Memory inductors on older Nvidia GPUs are failing

[TechyBen]

2 minutes ago, mr moose said:

Doesn't the whole card only cost $70 to build? But we are ignoring R+D costs etc. However that wasn't really my point,  the fact remains parts are failing, and if we improve one part then all we do is shift the weakest point to some other part on the board.

If it's a fault as simple as the inductors etc, they could easily* be swapped out. The difficulty is proving what part is at fault. Was it the VRM inductor failing, or was it a short in a poor PCB substrate???

 

*As to board repair, in general it's hard and a lot of skill needed. But on the scale of board repair, it's an easy swap for these boards by the looks of it, compared to other stuff that can go wrong.

[Carclis]

4 minutes ago, leadeater said:

I did just do a sanity check on what buildzoid was saying. He thinks the power delivery for the memory VRMs is from the PCIe slot and that the inductor failure would cause 12V on the output side and essentially a short circuit so we can do some really basic calculations on the power drawn from the PCIe slot.

 

I'm going to use 12A as that would cause failure.

 

12V x 12A = 144W

 

Correct me if I'm wrong but isn't the x16 slots in the PCIe spec rated for 75W? If this is actually the case I'm surprised no motherboards were damaged because more than 12A is likely to have been drawn for some of the failure pictures shown, more like 20A.

Well that is probably possible still. IIRC there were other cards that achieved numbers like that (might have been GTX 1050) but only momentarily. AFAIK the spec allows for this as long as it's not continuous draw in excess of the spec, like the RX 480.

[Trixanity]

2 minutes ago, leadeater said:

Correct me if I'm wrong but isn't the x16 slots in the PCIe spec rated for 75W? If this is actually the case I'm surprised no motherboards were damaged because more than 12A is likely to have been drawn for some of the failure pictures shown, more like 20A.

That's correct. The slot should be rated for 75W although I seem to recall some talk about you shouldn't even pull that from the slot if you can avoid it. The RX 480 pulled 75-85W, I believe, from the PCIE slot at launch which allegedly caused some cheap boards to burn out the PCIE slot. Huge oversight but it was fixed in firmware when it was discovered at least. You'd rather be a bit below than a bit above the spec. I think most cards try to keep it below 65W to be safe and pull the rest over the PCIE cables.

[leadeater]

2 minutes ago, Carclis said:

Well that is probably possible still. IIRC there were other cards that achieved numbers like that (might have been GTX 1050) but only momentarily. AFAIK the spec allows for this as long as it's not continuous draw in excess of the spec, like the RX 480.

Yea I'm just more surprised someone with a cheaper motherboard didn't lose both due to 200W+ being ramed through the slot when it goes to short like conditions.

[TechyBen]

Welp. This has put me off buying used (miner ) GPUs... but made me hopeful for "fixing" dead ones.

[Carclis]

5 minutes ago, leadeater said:

Yea I'm just more surprised someone with a cheaper motherboard didn't lose both due to 200W+ being ramed through the slot when it goes to short like conditions.

I think over-current is a lot more forgiving that over-voltage. Also here is the testing was thinking of. The 960 even goes over 200w at times.

[TechyBen]

My completely naive and untrained zero knowledge understanding of electronics is that lots of current makes things hot. So as long as it's short, quick and not constant, you get something hot, but it does not "melt". If it's for a longer time then *bad thing happen and the magic smoke escapes*. That or extreme current... but we are not getting that here.

 

Lots of voltage, that is really bad most of the time!

 

Also, is this NVidia, or the other brands? NVidia made the PCBs/main boards... have the other resellers put their own sub par inductors on? (Though these "reference" cards are also NVidia branded ones?).

[asus killer]

41 minutes ago, mr moose said:

Doesn't the whole card only cost $70 to build? But we are ignoring R+D costs etc. However that wasn't really my point,  the fact remains parts are failing, and if we improve one part then all we do is shift the weakest point to some other part on the board.

any product is as strong as it's weakest component, still that is no reason not to have better components, and even more if there cost is not that much relevant to the all product cost.

As i understand this is mostly a reference cards problem, so it even makes less sense as nvidia was a lot more money to spare.

[leadeater]

52 minutes ago, TechyBen said:

and the magic smoke escapes

I've tried multiple times in the past and I can never get the magic smoke back in to the failed part

[Carclis]

7 minutes ago, leadeater said:

I've tried multiple times in the past and I can never get the magic smoke back in to the failed part

That's the part where you make your three wishes.

[leadeater]

5 minutes ago, Carclis said:

That's the part where you make your three wishes.

Ohhh damn it! I've wasted so many wishes over the years, of course electronics parts are magic lamps but instead of rubbing them you have to over volt or over current them.

[Syntaxvgm]

These aren't old cards. Wow. 

[mr moose]

9 hours ago, asus killer said:

any product is as strong as it's weakest component, still that is no reason not to have better components, and even more if there cost is not that much relevant to the all product cost.

As i understand this is mostly a reference cards problem, so it even makes less sense as nvidia was a lot more money to spare.

But at what point do you stop using better components or use better QC?  People are assuming (and maybe rightly) that these parts are failing due to nvidia using/specifying underrated parts.  If that is the case then you can be rest assured it's not just the inductors that are underrated, they were just the weakest point for whatever reason. 

[asus killer]

26 minutes ago, mr moose said:

But at what point do you stop using better components or use better QC?  People are assuming (and maybe rightly) that these parts are failing due to nvidia using/specifying underrated parts.  If that is the case then you can be rest assured it's not just the inductors that are underrated, they were just the weakest point for whatever reason. 

ok, i'm just going by bildzoid analysis. I guess they may have used more cheap parts besides this, but that's ok if they aren't causing problems, this ones are.

I see your point, if this didn't fail other parts would be the weakest link and be the ones to fail, but that's just a theory. You could be on to something, no idea.

[mr moose]

Just now, asus killer said:

ok, i'm just going by bildzoid analysis. I guess they may have used more cheap parts besides this, but that's ok if they aren't causing problems, this ones are.

I see your point, if this didn't fail other parts would be the weakest link and be the ones to fail, but that's just a theory. You could be on to something, no idea.

The thing is we won't really know until we find out exactly why they are failing.   In my experience with such things, there are strong parallels to the white good industry, where brand X gets a reputation for part A failing while brand B gets a reputation for part C over heating.  We can sit and surmise that the washing machine just needed a stronger pump, but at the end of the day if they decided that pump was good enough then they likely also decided that the cheaper relay/controller board would suffice too.

[Toscomvl]

My 780 just died(R33 burnt), created an account here just to post that .

[SamHF]

Very sorry for bringing up an old topic but recently I beleive my 980Ti died. I took off the backplate and couldn't see anything which looked to be burnt, what exactly should I be looking for? Thank you.

 

https://linustechtips.com/main/topic/964629-very-depressed-and-worried-is-my-gpu-dead/

 

[ryao]

On 3/31/2018 at 12:20 PM, DrMacintosh said:

Nvidia Reference cards are having there memory inductors exploding!

Source(s): 

GPUs affected: GTX 780, 780Ti, 980Ti, and Titan X (Maxwell). And possibly more that are yet to be discovered. 

 

But its not the end of the world!

 

So it would seem that Nvidia is having more issues with older reference GPUs, granted these are old and reference but this should not be happening and is still a problem. It seems to be 100% a result of Nvidia cost saving measures. 

 

Its also a little interesting that LTT talked about on the WAN Show as to why Nvidia is not used in Apple products, Nvidia seems to have a tendency to cut corners to save cost, and they did it again. 

To be fair, ATI did too. My ATI AIW 9700 Pro had its fan die and then died. Then a ATI Radeon 9700 Pro that I purchased died killing the AGP slot in the process. These were built by ATI cards that very few people seem to remember existed.

[Just Monika]

That's one way to keep people upgrading their GPUs. I just hope my 1080 (reference PCB) holds out until the 30xx series.

[Stefan Payne]

On 31.3.2018 at 6:20 PM, DrMacintosh said:

Its also a little interesting that LTT talked about on the WAN Show as to why Nvidia is not used in Apple products, Nvidia seems to have a tendency to cut corners to save cost, and they did it again. 

The "Bumpgate" debakle and nVidias behaviour about that is probably one of the main reasons but also that they don't allow apple modifications.

AMD does and thus it is possible that Apple can do the GPU driver themselves and it is said that that is the case.

 

So there is a possibility that there are no AMD Drivers in Apple Products, only AMD Hardware

[asim1999]

wow now i figured out why my 980 Ti wont boot and it makes a short, high pitched click and then shorts out the system 

[Trik'Stari]

Would this happen on a reference card that is being water cooled?

 

 

nervously eyes his 1080ti FE

[Ashley MLP Fangirl]

so it's only reference cards? so my Asus DirectCU 2 GTX 780Ti should be ok?

[yian88]

nvidia is garbage, there is no month without a new controversy about their bullshit practices from any angle

[colonel_mortis]

This topic is from March. Please don't revive old topics, especially in Tech News.