| Detailed description |
|
|
Hello Mr. Imran
I have a card which has definitely two small areas in its ram since production which are faulty (I can see it in textures sometimes)
Its a GTX660Ti with 3GiB.
It is curious tho, since i can sometimes see information flow in those rectangles, as if two memory addresses go to the same destination, instead of being direct faulty rams.
Usually i try to "capture" this usually by opening a lot of browser tabs until i saw the error in one of the images, which worked great in windows xp, but since win8/win10 and its custom graphics management, it seems not to work reliably anymore.
I live with this card since around 4 years (?) and of course i am already on the go to buy a new one by now
however, not only did it spark multiple projects where i try to spawn textures until i find the faulty area, etc. it also created a lot of anger; and led me to your project which i now try to use to verify the problem.
Now to my idea. Given one can track down the "faulty areas" in a graphics card, is it possible with CUDA to reserve that area? E.g. even by creating a texture for an application, which keeps running? Would such a feature combined with the error check not be very useful to "temporary fix" the broken card?
Do you know of any tool which does this, and, since you seem to be very very versatile in CUDA, do you think, this is even feasible?
Greetings
Gabor Guzmics
gab@g4b.org
|
|
