Does anything stress a GPU more than FurMark?

[555BUK]

I just purchased a Powercolour 4850 and have been running a few stress tests. Everything runs fine at 780core and 1125mem (Accelero S1 + 2x92mm and 1.2v) except for FurMark which blank screens at anything over 720core after a few mins.

ATItool runs 780/1125 fine (max temp 47deg)
Crysis runs 780/1125 fine (30+ mins and 50deg max)
GRID runs 780/1125 fine (30+mins and 48deg max)
3DMark06 780/1125 fine (45deg max) Linky
FurMark will only run 720/1125 (30+mins and 58deg max temp)

So, Furmark gets my temperatures 20% higher than anything else at lower core clocks. Even if I lower the memory to stock it still will not allow higher GPU clocks. I guess it is giving the Shaders a pounding, but is there anything else that can stress a GPU more?

edit: Idle temps are 31deg with the S1.

 

[Greebo]

No

 

[555BUK]

No

 

[Durzel]

Furmark is Prime95/ORTHOS for GPUs, and the same rules apply imo. Not Furmark stable = not stable.

 

[MoRT489]

But if it's stable for a 4-5hr gaming session not really sure Furmark is all that important to u...

 

[TheRealDeal]

If my card ran all my games fine with no problems thats all i would be interested in. If you can game at those clocks with no problems then it should be fine as obviously the games won't be stressing the card as much. If furmark stresses the card that much and games don't then its another 3dmark06 situation as 4850 scores low compared with slower cards but in games does the business.

 

[Durzel]

But if it's stable for a 4-5hr gaming session not really sure Furmark is all that important to u...

I've had PCs that have failed Prime after several hours that will play games, etc for hours without issue. Doesn't mean it's stable, just means that the possibility of an error is sufficiently rare that you are lucky - nothing more - to not hit it during a gaming session, or whatever.

If you can't rely on the card operating at 100% usage consistently at a given clock speed then imo it isn't stable.

Each to their own I suppose - the same arguments come up with CPU stress testing ("it doesn't crash in X so imo it's stable"). Personally I wouldn't be happy running a PC that can be shown to be unstable in a given application and by extension could crash at any given time in any application if certain criteria were met.

If my card ran all my games fine with no problems thats all i would be interested in. If you can game at those clocks with no problems then it should be fine as obviously the games won't be stressing the card as much. If furmark stresses the card that much and games don't then its another 3dmark06 situation as 4850 scores low compared with slower cards but in games does the business.

It's totally different - we're not talking about a card that scores poorly in a benchmark, we're talking about an overclock that fails completely in a given application.

 

[555BUK]

My 3DMark06 score is fine. Over 14k with a dual core is about the same as a stock 4870 would get. With a quad CPU my score would be closer to 17k.

 

[555BUK]

I would be interested to know what clocks gurusan can get FurMark stable. I'm pretty sure he has the most extreme 4850 on here.

 

[bledd]

i stopped running benchmarks years ago, now i just look for the best overall drivers for stability/performance (by reading forum posts) then stick with them for a month before considering looking again

have much more fun playing games instead of worrying about 3dmarks

i used to test cpu stability using toast.exe, but then realised i'd never use it like that anyway, so just dropped my oc's a tad and used a 'set and forget' attitude, less tweaking = more time playing games!

-just realised, this is pretty off-topic

 

[JTrickle]

I was struggling to get my 4850 stable on Furmark at 680 this morning. Temps not exceeding 55c. No point worrying about missing 5% for the sake of stability imho. Leaving mine at stock speeds for now.

 

[jarvo1]

Check out RTHDRIBL (I think), bit of a mouthful I know but thats pretty intense when you max out the AA and run multiple instances of it, Furmark is a lot quicker and easier though for almost the same results.

 

[Vertigo]

Furmark is Prime95/ORTHOS for GPUs, and the same rules apply imo. Not Furmark stable = not stable.

And it's still total rubbish. Just as a CPU which can't run ORTHOS for 8 hours yet runs everything else fine is "stable", so is a GPU which can run everything bar Furmark.

People need to stop being so damn anal about this "stability" rubbish tbh

 

[Durzel]

And it's still total rubbish. Just as a CPU which can't run ORTHOS for 8 hours yet runs everything else fine is "stable", so is a GPU which can run everything bar Furmark.

People need to stop being so damn anal about this "stability" rubbish tbh

If you say so. If you're happy to have a system that is by definition unreliable then that's your perogative, it's not for me I'm afraid. I want to be able to rely on my hardware 100% of the time without having to worry about whether "benchmark X" will push it over a threshold.

In the absence of an obvious software defect something that causes hardware to fail could happen in anything else - just because it's a benchmark by nature doesn't make it irrelevant. The sort of stress Furmark places a GPU under could occur in the next game around the corner, and equally some other application could stress the CPU to the same sorts of levels that Prime produces (HD video encoding is an example that springs to mind). Saying "if it works in the games I want to play then it's stable" is like driving a car with tyres that are guaranteed to blow at 70mph and saying "well I never go over 60 anyway".

 

[MDPlatts]

f@h

 

[titaniumx3]

I think CPU stability and GPU stability are two different things. If your GPU is unstable then (for most people who just play games) the worst you'll get is some graphical artifacts, blank screen, freezing or crash, where as if your CPU is unstable, then your data can become corrupt ... which is by far the worse scenario.

 

[Cyber-Mav]

If you say so. If you're happy to have a system that is by definition unreliable then that's your perogative, it's not for me I'm afraid. I want to be able to rely on my hardware 100% of the time without having to worry about whether "benchmark X" will push it over a threshold.

i agree with you 100%, stability is very important for me. thats why i use that Intel Burn Test tool for cpu testing and i use furmark for a 12 hour run to stress out the gpu to the limits.


as for the op's question, furmark is the best tool iv found for stressing the gfx card. my x1900xt has a current consumption guage which reads the highest when furmark is run. your 4850 i think has a current consumption meter too which will tell you how much current the core is consuming. run the stresser programs and see which one manages to max out your power drain.

 

[thegoonden]

Nothing stresses it "more"[1]
But some things stress it differently. For example, I can run Furmark at a given clock for hours, then run Crysis, and get minor artifacts in that.

As for CPU stressing and stability. Why would anyone test 4 core stability when no game (assuming we're talking about a gamer), uses more than 2?

If it crashes on Prime95, but never ever on anything else, it's like the guy who goes to the Dr and says "Dr, it hurts when I do this", and the Dr says "well, don't do that then".
Having said that, if it bombs out of P95 in seconds, I'd say it's still worth adjusting your settings until it can at least stay conscious for a little while.



[1] Except possibly "feature test 2" on 3dmark vantage, which would SLAM my 8800GTX's fan to 100% within milliseconds, even at stock clocks and even if the fan hadn't moved from idle for the rest of the entire run.

 

[Greebo]

And it's still total rubbish. Just as a CPU which can't run ORTHOS for 8 hours yet runs everything else fine is "stable", so is a GPU which can run everything bar Furmark.

People need to stop being so damn anal about this "stability" rubbish tbh

Totally disagree. Its bad enough with badly coded programs and drivers trying to decide what is at fault with the system without factoring gpu/cpu stability.

I would rather know that my cpu/gpu is stable at 100% worse case scenario load and at least that is not then a factor when trying to find out why a game or application crashed.

Plus, instability is instability however you dress it up. I certainly, albeit however unlikely, don't want my data corrupted after I have spent a long time working on a spreadsheet/document/etc.

 

[Vertigo]

If you say so. If you're happy to have a system that is by definition unreliable then that's your perogative, it's not for me I'm afraid. I want to be able to rely on my hardware 100% of the time without having to worry about whether "benchmark X" will push it over a threshold.

The problem is that stability is an arbitrary thing. You could test your GPU with Furmark for eight hours and decide it's stable yet, had you left it running, it could have failed five minutes later. You can extend this as far as you like - leave it running for a week then stop it and, for all you know, it could have failed five minutes later.

Take my overclocked CPU. I tested it using ORTHOS and it failed at eight hours and twenty minutes. Now the accepted level for stability seems to be eight hours, so if I'd stopped it as soon as I'd reached that value I'd have been considered "stable" by everyone, yet because it was left to run and failed twenty minutes later, is it now "unstable"?

The point I'm making is that I didn't worry about this total irrelevance and instead just used it. In the six months that have followed it's never one failed, regardless of what I've thrown at it. I, and IMHO any other rational person, would consider this a stable overclock, despite the fact that it failed ORTHOS after 8h20m.