Much Ado About Everything

About two weeks ago I came home from work only to find my screen blank.  No amount of coaxing would let picture be shown on any of my two monitors, and only upon reboot I’d get to see the picture again.  But the picture was not right – as if the video card (an EVGA 7600GT CO) got sick; white letters of the boot menu were marred with random pink stripes all over.  This is usually a sign of damaged video memory.  I took a look inside my case and found that the cable feeding the video card’s fan was disconnected.  So it was likely the card died due to overheating.

But wait a second there.  This was no ordinary heatsink/fan combo – it was top of the line Zalman VF Cu-900 (or something like that) pure copper beast, which I got for the purpose of having as quiet of a video card as possible.  While the heatsink it comes with seemed appropriate – it was made from copper too and was reasonably large considering this is only 7600GT – the fan it’s equiped with is loud.  Very loud.  This Zalman was much quieter, though not as quiet as you’d expect.  But at the lowest setting it was acceptable.

As I keep fiddling with my machine constantly, I must’ve somehow pulled the cable without noticing.  In addition to this, this was on the hottest day this year so far.

But it’s not that simple.  You see, the card was doing nothing when it died.  Display was in shutdown mode and did not have to produce any picture.  Combine that with a very large copper fan (for a video card) and the fact that when I touched it I found it too hot to keep my finger on, but not as hot as to get instantly burned.  From my experience this means that the temperature was about 60C.  Now, my previous card which was passive out of the factory (no fan, just huge heatsink) – Gigabyte 6600GT – was running 70 to 75C in idle.  And about 110C while running at full load!  While 7600 is die-shrinked compared to 6600 (90nm compared to 150nm if I remember correctly), it should still be easily able to handle this kind of temperature.

I took the card out and sent it off for RMA – incidentally, its replacement has arrived but I still have to go to the post office to pick it up.  But I was thinking about what had actually happened.  I figured that it was really unlikely that the graphics chip itself died.  It didn’t run any hotter in idle without fan than it would be with fan and under load, so I couldn’t see a problem there.  I figured out that perhaps some other component died, most likely memory – especially since the picture pattern I saw indicated memory problems – because no other component has a heatsink.  So basically, the graphics chip can handle a lot but the fan is on it not just to keep it cooler but to keep the heat off the card as a whole.  That made sense.  But I still did not expect it to die – what if I was playing a game?  The graphics chip would get just as hot and while there’d be some airflow, it shouldn’t make such a difference to kill other components.

Quite some time later while reading anandtech forums I ran into a post claiming widespread problems with latest generation of Nvidia video cards.  It seemed like people were having their cards die on them after only a short while (I had it only for a month myself) and the frequency of incidents was far above the expected.  It looked like failures were of the same type that I experienced.  The thread went on and on, with lots of flaming in every direction – as threads such as this usually do – but at the end someone posted a link to an article on hardocp, which seemed to shut up the naysayers.  This article exposed this problem and tried to get to the bottom of matter.  And what they discovered is that some manufacturers were having problems because they released video cards overclocked from the factory, without paying attention to what the parts they overclocked are actually capable of.  Specifically, they overclocked memory beyond what it was supposed to be capable of.  Surprise, surprise!

Now my problem made perfect sense.  First, it is sometimes hard to buy a non-overclocked card because the market is so competitive that manufacturers keep releasing more-and-more overclocked cards in order to get “ahead”.  Although what they suceeded was getting ahead in the “I have more RMAs than you” game.  The card I got definitely has overclocked memory as well as core.  There are no heatsinks on memory, however.  I don’t think it’s a good idea to overclock memory when it’s right next to a really hot graphics chip  – especially if you’re not equipping it with heatsinks of its own.

So now I felt much better.  It wasn’t really my fault that the card died.  It was running at the very edge of its specs (actually, beyond) so the smallest thing could get it to die.  When the replacement card comes back, I’ll stick some heatsinks on the memory.  Or perhaps I’ll lover the clock of the memory so it’s actually within the specs.

Technorati Tags: video card, nvidia, overheating, rma