Repairabilty of an old GTX660Ti nvidia gpu...?

[eccerr0r]

I salvaged an old GTX660Ti but found out perhaps why it was dumped: it's not stable.

It boots fine, runs fine, but under heavy GPU load it works for a minute, but then artifacts up and hangs the machine. Sort of implies a heat issue but it's too consistent, crashes the same way though hard to predict exactly when it will crash, but crash it does.

This isn't exactly my first experience with dying video cards, though my RadeonHD 3650 and GeForce4 MX420 had a similar fate, but it has artifacts from bootup and truly unusable. Then there's the RadeonHD 5770 that won't let the computer even boot.

Is this nvidia card worth trying to see if it's possible to stabilize or is it ready for the dump as well? It's too bad, this thing should be significantly faster than any other video card I've ever had...

---

BTW the first thing I did was blow out dust from its HSF. It didn't seem way too dusty however, so dust clogs shouldn't have been the problem.

[stj]

Re: Repairabilty of a nvidia gpu...?

any non-polymer caps on it?
and dont overlook a failing ram.

[eccerr0r]

Re: Repairabilty of a nvidia gpu...?

Don't think it's ram, the way it fails is very consistent, just don't know when.
I haven't pulled the plastic air guide (course it's a double slot) so unsure of capacitors yet.

I tried bumping up the fan speed and didn't seem to help, still crashed. Forcibly reducing frame rate seemed to increase time between failures but this is only good on simple scenes... on dynamic scenery I may even have to reduce the frame rate to below that of my ATI R7, which of course means that I'd be burning more power to get fewer FPS - the R7 doesn't even need a direct connection to the PSU!

[stj]

Re: Repairabilty of a nvidia gpu...?

you need to strip it down, who knows what you may find in it!

[eccerr0r]

Re: Repairabilty of a nvidia gpu...?

Looks like all polys... as usual there's never an easy fix ...

[momaka]

Re: Repairabilty of a nvidia gpu...?

You can use GPU-Z or MSI Afterburner to watch the temperature as you put load on the GPU... though it's probably best to just remove the heatsink and check on the thermal compound. If it's dry, that might have broken the GPU already. And even if it's not dry, chances are the GPU chip is on its way out... or possibly the RAM. But with older cards like these, it's usually the GPU.

Easy "fix": give it a reflow and cross your fingers. If it does get "fixed" by this, make sure you keep the running temps nice and low. This should / could prolong the "fix". Aim for load temps under 50-55C... though I can tell you for a fact that this won't be possible with the stock cooler. So either you'll have to limit FPS / power draw or use a much bigger cooler (or water cooling if you have it?)

[eccerr0r]

Re: Repairabilty of a nvidia gpu...?

Unfortunately GPU-Z and MSI Afterburner don't quite work in Linux I don't think? nvidia-drivers does however have a utility to track temperature though tough to monitor it as it gets corrupted when it crashes.

Need a way to underclock this thing perhaps...

[ribcage]

Re: Repairabilty of an old GTX660Ti nvidia gpu...?

First run mats/mods to see if a memory channel is bad, then try https://gpuzelenograd.github.io/NVIDIA

[eccerr0r]

Re: Repairabilty of an old GTX660Ti nvidia gpu...?

hmm...the whole linux GPU scene is a royal mess IMHO, probably mostly because GPUs are so different from each other and there's no way to support them all in a unified manner. I did run across the vulkan memory tester but not sure what is needed to support vulkan and which of my cards support it or not...

[eccerr0r]

Re: Repairabilty of an old GTX660Ti nvidia gpu...?

I just tried nouveau on this card...while no artifacts, it still crashes -- but Nouveau seems to kill the X server when it gives up and I can just login again.
However nouveau seems sllllloooowwwww... at least for the benchmarks I tried. Perhaps I set it up wrong, need to check it again.

And eiew. binaries. that depend on systemd. eiew. my sysvinit system does not approve.

[stj]

Re: Repairabilty of an old GTX660Ti nvidia gpu...?

neuv' as far as i know has no hardware acceleration becaese NV wont opensource the drivers or internal register/commandset info.

[piernov]

Re: Repairabilty of an old GTX660Ti nvidia gpu...?

nouveau cannot change the clocks from their boot time defaults in a lot of cases, so you are stuck at like 200 MHz instead of 1 GHz of core frequency. This can also explain why there can be "less issues" running a dying GPU (or VRAM) with nouveau.
Even when it can do it (this GPU being a Kepler it may work), it does not do it automatically and you need to change power levels manually. Even with that, you cannot expect to perform as good as the closed-source Nvidia driver.