7900GT mem dying???

[DTU_XaVier]

I was testing my new 7900GT's oc-capabilities, and installed ATI-tool... Seeing that the mem was only reading as 330, I thought there was some driverfault, automatically decreasing mem-clock to half... So I started moving the slider up, until it started showing artifacts in windows... I understood my mistake, and backed down again.... kept testing, and hit 1630 after an hour or so... started playing Quake 4, but when I exited the game (was running the mem at 1500 then), the screen started blinking, not reacting to clicks with the mouse... Every time the screen flashed, it was as though Windows had trouble popping back in...
I couldn't restart, so I hit the power-button... Going back into windows, I tested with 3dmark05 to see whetever that made the same thing happen.. it didn't...
Today I tried figuring out what the course was, and came to the conclusion, that the GFX-RAM suddenly wouldn't even go higher than 675 with neither Coolbits, nor Rivatuner (using the driver-tests)... Before my mistake, the auto-oc utility in Coolbits took the mem to 1564 no problem... The card has no problem running at default speed, so my question is, if I have fried my GFX-mem's oc-abilities... No RMA is possible, as the card operates fine at stock speeds... Returning the card on my right of return would cost me a good deal of time and money, and with examns coming up, I do not have the time...
What do I do??

Best Regards

[NickS]

Why can't you RMA it? BTW this sort of thing happens to my Dad's eVGA 7900GT CO SC when exiting Far Cry. He's never OC'ed the card since he's had it 'cause of what happened to the last one..

Nick

[DTU_XaVier]

They would test the card at stock speeds, and as they would find no errors there, would just bill me for wasting their time...
Been testing a bit more the last few hours, and though I haven't had the RAM at 800 yet, they seem to be coming around again.. The internal tests show green at 750 at least... Will run a bit of Quake 4 to see whetever the problem persists

BTW...
Why does your dad own a 7900GT??

Best Regards

[alexio]

Been testing a bit more the last few hours, and though I haven't had the RAM at 800 yet, they seem to be coming around again..

This is normal. The IC's at the 7900GT cards are a little odd. Sometimes they suddenly loose stability at a certain speed and later they can run at this speed again, so no need to worry aslong as the card is always able to run the stock memory speed. This odd behaviour is probably the reason for the 1.4ns binning that looks very conservative in the first impression.

[DTU_XaVier]

This is normal. The IC's at the 7900GT cards are a little odd. Sometimes they suddenly loose stability at a certain speed and later they can run at this speed again, so no need to worry aslong as the card is always able to run the stock memory speed. This odd behaviour is probably the reason for the 1.4ns binning that looks very conservative in the first impression.

Well, I just thought it was my own stupid mistake that had blasted my mem-chips... will give it a day or two of testing, then we'll see... Probably just needs a bit of burn-in
Played some Quake 4, and it seems they're back to somewhat old performance, will try an auto-detect, and see what happens ...

*Update* Just tried an auto-detect... Locked the comp as described earlier... But otherwise the clocks seem fine... So I*ll just not use auto-detect ....

Best Regards

[[TAG]Imp]

is this still a problem with newer 7900 GTs?

[DTU_XaVier]

is this still a problem with newer 7900 GTs?

Mine was bought last monday, so I suppose it is

Best Regards

[[TAG]Imp]

Mine was bought last monday, so I suppose it is

Best Regards

agh... that's bad news...

[alexio]

Mine was bought last monday, so I suppose it is

Best Regards

I want to add something. To keep the 7900GT cheap, the power components on the PCB are just capable of supplying the power the cards need when they run at stock. Particulary the mofsets for the memory are not designed to cope with high current and they will overheat. Because the memory OC's so high percentage wise the current that goes true these mofsets can rise by over 50% from stock. They will heat up to above their mamimum operating temperature of 75C. At this point the voltage supplied to the IC's will fluctuate, and thus stability will decrease.

So when you overclock the memory to high levels you have to make sure there is enough cool air flowing over them. Sinking them is an even better idea.

[DTU_XaVier]

I want to add something. To keep the 7900GT cheap, the power components on the PCB are just capable of supplying the power the cards need when they run at stock. Particulary the mofsets for the memory are not designed to cope with high current and they will overheat. Because the memory OC's so high percentage wise the current that goes true these mofsets can rise by over 50% from stock. They will heat up to above their mamimum operating temperature of 75C. At this point the voltage supplied to the IC's will fluctuate, and thus stability will decrease.

So when you overclock the memory to high levels you have to make sure there is enough cool air flowing over them. Sinking them is an even better idea.

It's got fresh air going right over the MOSFETs, as it's being sucked in by the CPU heatsink/fan... But I have been thinking of sinking them, just have to figure out how

Best Regards

[DTU_XaVier]

A little update... Just cut up the stock heatsink from my ASUS A8N-SLi, and "glued" these onto the MOSFETs and power regulator on the card, using thermal paste... Have just started oc'ing again, and though Coolbits won't let me oc to 800 (had to use RivaTuner, no internal test for that), it does not start flashing the screen as it used to, now it just gives an error message as per normal... So seems those small flat pieces of metal were just what the doctor ordered

Best Regards

[ManagHead]

The mosfets you are talking about, which side are they on? I do also have one IC one the back which become very hot!

[DTU_XaVier]

They're on the back of the card, in the lower left corner (if we say the socket is the bottom)... It's a small cluster of black ICs....

Tested a bit more, and it seems the problem persists, to a degree... I played some Quake 4 (the game which first made the error present itself, so I test in that), and it ran clean, also back to the desktop... tried oc'ing again, just to see what would happen, and it locked again... it's as though the card is not aware there's something wrong with the MOSFETs, it just keeps going (somewhat)... But it's more stable now than before the sinking, though I might need som bigger, more professionally made ones for absolute stability@OC ...

Best Regards