OCCT GPU Error Check

[Rossi~]

Brilliant tool but is it necassary?
I can go beyond 785MHz on my GTX 470 core and be fine and stable in furmark (Kombustor) for at least an hour but anything over 725MHz gives me errrors in the OCCT GPU Error Check.
After browsing round it appears that this is the case for many people and most advise to stick to Furmark and other GPU thrashing tools and check for artifacts and crashes.

What do you guys think? It would appear that the OCCT GPU Error Checking is for a stable number cruncher e.g. F&H, and for pixel pushers you should stick to a Furmark-esque program.

 

[Rossi~]

The lack of response says to me that it may well be the case.

 

[novadaemon]

I've OC'd many cards and OCCT is my primary error checking tool. I'm well aware that nothing "visual" occurs in furmark, but to me an error is an error. I can get high OC's off of video cards, but I would never overstress components to where theres a risk that the errors turn into something worse.

What you need to do is isolate the errors and figure out if they are coming from the GPU or the memory. If a card survives 20 minutes OCCT stable using "Optimal" shader complexity with 0 errors then it's gonna survive anything you throw at it. I do use furmark for good measure though. just in case....

And yeah, I agree with your assessment. If you can't see an error while playing a game then it's no sweat technically... You don't want those errors when you are folding though.

 

[Rossi~]

Ah thanks for the reply dude I kind of agree with you about the errors turning into something worse, it kinda worries me. Then i think back over the years and how many graphics cards i have run into the ground with overclocking and they all survived up until upgrade.
That said, i'm going to push OCCT error check to the side and try and ignore it.

 

[[H]jimmy]

Personally I use occt error check for all of my video card overclock checking. It always shows my errors first as you have seen. Yes you can play games and pass benchmarks at higher overclocks but this is just like failing prime on your cpu and calling it "stable". Honestly I think everyone should run a full hour of occt gpu check with every overclock before posting it as stable. If you fail this check you will see artifacts sooner or later.

 

[Rossi~]

Damn you Jimmy, damn you to hell lol
I know i know, it would probably result in a much lower overclock but you can call it stable. But, but but... my overclock lol

I will ignore my girlfriend tonight and have a play around with my GTX470 tonight with OCCT error check and see whats what.

 

[martinje]

I will lower the mem a bit, and see if that helps. Usually I start with the GPU only until Im happy with the result, then I raise the mem and finally the shaders. Once thing at a time, I know its time consuming but thats how I do it.

 

[Xzyrus]

I think Jimmy is right. ATI's overdrive suggested 840/1080 (before I ran out of patience since it took > 20mins) for my 6850 and at that speed OCCT still detects errors although it was much fewer in the entire run compared to the first couple minutes at a higher clock rate.

I was able to run it at 950, 940, and 930 on the core with 1080 on the memory for pretty long periods of gaming/Furmark/benching before it 1) froze, 2) drivers crashed, 3) froze again at those respective speeds. I originally Furmarked for 10 minutes at each of those speeds before cranking it up by 10MHz at each step which obviously is not long enough but I was too "pumped up" with the new card.

Basically the less you stray away from what OCCT says is "stable" the less likely you'll bump into instability but then again we should kinda know that the moment we step away from stock settings lol.

Straight out of the OCCT FAQ:

OCCT stresses components alot. If an error was detected, your computer can generate those. they may happen everytime, usually under heavy loads. Thus, by ignoring this warning, you can have unexpected behaviour, for instance while playing, or while encoding a video...

Sometimes, this error isn't noticeable ( for instance, if in one frame of your favorite fps one point is not at the right spot, you probably won't notice it, and it won't crash your game). Sometimes, it is really noticeable ( if it happens in a critical CPU process, blue screen ! or if you were encoding, your video file will be corrupted).

Edit: P.S. When the card generates > 60+ fps in some games yet your monitor only refreshes at 60Hz, if my understanding is correct, there's a chance you won't see the frames with the artifacts.

 

[bezant]

Yeah, I thought I could push my Gigabyte until I tried OCCT. Anything over 900 Mhz, and I get errors. I see others pushing the cards over 1000Mhz, but it is hard to believe they have run OCCT.

 

[Elsie]

I just set it at something that doesn't crash 2 minutes of furmark and run games until it crashes then back off a tiny bit and carry on. Don't forget that neither occt or furmark really highlight memory instability (try bfbc2 or avp with some AA for that).

I always worry that running endless furmark and especially occt (vddc gets to 110C in seconds) sessions will significantly shorten the life of the card while only really being self-validatory. Obviously with this method I can't definitely claim that I am 100% stable (on the core) but then I've gpu crashed once in the last 5 months, which since I am only gaming I can live with, others cannot.

 

[Rossi~]

Wow more feedback! Thanks
I had a play about and it would seem that it's retty much core sensitive, memory doesn't have that much of affect on the errors produced, when i gets errors due to memory it's obvious in Furmark (as in it crashes or freezes). Also increasing vcore doesn't really help the errors, it's mainly lowering the overclock that gets rid of them.
Quite an anal stability test, but thorough.

 

[Ocellaris]

Yeah, I thought I could push my Gigabyte until I tried OCCT. Anything over 900 Mhz, and I get errors. I see others pushing the cards over 1000Mhz, but it is hard to believe they have run OCCT.

Well this same type of things always happens with CPU over clocking as well. There are a lot of people running "stable" over clocks that bomb out within a minute running OCCT Linpack. Once they see the OCCT failure, they claim it doesn't matter because everything else works fine for them.

 

[[H]jimmy]

The newest video cards will automatically adjust your video cards memory timings and settings if it detects errors, so you won't see many or any errors with the memory test.

You can test this by running benchmarks with higher and higher memory settings. At some point you will no longer see any improvement. This is your card relaxing timings. So your "1200 mhz!" memory might not be running any faster than at 900 mhz.

A lot of the benchmark runs you see guys on XS forums and such are bleeding edge suicide runs and will never hold up for lengthy gaming sessions.

If you are not passing an error check in occt, you will be having instability problems sooner or later whether you care about them or not. Why would anyone trust any computer part that can't compute numbers accurately?

 

[Elsie]

The newest video cards will automatically adjust your video cards memory timings and settings if it detects errors, so you won't see many or any errors with the memory test.

I don't find this, if I run at 1200 then I crash within minutes in AVP yet at 1185 all is well for hours. [H] also had a number of lockups in their original review during memory overclocking (http://www.hardocp.com/article/2009/09/22/amds_ati_radeon_hd_5870_video_card_review/12), so this EEC stopping vram related crashes is at the very least an exaggeration.

 

[sram]

Interesting topic. I was ganna write something like it but I was convinced by anandtech members that OCCT is actually the best tool to test your VC overclock and if you can't pass it, then your card have the potential to have problems sooner or later.

 

[[H]jimmy]

I don't find this, if I run at 1200 then I crash within minutes in AVP yet at 1185 all is well for hours. [H] also had a number of lockups in their original review during memory overclocking (http://www.hardocp.com/article/2009/09/22/amds_ati_radeon_hd_5870_video_card_review/12), so this EEC stopping vram related crashes is at the very least an exaggeration.

I did not mean that to sound like you could set your memory at say 10000 mhz and it would still run because it would self correct. I meant that if you benchmark every jump in memory speed increase you will see a point that your benchmarks no longer get better as you continue increasing your memory speed.

I have personally found that the memory on my gtx460s will go much higher than 1000mhz where it is set at right now. However I will not see any more performance from benchmarks or gaming or even memory throughput benchmarks.

At some point around 1200 or so mhz for my cards they will start artifacting or crashing.

Also please see this article, as this isn't new or previously unreported capabilities of the new cards.

http://www.hardwarecanucks.com/foru...6850-vs-gtx-460-1gb-overclocking-study-3.html

 

[Elsie]

Weird 'cos I just get a linear increase in performance right up until I reach massive instability, different card to you mind, maybe the EEC is just more aggressive on fermi.

 

[Rossi~]

Interesting topic. I was ganna write something like it but I was convinced by anandtech members that OCCT is actually the best tool to test your VC overclock and if you can't pass it, then your card have the potential to have problems sooner or later.

Indeed, i think it should be implemented as the way to test stability on graphics card overclocks, especially in reviews.

 

[NKD]

lol it is funny how some people say I dont like over stressing my components while they are doing the same thing running OCCT. when you get no errors running regular games means you are not pushing the card like the tool is pushing. I dont go by OCCT, i got my gtx 480 running at 850 and it has been stable under furmark I dont even bother with OCCT. it has been running fine for 3 months now and I have never had a crash.

Just because OCCT telling me I got errors doesn't mean the card is not stable. This things are meant to push pixels if you get no errors in that case then there is nothing wrong with running it at those speeds.

I think you will get errors if you are running the memory to high and I am not too surprised at that, if I bump the memory to high on my gtx it will heat up like crazy so I stay at 1000. I think the memory clock to high might give you errors sooner or later. I usually don't touch the memory until I have a stable core overclock. Becuase I don't think too much memory clock is going to help me when the gtx 480 is already pumping out 177gb/s bandwidth.

 

[HohnyF]

lol it is funny how some people say I dont like over stressing my components while they are doing the same thing running OCCT.

Idea behind that is to punish component(s) once, so they can run stable from there on.
And it's obvious you did some punishing too. With OCCT GPU that is

when you get no errors running regular games means you are not pushing the card like the tool is pushing. I dont go by OCCT, i got my gtx 480 running at 850 and it has been stable under furmark I dont even bother with OCCT.

Why should you even bother with furmark? It's not like it's your preferred day to day application.

it has been running fine for 3 months now and I have never had a crash.

Just because OCCT telling me I got errors doesn't mean the card is not stable.

Good solid wooden bridge is also stable.
Stable as long as it's peaceful villagers who walk across it, but he can't endure modern day traffic. Not the best of analogy I know, but "stable up to a point", would be more accurate description.

Large percentage of OC-ed systems are ORTHOS, prime, IntelBurn, linpack, whatever, unstable but they can last a lifetime and never crash due to hardware instability.
Some people are ready to tolerate this kind of "stability", some aren't.
Former can be pretty certain that CTD, BSOD or application crash is not due to their hardware being pushed to far, and later KNOW that they can't really throw anything on their system and be sure it will survive.

Now all that being said I was surprised that my everyday "stable" gtx 460 overclock 920MHz was far from OCCT GPU stable. In the end I had to settle for 870MHz

 

[[H]jimmy]

I already asked the question but here it goes again.

Why would anyone trust a computer processor that cannot calculate numbers correctly?

If you cannot pass OCCT error check you will have a crash sooner or later. Maybe a crash once every three weeks or months is fine for you, but its not a stable overclock. Furmark is probably creating errors if OCCT is, your just not seeing them.

 

[Ocellaris]

I already asked the question but here it goes again.

Why would anyone trust a computer processor that cannot calculate numbers correctly?

If you cannot pass OCCT error check you will have a crash sooner or later. Maybe a crash once every three weeks or months is fine for you, but its not a stable overclock. Furmark is probably creating errors if OCCT is, your just not seeing them.

Because its "good enough" for the user.

 

[Rossi~]

Exactly "good enough", though we here people throwing the word "stable" around, when they don't use OCCT Error Check, who knows if it's stable or not..

I mainly use OCCT error check now, i thought my overclock was stable when i used Kombustor, was fine in several games i played for hours on end then all of a sudden BAM game locked up, then a couple of days later it happened again. Did some research and found OCCT and used error check, errors all over the place so i droppede it back and all's good.
Have recommened it to so many people and it's found 2 cards to fail on stock clocks and had to get RMA'd.

But you know, if Furmark or Kombustor is "good enough" for the user then yeah fine, just be aware that it may crash someday/s or in extreme cases, never recover. Just please don't rave on about it being stable if your overclock were show errors in OCCT.