780 GTX overclock results

lordsegan

Gawd
Joined
Jun 16, 2004
Messages
624
I managed to get my Asus 780 up to 1020 OC for the GPU core and 6200 on the ram. Should I try for more? Nice and stable. Didn't tweak anything else.

I have an uber case with a pull fan on the back of the pci card area, but my temps are really low... Fan runs around 45% and card stays in the 70C range.

My CPU is maybe slightly limiting... i7 860 at 3.8ghz.....


... Waiting for hassswelllllll
 
I managed to get my Asus 780 up to 1020 OC for the GPU core and 6200 on the ram. Should I try for more? Nice and stable. Didn't tweak anything else.

I have an uber case with a pull fan on the back of the pci card area, but my temps are really low... Fan runs around 45% and card stays in the 70C range.

My CPU is maybe slightly limiting... i7 860 at 3.8ghz.....


... Waiting for hassswelllllll

what are you using to stress your card, test stability and error checking?... for that clock i see your temps kinda high.. that cards it suppose that perform really cold... about the OC.... keep that numbers its fine, you nor even need to OC your card with your CPU... you will not enjoy the card and their full capacities..
 
70C is fine. It just needs to be below 80C to keep it from throttling.

BTW, I got a similar overclock over and above what my EVGA GTX 780 SC ACX was already overclocked to. Still testing out the stability and trying to find the peak. I'll post specific settings as soon as I am confident in them.
 
Use the Kombuster program to test stability and temps.

Kombustor is a crappy way to test for stability on Kepler. You'll get power or temperature throttling before you ever seriously stress the GPU. Game demos or Unigine Heaven are a much better way to test for stability.
 
Kombustor is a crappy way to test for stability on Kepler. You'll get power or temperature throttling before you ever seriously stress the GPU. Game demos or Unigine Heaven are a much better way to test for stability.

Explain to me how you can "not stress" the cuda cores but yet raise the working temperature of the GPU?

So to avoid the throttle, raise the power and temperature limits, which you would do anyway if overclocking.

Kombuster is just one example. easily obtainable. I prefer to use any modern game's benchmark program and apply AA to give a more real world structure, as another example.:D
 
Explain to me how you can "not stress" the cuda cores but yet raise the working temperature of the GPU?

I don't understand the science behind it, but it's true. Kombustor puts a lot of power load on the GPU but doesn't stress it as hard as games or other benchmarks for that power load. I've seen it myself many times. An overclock could be Kombustor/FurMark stable but will fail miserably in a game or other benchmark. Kepler makes it even worse since you'll run into throttling at power or temperature limits.

So to avoid the throttle, raise the power and temperature limits, which you would do anyway if overclocking.

Max TDP limit is only 106%, so that's really not going to buy you much.
 
I use BF3 online to test my settings. I've found nothing that stresses a card more.
 
OK......

but I've used Kombuster to test cards a bunch......going back to "furmark".
granted, you don't get real world stress using something like kombuster, but for basic functionality it stresses a card just fine. YMMV.

I've found an OC on Kombuster generally holds up just fine on using a game run-through or built in benchmark.
and, this is testing with the latest NVidia cards.....ie GTX 680 and 670.

Yes the Kepler is a different breed, with built in Boost and power limitations, but the basic principles seem to hold true, for me at least.
 
Anybody seen any 780 modded BIOSes yet? Even just something to get rid of the Power Target or set it high enough to not be the limiting factor for most overclocks would be great.
 
Anybody seen any 780 modded BIOSes yet? Even just something to get rid of the Power Target or set it high enough to not be the limiting factor for most overclocks would be great.

There's a test version of the TechInferno one on their forums. A few people have said it worked well for them and got better overclocks on there and ocn, while one said it didn't improve clocks and made his card actually not clock quite as well when using that BIOS so he flashed back to the stock one. It'll bring your voltage max to 1.212v and power target max to 116%.
 
Explain to me how you can "not stress" the cuda cores but yet raise the working temperature of the GPU?

It thermally stresses the gpu with the hottest, hungriest paths, but does not tickle all circuit paths, notably the problematic speed paths. I suspect "stress" is used fast and loose here to mean "find the point of instability wrt clock speed".
 
Call me crazy... but if the game doesn't lock and doesn't have artifacts, and runs faster with the OC than without.... who cares if there are bit errors?
 
For my EVGA 780 GTX Superclocked, I can get another +100 core/mem before I get outright crashes running Unigine Heaven. I probably have some more wiggle room with voltage modifications but to be honest, considering its already overclocked, I am happy as is. 1189 core on full boost.

That said, I noticed some issues with fan speed. Sometimes it works sometimes it doesn't; I've tried doing a custom fan profile and forcing updates as well as disabling and letting self-auto work. Sometimes the fan just stays at 38% and the temperature rises up to 80c+ without the fan so much as noticing.

So for now I am running stock speeds until this problem is addressed - either via EVGA precision or nvidia drivers.
 
I don't understand the science behind it, but it's true. Kombustor puts a lot of power load on the GPU but doesn't stress it as hard as games or other benchmarks for that power load. I've seen it myself many times. An overclock could be Kombustor/FurMark stable but will fail miserably in a game or other benchmark. Kepler makes it even worse since you'll run into throttling at power or temperature limits.



Max TDP limit is only 106%, so that's really not going to buy you much.

The reason behind this is that Nvidia considers Furmark and similar programs a "Power Virus", and automatically throttles the card to prevent noobs from burning their GPU up and driving RMA rates through the roof..

Google it, there are some interesting reads about it..
 
keep that numbers its fine, you nor even need to OC your card with your CPU... you will not enjoy the card and their full capacities..

I am sick and tired of reading your broken English full of outright lies..You spout so much FUD about CPU limits and GPUs in general that you have ZERO credibility..

I have seen you do it in thread after thread, and all you are doing is telling people what amounts to a lie:rolleyes::rolleyes::rolleyes::rolleyes::rolleyes:...
 
I don't understand the science behind it, but it's true. Kombustor puts a lot of power load on the GPU but doesn't stress it as hard as games or other benchmarks for that power load. I've seen it myself many times. An overclock could be Kombustor/FurMark stable but will fail miserably in a game or other benchmark. Kepler makes it even worse since you'll run into throttling at power or temperature limits.

This is my experience as well, and is especially true in SLI. A benchmark stable overclock does not equal a game stable overclock. Power usage numbers are also inconsistent from real game play, although Kombustor usually drew more power than my games.

So far the Heaven benchmark with med or high Tesselation is about the closest measure (within 10% GPU/Memory) of what OC is "game stable" for me. Valley will routinely pass multiple tests with a significantly higher OC than what I can get away with in any game.
 
I am sick and tired of reading your broken English full of outright lies..You spout so much FUD about CPU limits and GPUs in general that you have ZERO credibility..

I have seen you do it in thread after thread, and all you are doing is telling people what amounts to a lie.

His posts usually annoy the hell out of me too, but you don't agree that a GTX 780 is going to be bottlenecked in quite a few games with a Core 2 Duo at 3.6GHz?
 
This is my experience as well, and is especially true in SLI. A benchmark stable overclock does not equal a game stable overclock. Power usage numbers are also inconsistent from real game play, although Kombustor usually drew more power than my games.

So far the Heaven benchmark with med or high Tesselation is about the closest measure (within 10% GPU/Memory) of what OC is "game stable" for me. Valley will routinely pass multiple tests with a significantly higher OC than what I can get away with in any game.

This is spot on generally... you can just about always bench higher than you can run in games by a nice margin. That's why I mainly care about 24/7 oc's, because let's face it, you're not gaining anything by managing to cobble a card through a bench run other than bragging rights (barring you being one of the sponsored guys who receive cards in packs of 10 from nVidia/AMD for butchering :p ).
 
Just an update with my OCing (re: evga 780 SC (non-acx))

Seems like my core doesn't like anything above +90, even with all 3 extra voltage increments. Heat is not an issue as 85% fan speed keeps temps below 70c. Beyond that, memory goes as far as +225 before performance numbers start dropping. I reduced the memory to +150 as the performance gain between that and 225 seems negligible.

A little disappointed that voltage doesn't help increase core but can't really complain. Already at or above titan performance.
 
Term-X, what is your actual boost clock in mhz as shown in Afterburner or Precision-X while gaming/benching?
 
Just an update with my OCing (re: evga 780 SC (non-acx))

Seems like my core doesn't like anything above +90, even with all 3 extra voltage increments. Heat is not an issue as 85% fan speed keeps temps below 70c. Beyond that, memory goes as far as +225 before performance numbers start dropping. I reduced the memory to +150 as the performance gain between that and 225 seems negligible.

A little disappointed that voltage doesn't help increase core but can't really complain. Already at or above titan performance.

You know, I'm wondering if the current 320.18 drivers are holding back our 780 OC's to some extent. When I think I've found my magic number (previously +135 GPU/+100 Memory) I'll encounter a random game or Firefox crash (just happened while I was writing this post). The Firefox crashes can occur while in or out of a game (when using 2d clocks). This prompted me to leave everything at stock for awhile and, sure enough, I've still had a couple of game and Firefox crashes.

Unfortunately since these are the only drivers to provide support for the GTX 780 (that I'm aware of), there's really nothing else to roll back to for comparison. I also can't rule out the possibility of just having a bad card but, due to the number of people reporting issues with these drivers, I'm hesitant to blame it on the cards.
 
@GoldenTiger

At +90 core / + 150 mem:

It boosts up to 1150 core during full load; memory just for the sake of reference sits at 6308.

@Artificiary

I wouldn't be surprised, though, I suspect that it is more probable that it's just a case of each graphics card and it's components can differ when it comes to headroom. I see more reviews showing the core going higher than 90 but not all have the same flexibility I've experienced when it comes to memory speed.
 
@GoldenTiger

At +90 core / + 150 mem:

It boosts up to 1150 core during full load; memory just for the sake of reference sits at 6308.

@Artificiary

I wouldn't be surprised, though, I suspect that it is more probable that it's just a case of each graphics card and it's components can differ when it comes to headroom. I see more reviews showing the core going higher than 90 but not all have the same flexibility I've experienced when it comes to memory speed.

Thanks :), was curious!

You know, I'm wondering if the current 320.18 drivers are holding back our 780 OC's to some extent. When I think I've found my magic number (previously +135 GPU/+100 Memory) I'll encounter a random game or Firefox crash (just happened while I was writing this post). The Firefox crashes can occur while in or out of a game (when using 2d clocks). This prompted me to leave everything at stock for awhile and, sure enough, I've still had a couple of game and Firefox crashes.

Unfortunately since these are the only drivers to provide support for the GTX 780 (that I'm aware of), there's really nothing else to roll back to for comparison. I also can't rule out the possibility of just having a bad card but, due to the number of people reporting issues with these drivers, I'm hesitant to blame it on the cards.

320.08 I think was a pre-release driver, but not sure if it's even publicly available, let alone if it would help.
 
You know, I'm wondering if the current 320.18 drivers are holding back our 780 OC's to some extent. When I think I've found my magic number (previously +135 GPU/+100 Memory) I'll encounter a random game or Firefox crash (just happened while I was writing this post). The Firefox crashes can occur while in or out of a game (when using 2d clocks). This prompted me to leave everything at stock for awhile and, sure enough, I've still had a couple of game and Firefox crashes.

Unfortunately since these are the only drivers to provide support for the GTX 780 (that I'm aware of), there's really nothing else to roll back to for comparison. I also can't rule out the possibility of just having a bad card but, due to the number of people reporting issues with these drivers, I'm hesitant to blame it on the cards.

After reading a lot of comments about issues with 320.18, I edited the ini file on 314.22 to install for the GTX 780. Just got my card in a couple of hours ago, but that seems to be working pretty good right now.

@GoldenTiger

At +90 core / + 150 mem:

It boosts up to 1150 core during full load; memory just for the sake of reference sits at 6308.

@Artificiary

I wouldn't be surprised, though, I suspect that it is more probable that it's just a case of each graphics card and it's components can differ when it comes to headroom. I see more reviews showing the core going higher than 90 but not all have the same flexibility I've experienced when it comes to memory speed.

Well, you have to remember, you're starting at a higher clock than vanilla, so your +90 core is actually more like +150 when compared to a vanilla card (edit: looks like it takes +148 on my vanilla Gigabyte to hit 1150 core).

Also, did you try to see how high your core would go with 0 offset? I'd get the core as high as possible, then start bringing in extra memory speed as it will allow. These cards seem pretty finicky about power draw and temperature (even if not at their limits) without a custom BIOS to get rid of those hurdles.

Oh, and one more thing, from the reviews and experiences I've read about, it seems like additional voltage hasn't really helped people's overclocks at all. I think it's probably an issue of making it hit the power target faster since power draw increases by a factor of the square of the voltage.
 
Last edited:
Up to 1202 core with memory at stock (6004) so far looping Unigine Heaven. Temps are sitting in the low 70s with fan speed at 62%.

edit: Ha, so the Caustic Caverns in Borderlands 2 are pretty good at sniffing out an unstable overclock. 1202 was pretty stable in Heaven, but I've had to knock my OC down to 1150 core to be stable in BL2.
 
Last edited:
Try the tech inferno bios if you want also, it is doing nice things for people so far.
 
Try the tech inferno bios if you want also, it is doing nice things for people so far.

Yeah probably will eventually. Just want to give it a good beating on the stock BIOS first and get some baseline OC numbers.
 
I ran both my cards against Uniengine last night.

I'm not done yet, but have had not one issue running the 320.18 driver plus:

power +106%
voltage +5
temp maxed
core = +130
memory +50
fan: auto

highest fan cycle 45% and highest temp was 75C, but this is on an open bench.

Under stress the core clock runs at 1163. I averaged about 80 FPS on the benchmark.

I think there is a lot more headroom but I ran out of time.:D

Great cards.....Galaxy branded.
 
After reading a lot of comments about issues with 320.18, I edited the ini file on 314.22 to install for the GTX 780. Just got my card in a couple of hours ago, but that seems to be working pretty good right now.

Good idea, I'd also like to try this. I just took a peek at the files in the \Display.Driver folder for 314.22. Is the 'ListDevices.txt' the only file I need to modify to make this work? If it's more involved I'm sure I can find a guide online.
 
Good idea, I'd also like to try this. I just took a peek at the files in the \Display.Driver folder for 314.22. Is the 'ListDevices.txt' the only file I need to modify to make this work? If it's more involved I'm sure I can find a guide online.

It's the nv_disp.inf file you need to update. Basically, you need to add the Device ID string to the [String] section for the 780 and add the Section ID (same as the Titan entries) to the various "NVIDIA_SetA_Devices..." sections.

I've got an already updated copy of it for the x64 version of 314.22 here. But I list this for reference only, don't take any responsibility if you blow your shit up, etc.
 
Is the EVGA scanner enough of a stress test?

Core = 1150
Mem = +100
Fan speed = 60%
Temp = 70
 
Last edited:
I have been reading a little about the TechInferno BIOS and I've got an EVGA GTX 780 SC ACX on the way but I'm a little concerned about bricking my card. I've only ever flashed a video card BIOS with my old 6950 that had a dual BIOS chip so it was essentially risk-free. What are the chances of bricking your card (assuming you follow all the directions) and what can you do if the flash fails?
 
Also, did you try to see how high your core would go with 0 offset? I'd get the core as high as possible, then start bringing in extra memory speed as it will allow. These cards seem pretty finicky about power draw and temperature (even if not at their limits) without a custom BIOS to get rid of those hurdles.

Oh, and one more thing, from the reviews and experiences I've read about, it seems like additional voltage hasn't really helped people's overclocks at all. I think it's probably an issue of making it hit the power target faster since power draw increases by a factor of the square of the voltage.

At stock, full load boosts the core to 1071. So it seems like it's more or less +80, even OC is +90. Not quite certain if this is a throttling thing, despite temps being held under 70C (which I believe is the first default throttling plateau). Even moreso, power target is at 106% temp target set at 90C, so that should eliminate the throttle until then? Unless I'm misunderstanding the point of these two settings.

But yea, the voltage adjustment doesn't seem to do much of anything at all.
 
Is the EVGA scanner enough of a stress test?

Core = 1150
Mem = +100
Fan speed = 60%
Temp = 70

No clue; I can't even get the EVGA OC Scanner X to stay "focused"; keeps disappearing when I try to access it from the task bar and/or has most of it's UI invisible. That said, the brief moments I got to see some parts of it, seems to use things like Furmark. Which can work to a degree but it isn't exactly relative to game load. Try to use something like Unigine Heaven to get a close idea of game stability.
 
I might be a little rusty when it comes to overclocking, but what side effects are you experiencing when going too high. On my SLI setup I am either working okay or the computer just restarts unexpectedly. I am wondering if I am bumping into some other kind of limit. On my last video card my game might crash, get artifacts, or something along those lines.
 
I might be a little rusty when it comes to overclocking, but what side effects are you experiencing when going too high. On my SLI setup I am either working okay or the computer just restarts unexpectedly. I am wondering if I am bumping into some other kind of limit. On my last video card my game might crash, get artifacts, or something along those lines.

My 260 (now sold), 570 and Titan all CTD (crash to desktop) but my 680 hard locked every time. I suppose it depends on precisely what circuit gets oc'd too much on the gfx card. Ideally it recovers, but that just can't be guaranteed.
 
I might be a little rusty when it comes to overclocking, but what side effects are you experiencing when going too high. On my SLI setup I am either working okay or the computer just restarts unexpectedly. I am wondering if I am bumping into some other kind of limit. On my last video card my game might crash, get artifacts, or something along those lines.

If you went too high, it normally crashes as soon as it gets stressed. Only side effect I can think of (outside of killing it, which is rare) is it begins to degrade slowly.

Barebone stock 3dmark11 (cpu & gpu):
http://www.3dmark.com/3dm11/6656271

I'm curious to see how much I'll be able to squeeze out of this card (66.3 asic).
 
It's the nv_disp.inf file you need to update. Basically, you need to add the Device ID string to the [String] section for the 780 and add the Section ID (same as the Titan entries) to the various "NVIDIA_SetA_Devices..." sections.

I've got an already updated copy of it for the x64 version of 314.22 here. But I list this for reference only, don't take any responsibility if you blow your shit up, etc.

Thanks for the file Parja. I was able to install the 314.22 drivers successfully but my issues turned out to be due to one of my cards being faulty. I've since replaced the bad GTX 780, reinstalled 320.18, and have experienced none of the same issues.

With two functional cards I decided to retest my overclocks. I can now (amazingly) overclock the memory to a significant degree and run a higher GPU OC at the same time.

Current "Benchmark Stable" OC (SLI) -
GPU: +175
Boost Clock: 1189 (according to Afterburner)
Memory: +500
Power Target: 106%
Voltage (Mv): +37

I did observe a few artifacts in Heaven with these settings, but lowering Memory to +450 eliminated them.

A couple of interesting observations during my testing:
  1. Increasing the Voltage does have a noticeable affect on the max Boost Clock (at least for me) as well as the GPU temps. With no voltage increase, and all other settings the same, my Boost Clock only went to1163 (while temps were roughly 7C lower).
  2. These cards appear to be Boost Clock capped at 1202. I ran into this wall when attempting +210 and +195 GPU. Both settings could complete about 6 scenes in Heaven before crashing and they both locked in at 1202 for the Boost Clock. This can also be confirmed by looking at a stock Bios rom file in Kepler Bios Tweaker.

Will probably use BF3 to determine my highest game stable OC.
 
Last edited:
Back
Top