Concerns about Video Memory

[nintendoeats]

I've been running FAH on my ATi 7970 almost nonstop for almost a year now. Recently I have started to develop concerns about my video memory from artifacts I've been seeing in some games, and that further makes me concerned about my Folding output. I ran MemTestCL and it gave me 0 errors except in the "random block" test where it reported astronomical numbers of errors. As such I have two questions.

1: What can I infer from that result?
2: Could my F@H client be sending back bad data?

Thanks!

[toTOW]

If your memory errors generate computation errors, the core should be able to detect it, and then end WU prematurely with an error.

In you FAH log, do you find WUs that don't end with the FINISHED_UNIT status ?

[nintendoeats]

Nope, it's totally clean and I have never noticed any prematurely ended units. I will be cautiously optimistic.

[Joe_H]

One other thing that might be causing the high rate of errors, insufficient cooling. Have you cleaned out dust and lint from your video card and the rest of your system recently? I ask because a number of cards I have seen are pretty good about cooling the GPU chip itself, but depend on unobstructed air flow over the video RAM to cool it sufficiently. A little bit of dust or lint there can lead to the RAM overheating and increase the error rate.

[nintendoeats]

I'll double check, but I doubt it. I have an aftermarket GPU cooler that is much more efficient than stock, so even if it was less than optimal I would still expect it to provide adequate cooling to a stock clocked card.

[nintendoeats]

I just took a peek. I did discover that I'm out of compressed air, but the cooler was totally clean. The industry standard "touch the heatsinks" test also revealed that the RAM heatsinks are dissipating heat. I've put a great deal of effort into perfecting airflow and cooling in that case, so I have to assume that the problem lies elsewhere.

[nintendoeats]

Hmmmmmmmm. It would appear that the random blocks test is known to give false positives on ATi cards. Given this, I am going to run it overnight and check for anything OTHER than random block errors.

[foldy]

I had problems with my HD7950 after some years where F@H or Furmark made my PC crash.
The temperatures looked good but to be sure I replaced the thermal paste on the GPU which can dry out after some years of heavy use.
No problems anymore till then

[nintendoeats]

I had problems with my HD7950 after some years where F@H or Furmark made my PC crash.
The temperatures looked good but to be sure I replaced the thermal paste on the GPU which can dry out after some years of heavy use.
No problems anymore till then

That's interesting to know, though given that I put that cooler on at about the same time I started folding that isn't likely to be the problem here.

[djay86]

I also have this Graphic cards on wich i've folded more than 40 million points with it, and I know that fah doesn't put that much stress on your GPU not as much as some game and benchmark.
I suggest you raise your vcore a little with saphire trixxx, since the more load you put on your GPU the more the vcore will drop. the vcore should not go under 1.15volt with stock frequency.

[wilding2004]

I also have this Graphic cards on wich i've folded more than 40 million points with it, and I know that fah doesn't put that much stress on your GPU not as much as some game and benchmark.
I suggest you raise your vcore a little with saphire trixxx, since the more load you put on your GPU the more the vcore will drop. the vcore should not go under 1.15volt with stock frequency.

I would have to disagree. F@H put's a lot more stress on the GPU than any game ever has. The GPU is under a constant load of 90%+ when running F@H. Can't think of any game that has ever made that sort of GPU demand. What games do stress far more than F@H, is the graphics cards memory system.

[bruce]

True.

Moreover, when a GPU is reportedly 90%+ busy, that one number represents some combination of (memory_busy + shaders_busy + whatever_busy}. In other words, all 90%+ busy conditions are not the same. Putting it another way, all benchmarks are not the same.

[djay86]

When I say load it is not the percentage load that you can find on GPU-Z, it is actually the amount of current the GPU draw.
While folding it never exceed 75 amps while in the benchmark "Valley Benchmark" it actually draw 100 amps on average and it goes up to 120. I'm sure some game now can make the GPU draw this amount of current and make the GPU voltage drop a whole lot.
Some HD7950 don't have that kind of voltage drop but some do, mine do and probably the OP.

[davidcoton]

amps or watts?