Would it be possible to accidently undo the stock-overclock on a EVGA FTW card?

Cyber Akuma

Cyber Akuma

Gawd
Joined
Jan 3, 2009
Messages
646
I have an EVGA GTX 670 FTW card, it comes stock overclocked.

Now, I know that the cards pretty much come overclocked already, but I downloaded EVGA's Precision X tool to see if it can be pushed any farther, in small increments. Thing I want to know is, is the stock-overclock set as the card's default clock and thus any overclocking offsets are set based on this already in-place overclock... or is it just a configuration preset and its possible to accidentally wipe it and bring the card down to reference stock settings by trying to set the clocks back down to their default values with the Precision X tool?

The card actually seemed to perform much much worse with any attempts to overclock with the Precision X tool, is this because its pointless to overclock a FTW card or because the overclock's offsets in the Precision X are based in it's reference clocks instead of it's stock overclock and I was effectively underclocking?
 
Last edited:
Yes, the stock overclock is set in the BIOS as the "default clocks" on pre-oc'd cards like you have. So, your offsets are based on the already in-place overclock. :)

EDIT: In response to your edit, OC'ing the memory can result in loss of performance if you go too far with it. With the GDDR5 specification the error correction will kick in to prevent it from crashing (unless you go obscenely far), but it will reduce performance as it discards any errored data over and over.
 
By the way, this is a new build and is currently running a junk install of Windows 7 that I am just testing different apps, drivers, and settings on. Once I find a stable overclock point, does it save these settings on the GPU card itself somehow or would I lose them once I reformat and re-install windows from scratch? I know the CPU will remain overclocked since that's set on the motherboard but what about my GPU?
 
It is not written to the GPU hardware in any way, just the program you use to overclock as a stored settings file. So, if you reformat/reinstall you will lose those settings; make sure you write them down first :)!
 
Hmmm, in that case, how would the GPU remain overclocked if the program wasn't running?
 
Cyber Akuma said:
Hmmm, in that case, how would the GPU remain overclocked if the program wasn't running?
Click to expand...

The driver keeps them set until it is reloaded, at which point the clocks would revert to the ones stored in the video card's BIOS (i.e. logging off of your windows session, rebooting, etc.). You normally keep the program running while using an overclock and let it reapply it every startup.
 
Ok, getting some weird results from the EVGA Precision X overclocking tool and the EVGA OC Scanner testing tool.

The "Power Target" setting apparently is supposed to be a sort of ceiling percentage, where if your overclock settings hit that max percentage it won't push the card harder even if your settings are set higher. Every overclocking guide and post I saw said to just set this to as high as it will go since all it will do it bottleneck your overclock if set lower.

Problem is, in the OC Scanner tool I get a massive framedrops if its set anything above the default 100%, get the GPU clock doesn't even reach 1000Mhz if I have the Power Target set to 100% in stress test mode:

http://i49.tinypic.com/hrh1sy.png

http://i48.tinypic.com/348fzfo.png

(Notice the FPS and core speeds there, its at 289FPS with a core of 992Mhz at 100%, and at 86FPS with a core of 1175 at 145%)

This happens even if all my settings are at stock, I get around 260FPS on the Fur+Tess benchmark when its at 100%, and around 88 FPS when its anywhere above 100%, whether its 110% or 145%, despite the fact that since the GPU and memory clock are both at offset +0 raiding the Power Target should have no effect.

.... help? I am very confused.
 
Raising the power target is like a two sided blade you know, the core voltage will go up too and then the GPU will reach temperatures above 70*C and the card stars to throttle...

The best GTX 670 OC guide is THIS, please read it carefully, it explains a lot of things.
 
Ok, so I found a stable GPU overclock point, and still scaled iot back a little bit just for some leg-room as well as since it was still a faster overclock than I expected, and now getting started on finding a stable overclock point for the memory.

Problem is, I can't...

Rather, to be more accurate, I can't find an UNstable overclock point for the memory.

I have never overclocked a 3D card before so I don't know what would be considered an average memory overclock for them, but I have it now at an offset of +500 and after hours of stress testing, several runs of Heaven on max settings eyeballing for any artifacts, and bench-marking I have had no problems. Is 500 a low overclock, average, or high? All guides I read recommended starting out at 100 and increasing it if +100 is stable, so going by that guide 500 seems excessive. I am not sure if I should try for higher or not, or if it would even make a difference that much higher or if I am risking damaging it by going that high.
 
Cyber Akuma said:
Ok, so I found a stable GPU overclock point, and still scaled iot back a little bit just for some leg-room as well as since it was still a faster overclock than I expected, and now getting started on finding a stable overclock point for the memory.

Problem is, I can't...

Rather, to be more accurate, I can't find an UNstable overclock point for the memory.

I have never overclocked a 3D card before so I don't know what would be considered an average memory overclock for them, but I have it now at an offset of +500 and after hours of stress testing, several runs of Heaven on max settings eyeballing for any artifacts, and bench-marking I have had no problems. Is 500 a low overclock, average, or high? All guides I read recommended starting out at 100 and increasing it if +100 is stable, so going by that guide 500 seems excessive. I am not sure if I should try for higher or not, or if it would even make a difference that much higher or if I am risking damaging it by going that high.
Click to expand...

GoldenTiger said:
EDIT: In response to your edit, OC'ing the memory can result in loss of performance if you go too far with it. With the GDDR5 specification the error correction will kick in to prevent it from crashing (unless you go obscenely far), but it will reduce performance as it discards any errored data over and over.
Click to expand...

As I explained earlier :).
 
I am not experiencing any performance loss though, I am actually gaining FPS as compared to just having the GPU overclocked.
 
Cyber Akuma said:
I am not experiencing any performance loss though, I am actually gaining FPS as compared to just having the GPU overclocked.
Click to expand...

Then it's still working properly... however different games/apps will be affected differently, so make sure you're gaining performance still in multiple ones.
 
Cyber Akuma said:
Rather, to be more accurate, I can't find an UNstable overclock point for the memory.
Click to expand...

Just set the memory to 3300, and be done with it. Sure you can spend a bunch of time squeezing out an extra 1/2 fps, but is it worth the effort at this point?
 
I am just going by increments of 100, I am not trying to fine-tune to the 10ths or anything.

I did want to ask one thing though, I ran hours of stress tests on this thing at a clock of +600 and it detected no problems or errors, FurMark and OCCT passed fine too, however, eyeballing it in Heaven Benchmark for about 5 minutes showed clear artifacts.

Even at what I assumed was a stable clock at 500, I just barely by chance noticed an artifact once in Heaven Benchmark when several runs before had wielded nothing... that or I never noticed them.

So then, since I can't just sit staring at my screen for hours looking for artifacts without shooting myself in the head out of boredom, is there any app similar to Prime95 or Linpak but designed to run on a GPU?

What I liked about Prime95 and the Linpak-based IntelBurnTest is that it both stressed my CPU and my memory and instead of just simply putting load on it to see if it crashes.

All the stress testing and benchmark apps I saw just seemed to stress the GPU, not check if the calculations were accurate. So is there anything that will task the GPU and it's memory with similar calculations that most videogames would perform AND also check to make sure its performing them correctly?
 
Try eVGA OC Scanner X to find automatically artifacts ;)
There is a CUDA and OpenCL memory stress tester (like memtest86/memtest86+): LINK.

But - and that's a big one - the best OC testing program IMO is BF3 on 48-63 servers. Trust me, many have failed with "stable" OC's with that "utility"... LOL
 
That eVGA one is actually what I tried, it didn't seem reliable to me, running it for several hours would show no errors or artifacts but a few minutes of Heaven and it was obvious.

Are there any pre-compiled windows binaries of that gpumemtest one?
 
You must log in or register to reply here.
Back
Top