Can a overclocked GPU become unstable without overheating?

smjpl

Hi. I have Radeon HD 4890. I let AMD Catalysts Control Center do an auto overlclock and it increased the:
GPU clock from 850MHz --> 980MHz
&
Memory clock from 950MHz --> 1105MHz

The temperature is still only 33C @ idle & 52C @ full load. (Water cooled)
Can the GPU still become unstable even though the temperature is so low? These cards can keep working up to around 100C.

I ran a torture test with FurMark.
Results:
(See attachment but it doesn't really tell you that much.)

The test didn't really give me much info. I didn't see any streaks in the doughnut. Does that mean it passed? Or is is just a case that if the computer crashes or overheats, its a fail?

I ask this because I was playing Battlefield 3 (on fairly low graphics settings) and it crashed but unlike any crash I've every seen before (usually BSOD).
The screen that BF3 was playing on went completely black and the second screen had grey & black horizontal stripes across it. (I have dual screen set up with extended display) I just had to reboot the PC and she was back to normal.

This could be a coincidence that it just crashed like that after I OC'ed the GPU but I'm fairly clueless when it comes to overclocking so I thought I'd check with you all first.

Any help or advice is appreciated.
Cheers all.

ED E

Noob OCer myself but AFAIK you can only ever go so far on stock core voltage, have you increased yours?

With a watercooled setup you should be able to push the envelope significantly, I only joined the ATI team at 5 series(5770Hawk)(excluding my laptop but we wont go there) so I'm not sure what one should expect from such a card. Googling didnt throw up much at a glance other than somebody mentioning getting 1ghz.

/uselessrant

eF. Godlike

ED E is right, if you want to go further you could increase the voltage supplied to your GPU with software like MSI's afterburner. (assuming the 4890 is supported that is, if not you'd need to use a different program but AB is really easy to use)

To answer your question, yes a card can be unstable regardless of it's operating temperature, keeping the card cool will help you to overclock further but it doesn't give you limitless headroom.

It sounds like catalyst just tried to push your card too far and it's unstable at those clock speeds.

Also, furmark is great for heating things up and checking if temps are okay but it's not very useful as a stress test in my opinion.
I can run furmark at 940MHz core on my GTX470 but I need to back down by at least 20MHz to play BF3 or pass an OCCT artifact scan so chances are that catalyst pushed it too far.

If you can get atitool it has a built in stability test with an artifact scanner but it's pretty old at this point and no longer supported.

I'd also recommend OCCT's GPU test but only if you've got decent cooling on your 4890's VRM, it and things like furmark have been known to kill some video cards due to the amount of power they take to run.
If you've got a full-cover block or if your VRM has a decent heatsink on it you should be fine.
Explanation of VRMs in case you need it.


If I were you I'd probably work the clock speeds up slowly and only change the memory or GPU clock between stability tests as it's easier to diagnose the problem when you've only changed one thing.
Run short stress tests between the increases until it fails. (a minute is what I do)
When you reach a point where you can't increase any more you can either use an overclocking tool like rivatuner or afterburner to add voltage and continue upping the clock slowly and testing or you can back the speed down a little and run a stress test for 15-20 minutes, if it passes with no errors you've found your stable overclock.

Sorry for the long post and what is probably a confusing format, I was editing as I went along.
If you're unsure on anything ask and chances are I'll be back here at some point.

ED E

+1 on Afterburner, its clean and precise and has a plugin for my G510's display which is pretty fecking handy.

smjpl

Cheers for the reply lads. I have had a few headaches since I posted that message as I went with the self thought overclocking method. My card is still alive so all isn't that bad at it.

Yea I got MSI Afterburner and started messing around a bit. The thing is, the stress test with MSI Ab (Kombustor), makes my card unstable even when it is at stock core voltage, core clock & memory clock. After a minute or so the screens just go black and I have to restart the PC (I take it this is instability). That doesn't seem right that it should fail on stock everything. Temperatures are still not a problem. If I up the core voltage from 1.312V --> 1.325V it does become stable for the Kombustor test. The max voltage I can apply is only 1.35V so there is no room left to up the core and memory clock. If I try increase them it becomes unstable so I need to up the voltage and I don't really want to put the card up to its max voltage. At this stage I think I will just leave it. I don't think I need to OC it. Its an old card. Also after going through that article "Explanation of VRMs in case you need it.", I realized that I have no heat sink on the VRM or the memory. There is a good airflow over it but no point pushing my luck. Think I will just wait till I upgrade my PC.

Cheers lads.

ShadowHearth

I started messing around with my gtx 570 and I am not impressed...

When it says that ideo drivers stoped responding does that mean its unstable, right?

There is only one setup I managed to get working without crashing heaven and it's very very little increase on stock voltage. It looks like I get less performance with higher voltage.

I tryed a few setups other people had stable on 570 and for me it just goes to same " video drivers not responding " thing

A bit annoyed...