what it is that is stopping the boost clocks raising higher.
At stock volts & stock clocks, according to Afterburner it was voltage limited (as you would expect from a card running completely at stock). Yet it managed to go much higher (1097MHz) when I increased the GPU clock (and left the voltage untouched). It doesn't make a lot of sense. Do you see what I mean? Similarly, on other occasions I've triggered the power limit when increasing volts & clocks but if I ignore it and keep increasing the GPU clock, the boost speed seems quite happy and continues to rise, despite reaching the supposed "power limit".
