Question What's all this PSU/GPU Ungracefull Failures , Anyhow ?

[antonr]

Hi

I am sort new here on TH and I very much enjoy perusing the GPU sections. What is noteworthy is the reports of GPU failures related to PSU fault modes.

That worries me . It is 2020 and GPU's are very very expensive , a sort of most valuable asset . Also relates to CPU/Motherboard of course but I want to use the GPU as an example. I have read a number of TH articles and is most interesting and relevant . I understand the reasoning behind a tier list , and I understand that
a very good PSU is very good PSU , it will for sure increase MTBF (mean time between failure). I get that . I understand why the senior members on forums will 10 out of 10 propose a good quality power supply.

... BUT ...

1. There is a producer (PSU) and a consumer (GPU) . The consumer must be equipped to also look after itself . I infer from failure modes of GPU that GPU's are susceptible to out-of-spec transients or low power capability of PSU. This makes no sense to me , that the GPU does not look after itself on the 12VDC feed rail (arresting transients , detecting low voltages and gracefully reject, limit current (where is the good old slow blow fuse ?).

2. The OVP and OCP fetures of the PSU are most effective , most of the times . But they themselves are electronic subsystems which can and will fail . Where is the good old slow blow fusing workhorses to complement that . A lot of people are talking about wires that burn . That should never allow to happen.

3. A top quality Seasonic PSU can and will increase reliability , I get that , but are we not relying 100% on this device , whereas other factors must be also taken into account in order to prevent an ungraceful degradation ?

Am I missing something , perhaps the gurus can help me here ?

 

[13thmonkey]

You are right, components should have enough protection to look after themselves, but do they? Probably not. As to the idea of a slow blow fuse, when dealing with Xnm features on a GPU i suspect that there are micro seconds before damage becomes serious, this may be just too expensive to implement more than once in a system. How far do you go? with a PSU the voltage it could output for a fraction of a second under a fault condition could be kilovolts, with the wrong physical location it could even bypass the wiring and jump to the board. There will be ways of hardening, but these will be expensive for the level of rarity when using good components. All you can do is mitigate risk (although this is just true for life in general). There might even be specialist insurance policies out there..

 

[antonr]

You are right, components should have enough protection to look after themselves, but do they? Probably not. As to the idea of a slow blow fuse, when dealing with Xnm features on a GPU i suspect that there are micro seconds before damage becomes serious, this may be just too expensive to implement more than once in a system. How far do you go? with a PSU the voltage it could output for a fraction of a second under a fault condition could be kilovolts, with the wrong physical location it could even bypass the wiring and jump to the board. There will be ways of hardening, but these will be expensive for the level of rarity when using good components. All you can do is mitigate risk (although this is just true for life in general). There might even be specialist insurance policies out there..

Thank you , I copy that as good answer , you probably right , with a GPU of > 2048 cores one could expect some nasty rate of current changes on the 12v feed . I get the feeling that 12v must be on its last leg , as GPU's enters the 150W - 400W consumption . Will not be surprised if somewhere along the line of new specifications , that we might see a bump to 24V , which I think will address a lot of issues , for starters current will be half at the same power requirements. To draw these currents from 12v is becoming ridiculously high . 48V will become too high as we start getting voltage isolation issues.

Well the long and short of all of this is that I will have to replace my 5 year old perfectly working PSU with a new one in order to proactively protect my valuables. I think 5 years is enough service , do you agree ?

 

[Mezoxin]

A GPU as the rest of the PC components has an operating envelope that they are designed to work with safely , which is normally set by industry guidelines such as the ATX and intel guidelines , such guidelines are the results of decades of industry experience , and they decide which is the most practicable methods of reaching their reliability goals

Sure its possible that a gpu manufacturer can add discrete protections to the GPU to be additional or redundant to that of the PSU , but that would be very costly and therefore not practicable , for example the PSU has a lot of real estate to occupy larger components that can filter out large ripples , the GPU however have a much smaller real estate dedicated for that and cant host the same components.

Also you have OCP , OPP , UVP , OTP all of these are protections features that to some extent offer some sort of redundancy to each other in certain failure scenarios so the probability of failure of all of them at the same point of time is quite low

you have to remember at the end of the day these are products for a consumer pc and not a nuclear power plant, for a consumer PSU with the relevant state of the art Protection features would in the worst case scenario cause the average user losses of several hundred dollars , but even then this is a low probability scenario

 

[13thmonkey]

it's only 12V before the VRM's, after the VRM's (i.e. on the way to the chip) the voltages drop further, and current increases, so you're only getting the benefit of lower current in the places where you can apply better heatsinks and have the the space to work, i.e. on the board not the chip.