5870 goes to no video during OCCT

[hat]

So I bought a 5870 off my friend for cheap, but he says it had some minor issues. Frame drops and such, but just in one particular game. I just got it today and ran it through the OCCT: GPU torture machine to see what's what. It ran strong for about 15 minutes or so, no errors... then, no video. I ran this test once before today while I was away and I came back to no video, but the fan was still screaming. My guess is, the test was still running all that time (because of the fan speed), but I just wasn't getting any video. What could cause a card to suddenly output no video like that?

[Flibolito]

15 minutes huh how are temps for those 15 minutes. Could be over heating. Usually clock related issues cause drivers to recover and you should have signal after a few seconds so I'm guessing stock clocks is what your testing at. Sounds like temps or maybe a 12v rail getting weak. Depending on the PSU I would see if another rail (if there is one) could make that go away.

Other ideas would include underclocking and driver changes and possibly other stress tests such a unigen heaven for a few hours to see if it happens there too.

[hat]

I have no idea how the temps were. I had GPU-Z open, but not logging. I'll run it again soon with a log.

I don't have the ability to change rails on my power supply. I have one PCI-E at one end of the connectors, and one far end of the other side. They all use a generic 6-pin pci-e lookalike connector. My friend had the fps loss issue in his rig too. I'm not sure what power supply he had... I know it was Ultra brand, not sure of the wattage... I think maybe 650w. I know my GTX260, which also takes two pci-e power connectors, same as the 5870, didn't give me the no video issue, and I even had room to overclock and overvolt it, although not as far as others claim to have gotten...

Yeah, stock clocks. I'm not interested in underclocking it... but another stress test seems like a good idea.

[Flibolito]

yeah if you got a 260 overclocked and volted and it fine that the PSU should be solid. Try Heaven and let it run on stone giant which is nice too if stable after 1 or 2 hours I would try some games, if everything except occt is stable i would just not worry about it. As long as you can game for hours and have no issues you should be ok, let us know what happens.

[Mussels]

my guess is power. i had similar issues with my 4870's and a 600W OCZ PSU - it'd trip the OCP on one of the rails on the PSU.

[hat]

Originally Posted by Flibolito

yeah if you got a 260 overclocked and volted and it fine that the PSU should be solid. Try Heaven and let it run on stone giant which is nice too if stable after 1 or 2 hours I would try some games, if everything except occt is stable i would just not worry about it. As long as you can game for hours and have no issues you should be ok, let us know what happens.

It will be used for folding, so I want to make sure it's rock solid godly stable before I put it to work doing that.

Originally Posted by Mussels

my guess is power. i had similar issues with my 4870's and a 600W OCZ PSU - it'd trip the OCP on one of the rails on the PSU.

You had two? Doesn't the 4870 use more power than the 5870? My power supply is only 50w under that and a quality model too, so I think I should be in the clear there.


//I ran a test while taking a walk. It lasted 30 minutes, so if it ran anything like the last 2 tests, it blanked out to no video during the test. When I came back, I had video, and it said the test completed without errors. I checked the GPU-Z log and the highest temp I saw way 85c...

[Mussels]

it can still be a good PSU and FAIL, quality just lowers the odds of it failing. was just saying i had similar issues in the past, and it was the PSU.

[hat]

Looks to me like the card just "shuts down"... check the gpu-z log

[hat]

I ran Unigen, and my screen blanked out to "Not Support" instead of "No Video"...

[Mussels]

Originally Posted by hat

I ran Unigen, and my screen blanked out to "Not Support" instead of "No Video"...

was this immediate, or after a few minutes?

[hat]

I managed to run the benchmark past the 10th stage or whatever the first time. Now it seems to happen soon after it loads up, without even bechmarking it.

I had an idea... what if I took the power supply out of my server (ePOWER EP-500NE-XV 500W ATX12V v2.2 / ESP12V v2.91...), jumped it and used it to help power the video card? It does have 2 PCI-E connectors, but one is daisy chained off the other. I figure if I use the e-power to power one of the PCI-E connectors on the card, and my main unit on the other, it should be enough to make it work if it's a power issue...

[Mussels]

Originally Posted by hat

I managed to run the benchmark past the 10th stage or whatever the first time. Now it seems to happen soon after it loads up, without even bechmarking it.

I had an idea... what if I took the power supply out of my server (ePOWER EP-500NE-XV 500W ATX12V v2.2 / ESP12V v2.91...), jumped it and used it to help power the video card? It does have 2 PCI-E connectors, but one is daisy chained off the other. I figure if I use the e-power to power one of the PCI-E connectors on the card, and my main unit on the other, it should be enough to make it work if it's a power issue...

its a decent idea to test.

[Jack Doph]

^
Probably a good idea, although I, personally, have zero trust in OCCT, as it throws me an error ALL the time, no matter what setting.
IOW, I can pass all tests in everything (including F@H) and *never* have an issue, but OCCT tells me I have errors regardless.
Perhaps a glitch in the program?
In any case, a spare PSU is always worth a go

[hat]

To satisfy my curiosity, I downclocked to 600MHz in CCC (left the RAM alone) and it ran Unigine all night, never crashed. I'm going to work soon, I think I'll try 700MHz.

[Mussels]

Originally Posted by hat

To satisfy my curiosity, I downclocked to 600MHz in CCC (left the RAM alone) and it ran Unigine all night, never crashed. I'm going to work soon, I think I'll try 700MHz.

well... at least it doesnt crash when you do that. usable workaround.

[cheesy999]

Originally Posted by hat

Looks to me like the card just "shuts down"... check the gpu-z log

woah -1c - what sort of cooling do you have on that

it could be to do with the fact that you've got 3 psu rails, that dosn't really leave many amps for each individual rail and your card might be tripping some sort of protection in the psu

its probably to do with the cards VRM however as all of the recent dual slot cards end up crashing under furmark etc - thats why they brought in power limiters - just don't run occt and you should be ok, run crysis or Metro_2033 as they usually stress the card quite well

[hat]

Originally Posted by Mussels

well... at least it doesnt crash when you do that. usable workaround.

I wouldn't be too happy with running it 250MHz slower all the time...

[Mussels]

Originally Posted by hat

I wouldn't be too happy with running it 250MHz slower all the time...

i'd be happier than not running it at all.

[hat]

What do you guys think about this power supply?

SILVERSTONE ST75F-P 750W ATX 12V v2.3 / EPS 12V 80...

I see there's no power switch... I suppose if I ever ran into a situation where I needed to cut the PSU power I could just unplug it from the wall. Would that be any worse for it than flipping the power switch usually found on the back?

Originally Posted by Mussels

i'd be happier than not running it at all.

True, but then I may as well run my GTX260... and there's always the looming thought that the card itself is borked and a new PSU won't fix it.

[hat]

Originally Posted by hat

I managed to run the benchmark past the 10th stage or whatever the first time. Now it seems to happen soon after it loads up, without even bechmarking it.

I had an idea... what if I took the power supply out of my server (ePOWER EP-500NE-XV 500W ATX12V v2.2 / ESP12V v2.91...), jumped it and used it to help power the video card? It does have 2 PCI-E connectors, but one is daisy chained off the other. I figure if I use the e-power to power one of the PCI-E connectors on the card, and my main unit on the other, it should be enough to make it work if it's a power issue...

About this idea...

if I jump the psu, then it would be always on, or at least on before I turned the computer on. I would be feeding power into my 5870, even though it's off. Doesn't sound like a good idea...

[Flibolito]

There are some ways of wiring it up to where they both start and stop at the same time. I have done this before but it will void any warranty whatsoever. Do you have a friend with a decent system you could test it in. Running below stock clocks is definitely not a long term solution but as a previous post said the VRM might be weak or dying, does the card make any squealing noise under heavy load? My gtx 480 does during OC scanners that run 6000 fps but i don't use those for that reason. If it squeals alot under heavy load without a voltage bump/super high clocks it probably the VRM crying and might be the problem. I think the PSU and rest of the system is fine but to be sure another system would rule out ALOT of what it could be.

BTW mine squeals because its @980MHz @ 1.152 volts with the VRM @ 427 MHz.
At stock there should be no squeals.

[Mussels]

Originally Posted by hat

About this idea...

if I jump the psu, then it would be always on, or at least on before I turned the computer on. I would be feeding power into my 5870, even though it's off. Doesn't sound like a good idea...

IIRC you link up the green wire, and they power on/off together.

[Flibolito]

Mussels = correct

[hat]

Sounds like it involves cutting up wires on both power supplies... I would really rather not do that. This is a temporary test, not a permanent solution.

I've underclocked and undervotled my CPU to 1.6GHz at 1.0v. By my calculations, it should be taking less than 20w, as opposed to the 80w it was taking when overclocked.
65(1600/2800)(1/1.425)^2 = ~18.3

If it were my power supply, that should free up enough room for the GPU to eat up more power...

[Mussels]

Originally Posted by hat

Sounds like it involves cutting up wires on both power supplies... I would really rather not do that. This is a temporary test, not a permanent solution.

I've underclocked and undervotled my CPU to 1.6GHz at 1.0v. By my calculations, it should be taking less than 20w, as opposed to the 80w it was taking when overclocked.
65(1600/2800)(1/1.425)^2 = ~18.3

If it were my power supply, that should free up enough room for the GPU to eat up more power...

lol, no cutting. personally i used something like the leads on a multimeter with the 'needle' point on the end into the main ATX connector, and then a paperclip to connect the other end of that cord to the aux PSU.

remember that its just for on/off, no real power is flowing through it (still, use insulated connectors/wires)


oh and since you didnt seem to get it - i mean the end the wires come out of the ATX connectors, not the plug end. you can still get contact from there.