LTTLabs Article - Power Supply Testing Method Update - Protections

[Lucas - Labs]

In the recent LTT video discussing our power supply testing methodology we shared a relatively dramatic failure rate of 23%. This isn't the rate of power supplies that have failed to meet an efficiency tier, it is the rate of power supplies where two samples have fully stopped functioning during testing.

 

While our tests don't involve any outrageous abuse and I consider them to be quite fair, they are certainly more onerous - creating a larger testing load/burden - than typical use in a gaming rig or workstation. There are pros to more onerous testing and strict requirements like finding only the best of the best and ensuring that anything we "recommend" is essentially bulletproof. However, our primary goal is to provide useful information to the audience, and sometimes that is best done with testing that is more representative of typical use. There is a balance to be struck.

 

continue reading on the LTTLabs website...

[Toryze]

Adding a failure analysis would be VERY appreciated! With that, I'd also expect a parts analysis of the PSUs to be done too, and that would be absolutely fantastic and greatly appreciated, especially for power-users. For example on your failure analysis, I would like to know specific parameters on what went wrong and where (example being with the Asrock Steel Legend in its respective article, since the only reference I have for it dying is mentioned with "brownout testing" in a video about dangerous PSUs at the 9:57 mark, or with the Gamemax PSU exploding (which I can only assume something horrifically failed on the primary side).

About your 100PLUS rating, I personally think that would be more trouble than it's worth. It would introduce extra logistical problems (creating a grading criteria just based on protections functions and getting brands to recognize and use it) which I don't think would be necessary compared to just verifying if the protections work in the review. 

[Lucas - Labs]

Thank you for the input Toryze.  We have considered a parts analysis, but I don't think that people are actually that interested in a part-by-part component analysis.  99.99% of readers will have no idea what the parts are or any reference on if they're good or not.  It also requires quite a bit of effort in actually identifying the components, they can be potted or just obscured by other components.  Even if a power supply comes with all of the correct components, a lot of the performance and reliability relies on the implementation which can be even more difficult to fully determine or describe.

 

The Asrock is a peculiar example because both samples we tested actually did just die in the first test that we run, the brownout testing.  We can sometimes identify components that have exploded or been scorched, but often the failures aren't particularly catastrophic and it requires in depth knowledge of the circuit design to diagnose an issue.  Then in the end, I'm not sure if the regular audience cares that much about what exact component failed, they will just "write it off" as an option when they know it has failed.

 

I'm sorry to disappoint, but the 100PLUS rating was just a joke, I really don't think that people are making purchasing decisions on the protections unless they're particularly bad.

[OhYou_]

I've always wondered what your methods are for confirming the psu is dead. In my experience, I have killed a few and just set them aside, only for them to start working again after like 30 minutes.
I assume this is because of a PTC having to cool down and reset, or some other form of safety delay, so I'm curious if this is accounted for in the tests or if it is pronounced dead the moment it doesnt turn back on after shutting off.
23% seems surprisingly high is why

[Lucas - Labs]

A very good question OhYou_, and to give an unsatisfying answer, it really varies.  Some power supplies lose their magic smoke or will engage the protections on our AC source when power is applied but others just don't react to being connected to power.

 

When a power supply has just "died" in the middle of testing, the 'trouble-shooting' order typically goes through just trying to restart the automated testing a couple of times, disconnecting it from the test apparatus and powering it from regular 120V wall power on a bench, opening it up to inspect for scorched components, and then trying to apply power again.  I don't have a specified wait period to retry applying power, but because I'm doing other things it is often some time in between attempts to apply power.  Some power supplies are then sent to the manufacturers for further investigation.

[OhYou_]

45 minutes ago, Lucas - Labs said:

A very good question OhYou_, and to give an unsatisfying answer, it really varies.  Some power supplies lose their magic smoke or will engage the protections on our AC source when power is applied but others just don't react to being connected to power.

 

When a power supply has just "died" in the middle of testing, the 'trouble-shooting' order typically goes through just trying to restart the automated testing a couple of times, disconnecting it from the test apparatus and powering it from regular 120V wall power on a bench, opening it up to inspect for scorched components, and then trying to apply power again.  I don't have a specified wait period to retry applying power, but because I'm doing other things it is often some time in between attempts to apply power.  Some power supplies are then sent to the manufacturers for further investigation.

oh that seems pretty solid then, thanks for the clarification.

[sgreenberg]

Please test the protections and how much the psu goes over its recommendations. After the LTT video where Linus took about a 500w seasonic psu to 700w I wonder about others. 
 

Regardless, protections testing is needed for those who don't understand counting wattage when getting a psu (or are just pushing it but want to remain safe).

[PainoCat]

If you genuinely think that the power supplies that have been failing are perfectly good, then I would go for it. I personally think that as long as some ~$100USD (affordable) power supplies can meet the standard testing that you have put in place, then they should be adequately praised and recommended. Such praise might be diminished by saying similarly priced power supplies that wouldn't complete the test are a "pass". Definitely not a simple decision and I think it's awesome that you're reaching out for feedback! 

[Narr]

If I bought a passing PSU because of your testing, I might not want it to listed beside perfectly good PSUs when it’s known to fail in some permanent way when mildly overloaded/browned out or risk having it take something else out.

 

if you’re testing the protections, they’re advertising protections, and you know they’re not meeting that maybe at least give any passing grade a huge asterisk and describe what it will kill when given the chance.

[SiNevesh]

This sounds like a solid idea as long as it's still clearly marked that it *did* fail in some way, keeping the "Pass" score clean of failures. The blog post's example of "Protections Fail" would satisfy that. Thanks for the PSU testing you guys do, my latest purchase was guided by your reports!

[AndrewAsd]

I might've mentioned this before but I would like to see more testing of cheap power supplies, particularly ones from known brands, like the Corsair CX/CV series, EVGA W1/W2, Seasonic S12 and so on, given they're the most likely to have issues and are fairly popular, as oppossed to the higher end models like the Corsair AX/HX/RM series which are less likely to fail. It would be nice to know which (if any) cheaper power supplies can be trusted. 

[JohnMcBane]

If a PSU produces magic smoke or behaves in any unpredictable manner, the situation is pretty clear. But are PSU generally expected to recover from a OCP/OPP situation ?

Maybe higher-end models are designed that way, but cheaper ones may be not. I wouldn't categorize it as a failure when a device behaves in an intended way, I'd rather call it a missing feature. And if the warranty terms covers those situation, which for PSUs are generally very long, I don't see it as problem. The manufacturer should just clearly communicate, what the consequences of a OCP/OPP situation are.

[elfensky]

It depends on what the goal is:

 

- If the goal is primarily to help users find a good PSU they can buy, I don't see a need for a separate category - why would I ever risk buying a PSU that partially failed. I will simply filter them all out and only look at the good options. 

 

- If the goal is more of a wiki/information aggregation, then sure - having more filters and subgroups to filter by can be useful. 

[Seccedonien]

I personally think it would be a very useful distinction especially for people who want to check if a PSU they already own and use is safe enough or if they should (consider) replacing it based on this extra information.

[silentdragon95]

On 4/29/2025 at 5:46 PM, Lucas - Labs said:

We have considered a parts analysis, but I don't think that people are actually that interested in a part-by-part component analysis.

I am, a lot even! I was a little disappointed by the power supply tests on the labs website to be honest, since I was hoping for something on the level of jonnyguru.com (4+ years later their level of testing is still unmatched).

 

However, I do realize that there are probably literally dozens of us, so you guys have other priorities. I'd like to point out though that a parts analysis can provide helpful information about the expected lifespan of a PSU. A cheap PSU might perform just fine in the tests, but cheap capacitors for example will fail earlier than high quality ones. A parts analysis could also help to determine whether a higher price is justified or not.

[Toryze]

12 hours ago, silentdragon95 said:

I am, a lot even! I was a little disappointed by the power supply tests on the labs website to be honest, since I was hoping for something on the level of jonnyguru.com (4+ years later their level of testing is still unmatched).

 

However, I do realize that there are probably literally dozens of us, so you guys have other priorities. I'd like to point out though that a parts analysis can provide helpful information about the expected lifespan of a PSU. A cheap PSU might perform just fine in the tests, but cheap capacitors for example will fail earlier than high quality ones. A parts analysis could also help to determine whether a higher price is justified or not.

Doing a failure/part analysis as well as being transparent about their methodology would also open their reviews to third party/peer scrutiny, which I believe is critical to maintaining the credibility of LTT Labs' claims about a PSU dying. A big part in properly testing stuff involves substantial evidence to back up claims. Stuff like reviewing PSUs are really complicated, and despite LTT Labs' desire to make PSU reviews as digestible as possible to everyone, I do strongly believe it's against the best interest of everyone if their data simply doesn't align with others' tests and claims without proof.

[Toryze]

On 5/3/2025 at 9:31 AM, JohnMcBane said:

If a PSU produces magic smoke or behaves in any unpredictable manner, the situation is pretty clear. But are PSU generally expected to recover from a OCP/OPP situation ?

Maybe higher-end models are designed that way, but cheaper ones may be not. I wouldn't categorize it as a failure when a device behaves in an intended way, I'd rather call it a missing feature. And if the warranty terms covers those situation, which for PSUs are generally very long, I don't see it as problem. The manufacturer should just clearly communicate, what the consequences of a OCP/OPP situation are.

At its very basic function, it should prevent the death of other connected components in the event a protection must kick in. If it recovers itself and the connected parts safely and within the ATX specification, that is preferred. The exact consequences of an OCP/OPP situation depend on how the protections were designed or tuned. For example, some units have implemented OCP, but it is set too low to pass power excursion tests, so some units may provide the allotted power, but don't account for power spikes. Other times (more often than not), it could set too high or missing, which could pose a risk to the PSU itself or other connected components, and/or allowing it to drop out of ATX spec.