Errors on MemtestG80, not on MemtestCL?

[Wolf]

Sorry...I'm not sure if this is the most appropriate forum for this or not...

I've been suspecting something is wrong with my video memory for a couple years now, as I get intermediate crashes, everything else checks out, and the crashes USED to be proceeded by weird graphical corruption...yet my GPU (a Geforce 9650m GT) seems to work fine...like it shrugs off Furmark, and usually is okay with GPU units (although often when I exit a unit the video drivers get rebooted)

Yesterday I stumbled across memtestg80 and cl. First run of g80 crashed my system almost immediately. Second run spit out an astonishing number of errors...I didn't even bother writing it down because it looked like hundreds of thousands or millions.

For the heck of it, I tried memtestcl this morning...and so far it seems to be running with no problem, and no errors.

So...I'm not really sure what, if anything, to make of this.

I've got Seagate's Momentus XT, the original one, and I've heard of some people having issues with them causing crashes, but the Firmware is all patched up, and the drive itself checks out with Spinright, so...hopefully that's not it. System RAM checks out with memtest...

I was hoping maybe I'd finally found what's wrong, but g80 spits out so many errors it's hard to believe my system would even boot like that, while cl isn't reporting anything, so... not sure.

[Wolf]

More info:

memtestG80 gives 4294961546 errors. It crashes my GPU (which Windows reboots), and lists that many errors. The CL version doesn't crash, and lists zero errors, at least in 50 rounds.

Soooo I don't know if these results mean anything at all.

[bruce]

MemtestG80 is the older code and was written strictly for the hardware interface with G80 compatible chips. I wouldn't trust it on your Geforce 9650m GT.

MemtestCL was a code rewrite to a general standard so it will run on a much wider variety of devices. There are some restrictions, however, in it's ability to address all of VRAM so if in the unlikely event you have a problem in an untestable region, it wouldn't find it.