I've seen several times the PSU tier list being put here, and they just a gross oversimplification of how things work with several misunderstandings at best, or complete marketing BS at worst, So I'll go into detail.

First, lets dive into the methodology of the tier list at LTT.

Links: https://docs.google.com/spreadsheets/d/1eL0893Ramlwk6E3s3uSvH1_juom7SMG5SCNzP2Uov8w/edit#gid=1214219159

https://linustechtips.com/topic/1116640-psucultists-psu-tier-list/

Tier A and above - required to have ZVS (LLC or phase shift resonant topology), have ripple below 50mV under all load situations including 110% overload, <3% 12v voltage deviation from nominal. Have fan with =>30k hours MTBF fan (generally non-plain-sleeve bearing) or be fanless. Have all 105°C capacitors.

There are several problems with these requirements:

For one, ATX specifications set the requirements perfectly, any PSU that meets them is complaint and safe and can be used in any PC. (In fact it would be a problem of the rest of the PC if it needed a power supply with ''better'' specs than that).

For example, noise at 12V rail at 120 mV peak to peak max. (btw 120 mV peak to peak of noise is already a very good value in general switch mode power supply design, and this is the noise of the power supply alone tested in special conditions (PSU alone connected to a resistive load, a few capacitors IIRC), turns out that graphics cards and motherboards add a lot of capacitance to the system, greatly reducing the noise after all.

What I mean is that, even a power supply that has a noise above the ATX spec, it will most likely still work without issues on most PCs. And here we are in 2020 creating list of power supplies with ripple under 50 mV xd.

If you want a PSU with that low ripple go for it, but it has to be clear that you're not getting anything from it.

And more importantly, a PSU meeting those tier requirements does not tell you if the power supply is more reliable or safer by any mean. The PSU being LLC resonant just means that it is more efficient (and also more complex and expensive).

It could be argue that a LLC resonant design is in fact less reliable than a old half bridge due to its complexity, in those power supplies they even use ICs as bleeder resistors and mosfets controlled by ICs do to the work of a simple diode.

And it also could be argue that requiring DC-DC would make the PSU less safe, because it is easier to have a overvoltage fault in this case.

For example all you need to have 12V at your 5V rail is the transistor of the DC-DC buck converter failing short (and that can be caused by ESD). In older power supplies without DC-DC it is not that simple, and basically impossible because the 5V and 3.3V regulation is done at the primary side. In other words, if the primary side transistors that regulate the secondary voltages fail in those older power supplies, what happens is that the power supply stops working, because the primary is isolated from the secondary by the main transformer, the transistors at the primary either stop switching or fail short which in both cases translates to the secondary having the output voltage dropping to 0, you can't have DC thru a transformer.

In these old power supplies the only way for a overvoltage to happen is via a open feedback fault, which is a very VERY rare thing (the feedback loop is not something that is stressed in a power supply), and if anything you quadruple your chances of a open feedback loop in a DC-DC power supply because:

There's the main feedback loop thru optocoupler for the 12V rail.

Plus:

A feedback loop for the 5V buck converter.

A feedback loop for the 3.3V buck converter.

A feedback loop for the -12V voltage inverter.

Now of course I'm not saying that a DC-DC power supply is not safe, after all the power supply has the monitor IC that will shut down the computer if that were to happen (And they also often have open feedback protection on top). But with this information it can be said that the power supply with DC-DC is less safe.

Moving on, The <3% regulation requirements for the 12V rail sounds pretty and all, but useless in practice. contrary to popular belief computers in general are in fact very tolerant to over voltage and under voltage situations.

For example, your modern CPU has a vcore is about 1V, how this vcore is created thru a synchronous buck converter in the motherboard, the so called VRM, which I don't like as a term due to its generic nature, it is like calling a GPU a computer component all the time. And the funny thing is that when we talk about DC-DC power supplies here you do often see the term buck being used, but whatever, moving on.

It works on the principle of negative feedback (feedback loop),just like modern power supplies can work with a voltage range from 90 to 265V, your computer internally also has quite a range. testing the actual limits is something I always wanted to check in a lab. in theory if all protections can be disabled (because turns out motherboards are also adding over and under voltage protections) most PCs might be able to work with a voltage range from 8 to 16V. The lower limit would be determined by the maximum duty cycle that the internal buck converters of the PC can have, and the upper limit is due to some having 16V capacitors in the 12V rail.

The only things to watch out for would be the hard drive motors, No idea of what would happen if the motor of a hard drive spins faster or slower than usual.

ATX spec sets the overvoltage protection at a maximum 15.6V.

The internal ICs and transistors of motherboard often have a max voltage of 25V, it might be possible that a few motherboards might be able to work with a input voltage all the way to 20V. xd

And If anything, given how nvidia measures the power consumption of their cards, having a higher input voltage leads to better performance. There was a video of a guy testing a gpu with and without daisy chained power connectors and found out that with daisy chained connectors the GPU performed a bit worse, this is because of the voltage drop in the wires, at lower voltage the GPU (due to the fact that it also uses a synchronus buck converter) increases its duty cycle and draws more current from the power supply, and since the GPU measures its current input to determine its power draw, it will think that it is at the power limit sooner at 11.5V than at 12.6V for example.

I feel like I'm giving overclockers an idea, this is something that has to be done carefully, you would need to reverse engineer a good part of the motherboard/GPU to determine if it can take something like 18V.

I think this obsession with voltage tolerances and ripple levels can be traced back to the old times when PCs where significantly different than now. In the old times your CPU vcore was 5V or 3.3V, basically the voltage being put out by the power supply was the voltage getting to your CPU directly, in those cases noise levels and voltage tolerances played a major role. But now that everything has its own regulation not so much.

And the minor rails are being phased out, nobody uses them anymore, server power supplies got rid of them, and OEM makers were implementing 12V Only power supplies into their systems for a decade already, and now Intel has created the ATX12VO standard that hopefully will be adopted quickly.

And here we are in 2020 measuring the performance of power supplies in unrealistic cross load scenarios. In fact this whole better cross load performance was the thing that lead to the DC-DC creation, basically we created and problem to sell a solution.

Now on capacitors, there are a few things that need to be clear out. the current obsession of capacitor quality was the result of the capacitor plague.

https://en.wikipedia.org/wiki/Capacitor_plague

The thing is that the capacitor plague ended over a decade ago, there's no longer a huge failure rate in cheap capacitors, if anything what has happened is that Japanese manufactures took it as an opportunity to start price fixing:

https://passive-components.eu/japanese-electrolytic-capacitor-manufacturer-settles-criminal-antitrust-case-and-pays-60-million-fine-for-illegal-price-fixing/

And fans, well I don't even know where to begin, we never had a fan plague or anything similar, I still have several sleeve bearing fans that I removed from very old almost 20 year old power supplies that died as result of the capacitor plague that still work perfectly.

And you would be surprised of how common this is in general that old recommendations still persist. I still often see people recommending practices that help with the lifespan of NiMH cells on lithium devices, even though NiMH hasn't been a thing in portable computers in the last 20 years of so. One of those is 'wait for the battery to drain completely before recharging' which was a thing to avoid the memory effect of the nickel batteries, it is not a thing in lithium or any other battery chemistry.

Tier B and above - required to have OTP (just claimed is enough), be based on ACRF or ZVS primary, DC-DC secondary topology, and have APFC with full VAC input range; voltage regulation and ripple should be in spec under 110% overload; 12V ripple below 100mV under 110% overload on all rails; should be rated for continuous operation at at least 40°C ambient

Now here it is something good, requiring OTP, now it is important to be clear that not having OTP while not complying with the ATX standard it is still safe if the power supply is able to give its rated power at high ambient temperatures, and that is very easy to make (just use overrated parts in your design, it can be in fact cheaper than just implementing the OTP).

The funny thing is that this tier list says that the N1 is Potentially dangerous in multiple scenarios.

Even though it passed all tests: https://www.jonnyguru.com/blog/2015/07/13/evga-400-n1-power-supply/

While it fails to mention that the B3 series had a problem where they would fail if overloaded which I really can't stress of how much of a design error that is.

The B3 series blows up because the maker didn't calibrate the OPP correctly, in other words, the power supplies will blow up because someone didn't the right value resistor in them.

https://www.tomshardware.com/reviews/evga-450-b3-psu,5160.html

https://www.tomshardware.com/reviews/evga-750-b3-power-supply,5229.html

https://www.tomshardware.com/reviews/evga-850-b3-psu,5186.html

So now that you read this wall of text you might be wondering how do I know to pick a good and safe power supply, and well this can be resumed in a very short list.

Comply with all ATX12V requirements.

And even if it doesn't (lacking OTP) as long as the power supply can give its output power at high ambient temperatures (like the N1) it will work.

And for those advanced users, please try to learn a little on how the power supplies themselves work instead of just comparing designs, like going into the schematics, reverse engineering, etc. I was reading comments on that tier list and I often find people talking about the changes of several revisions of PSU manufacturers, like they see one model and already know who made it, while at the same time provide say nothing regarding on how the power supply works, advantages or disadvantages of topologies, etc.

I really feel like the PC enthusiast community is turning into something like the audiophile community where you have people buying all types of expensive equipment based on complete misconceptions of how DACs or Opamps work.

Like just to give an example, part of the audiophile community doesn't like digital sound because apparently 'you can hear the steps' on the wave. This is something that even hardware makers at taking advtange of in their marketing.

For example when comparting 16 bit vs 24 bit you will often see less steps on the 24 bit waveform, and that leads to the conclusion that 24 bit sounds 'better'.In reality that stepped waveform goes to the low pass filter of the DAC, and you end up with a smooth waveform that was 99.99% like the original one (and the difference is due to noise at -90 dB or so). This is the whole Shannon–Hartley theorem.

WHILE AT THE SAME TIME, THESE PEOPLE OFTEN PREFER NOS DACs OVER REGULAR ONES.

NOS DACs are DACs without the low pass filter, those DACs literary output the stepped waveform, which just means there's a lot of high frequency noise in it, and they buy into that bs because it 'has a higher bandwidth'. omg.

I really don't want the PC enthusiast community to turn into something like it, so that's why I wrote this whole text.