[stratcatprowlin]

I'm having a little trouble understanding what this guy is really saying.
Is he saying that by overclocking the gpu we are making drops in fps worse?

Example:A game runs at 60fps but sometimes dips to 20.I then overclock the gpu and now get 63 fps but dip down to say 17fps?

[theELVISCERATOR]

Originally Posted by stratcatprowlin

I'm having a little trouble understanding what this guy is really saying.
Is he saying that by overclocking the gpu we are making drops in fps worse?

Example:A game runs at 60fps but sometimes dips to 20.I then overclock the gpu and now get 63 fps but dip down to say 17fps?

you got it.

[stratcatprowlin]

I never looked into this.I always figured overclocking the core would increase minimum framerates.I wonder if this is a rule or if performance varies.I guess I would have to find out for myself if this occurs with my card.

[MSbMeaner]

I have always held that ocing the memory first then the gpu gave better gameplay ocing. Now I know why, if this is true, and I plan to test it. If this can be proven than I will feel vindicated when so many times others ostercised me for telling others in forums thats the best way to oc vid cards.

[austinbmxnig]

yea im pretty sure that memory clock speeds are more important than the core as far as real world gaming is concerned....benchmarks seem to like a higher core though

[Sjaak]

I believe the cause for this drop is to be found in the GPU architecture...though i'm now known with it.

[Voodoo Rufus]

I thought this could be caused by GPU instability at highest operating speeds. Small errors causing decreases in performance? Sometimes I've run into the same problem on my old GF3.

[stratcatprowlin]

Yeah,I had a couple of video cards that were like that.Losing performance before the max overclock.I think every card is different and the highest overclocks may or may not be best for each one.
Nonetheless,I have never tested for this minimum framerate loss phenomena.

[MSbMeaner]

Im too busy right now so could some one test this on 6800gt? I noticed the test this guy ran was on a geforce card.

[obsolete]

I think it depends on the card like stated above. If you have done a lot of video overclocking. At some point or another you may have noticed that you can squeeze the extra 5 or 10 mhz out of a given card without artifacts, but the performance may actually decrease.

I think this is pretty much along the same lines.

I thought this could be caused by GPU instability at highest operating speeds. Small errors causing decreases in performance? Sometimes I've run into the same problem on my old GF3.

[RJARRRPCGP]

With a Radeon video card, may be because of SmartGART dropping the AGP transfer rate. An unstable video card overclock may cause SmartGART tests to fail.

[macklin01]

Originally Posted by Voodoo Rufus

I thought this could be caused by GPU instability at highest operating speeds. Small errors causing decreases in performance? Sometimes I've run into the same problem on my old GF3.

I think this is it. Here are a few points to connect:

1) When a gfx card is yielding its minimum frame rates, its GPU and memory are under the greatest stress. (highest polygon counts, greatest T&L complexity, etc.)

2) On all the graphs, the min framerate increases with clockspeed until just a bit before the max overclock. In the first plot, this was around 195 MHz. In the second, this was around 220 MHz.

3) The improvement from the first to second graphs was attained by improving the cooling.

4) The article states that at 230 MHz in the second configuration, the gfx card felt unstable. For the first configuration, greater than 210 MHz crashed the card.

Putting these points together, we see that the apparently max stable overclock isn't. At 10-20 MHz below that OC, the card is experiencing instability when at its fullest load / stress (i.e., the times when the framerate would drop to a minimum), and when the OC is further increased, the card is unstable more of the time. This is not all that dissimilar to the case where a CPU overclock is apparently stable, but it crashes in superPI or certain folding WUs; it's not actually a stable overclock.

I think what he found is that the minimum framerate gives a better measure of how close one is to the max stable overclock, as a drop in that framerate is the very first manifestation of GPU instability. -- Paul

[mjw21a]

Hmmmm, this makes me wonder whether the same thing can happen with CPU's also....... Apparently stable, but actually performing worse.....

[Voodoo Rufus]

I've never run into anything like that on CPUs. To me, Folding either runs or it doesn't when it goes unstable.

[Quailane]

Hey, I noticed this in America's Army. When I had my 9700pro overclocked as high as it would go, my framerates would hover at 12-18 outside of the hospital, but now without overclocking it stays above 18 no matter what, with the average in that spot 25 to 30.

[@md0Cer]

This has sparked my curiosity. I am doing some testing of my own.

[Sjaak]

Originally Posted by @md0Cer

This has sparked my curiosity. I am doing some testing of my own.

let us know what you find

[@md0Cer]

Originally Posted by Sjaak

let us know what you find

I will have a full review with charts and all. So far in my testing FarCry has proved this persons theory, however, Doom3 has not.

[GUNNaH]

This is a great thread ... I thought I was going crazy but I see, clearly, that many people have run into the same situation as myself. Looking forward to your stats @mdOCer.

GUNNaH

[diehrd]

I have noticed this in a CPU and GPU,,And the way to tell for a CPU is with Si Soft Sandra..You will notice the scores decrease at a higher over clock.And the system may even run stable...

With a GPU you will notice in the 03 series the nature GPU test when naturally the frame rates drop as a result of the screen details involved ...The nature test is a good example i have noticed going from 112 frames down to 12 .. And i have noticed a lighter over clock go from approx 112 to 19 in the same section of the test.

GPU
My best score is 11349 stable run 24/7 However if I increase my over clock I see it drop in score to 10689..Big difference with a small increase in overclocking...Could be heat ? Could be a vCore issue who knows but the drop off is present and I have witnessed it on every system I have ever overclocked.

CPU
Si Soft sandra at 3291 per procesor I get like 20000 if i break 3300 I score between 17500 and 18300..(Arithmetic test)

SO when you get close to max O/C it is possible to witness this decrease in performance..And a reason WHY it is a must to check every o/c with a bench mark test from the beginnning to end,,