[EOL] PSU Tier List rev. 14.8

[Juular]

5 minutes ago, Mezoxin said:

dialing in that long window that would be sufficient for transients  isn't really possible with the type of hardware that is readily available in current analog PSUs thats why its not done by manufacturers , it is possible though with digital PSU's but the feature isn't available 

But still, connecting a single GPU with two connectors should lower current requirements for OCP to quite mundane ~25A in transient for even the most power-hungry GPUs out there so my point is still valid.

[Mezoxin]

2 minutes ago, Juular said:

But still, connecting a single GPU with two connectors should lower current requirements for OCP to quite mundane ~25A in transient for even the most power-hungry GPUs out there so my point is still valid.

yes that would be best but that would make the psu more expensive (single 8 pin pcie  per cabe = more cables = more cost and less cable management) and there will be no market for it  as consumers care more for cable management and aesthetics

 the problem is that you would be wasting two pcie cables with 4x8 pin connectors of that psu to power a single gpu , which is not the best solution to accomodate for ocp trips because of 10ms spikes 

 that extra cost could be directed towards making extra circuitry to implement what you said earlier ( also JG said the same) of having a time dependant ocp  

[Juular]

9 minutes ago, Mezoxin said:

yes that would be best but that would make the psu more expensive (single 8 pin pcie  per cabe = more cables = more cost and less cable management) and there will be no market for it  as consumers care more for cable management and aesthetics

Most (all?) high-wattage multi-rail units that can accomodate for multiple-GPU setups already have enough cables to connect up-to three GPUs with 2 cables each.

9 minutes ago, Mezoxin said:

 the problem is that you would be wasting two pcie cables with 4x8 pin connectors of that psu to power a single gpu , which is not the best solution to accomodate for ocp trips because of 10ms spikes

That's kinda pointless to argue over since we don't manufacture PSUs here and can't change it but if we set say 25A OCP, even with insane 40% transient spikes that makes about 200W power draw room per cable, 400W with two of them, more than enough to power 95% of GPUs available. Well, and if you're powering smth even more power hungry it's kinda warrants getting high-end digital unit anyway so you can configure OCP a bit higher if you still manage to get it tripped.

[Mezoxin]

1 minute ago, Juular said:

Most (all?) high-wattage multi-rail units that can accomodate for multiple-GPU setups already have enough cables to connect up-to three GPUs with 2 cables each.

That's kinda pointless to argue over since we don't manufacture PSUs here and can't change it but if we set say 25A OCP, even with insane 40% transient spikes that makes about 200W power draw room per cable, 400W with two of them, more than enough to power 95% of GPUs available. Well, and if you're powering smth even more power hungry it's kinda warrants getting high-end digital unit anyway so you can configure OCP a bit higher if you still manage to get it tripped.

What I meant is if a manufacturer released an analog psu with multiple 12 v rail psu with ocp set at 20-25A , would he put 2 daisy chained 8pin pcie connectors on it or just one ?

 

if he put 2x8pin , you will find end users ranting on amazon and newegg and giving it a 1 star review saying how terrible this psu is because its shutting down during load. 

 

if he wants to play safe ( avoid all the bad customer reviews ) he will just put one 8 pin pcie connector per cable , but in order to be able to keep up with the competition he must produce a unit with the same number of pcie connectors as them , so the end result will be more  pcie cables ( double )

 

 

[Juular]

7 minutes ago, Mezoxin said:

you will find end users ranting on amazon and newegg and giving it a 1 star review saying how terrible this psu is because its shutting down during load.

That's the problem, 40A solution with daisy-chained cables are fool-proof in a sense that it will not trip, but it's also dangerous and just plain wrong to power very power hungry GPUs with just one cable. So manufacturer needs to decide, to play safe and keep things as they are now or change OCP to 25A, include just a couple daisy-chained cable for whose who power not so power-hungry GPUs that still require 6+6\8 pin cables and all other cables single 6\8-pin for those who understand that you need to plug two separate cables if your GPU needs two PCI-e power connectors and possibly slap a sticker on PCI-e ports \ bag with cables that this unit wouldn't deliver more than 200W per connector because it's safer (no-one reads manuals anyway).

[Mezoxin]

@Juular

as you said nobody reads , and although that would be the end users fault , in the end it doesn't matter as the end result would be the users ranting about the product and giving it bad reviews and affecting the sales of this unit ( you would be surprised about how many people buy products based on customer reviews only )

so the only way to avoid this would be to maker the implementation fool proof , which is a single 8 pin per cable or time dependant ocp , which equals more cost for the manufacturers for a feature that almost nobody  in the market cares about or even asking for it 

 

 

[Juular]

1 minute ago, Mezoxin said:

@Juular

as you said nobody reads , and although that would be the end users fault , in the end it doesn't matter as the end result would be the users ranting about the product and giving it bad reviews and affecting the sales of this unit ( you would be surprised about how many people buy products based on customer reviews only )

so the only way to avoid this would be to maker the implementation fool proof , which is a single 8 pin per cable or time dependant ocp , which equals more cost for the manufacturers for a feature that almost nobody  in the market cares about or even asking for it

Nobody asking for all these protections modern PSUs have, nobody asking for independent regulation and sub ATX-spec performance, no-one asking for multi-rail in the first place, yet all this are here. But i agree that these things are hard to change, industry as a whole needs to work toward this, maybe it should be in ATX spec that per-rail OCP on multi-rail units shouldn't be higher than 30A but giving that Intel for some reason pulled multi-rail from ATX spec in the first place it's highly unlikely that this would happen ...

[Mezoxin]

Some of these are raised by technical departments and if they are feasible the get the green light to implement them ,

for now what I am trying to say is that we should trust expert opinions when they say a prime platinum or AX is better than X unit that has multi rail . Because he knows what he is talking about 

[Spotty]

1 hour ago, Juular said:

But i agree, 40A OCP on multi-rail are almost certainly too high, there's very few GPUs that need more than 20A through PCI-e connectors, and if you have a GPU this power hungry you should hopefully connect it with two separate power cables anyway. So per-rail OCP should more commonly set to 20A long-window.

 

1 hour ago, Juular said:

That's why i said '20A long-window', it could still have 40A short-window OCP to account for transients, but again, if your GPU draws 40A on transient, you should use two PCI-e power connectors and bam - problem solved.

Edit: i do agree that it probably commonly set 40A just to be a fool-proof, average gamer who just bought his second-hand AMD Fury would most likely use just one PCI-e cable because it's easier and it has two connectors on it, so why not ? That needs to change.

 

54 minutes ago, Juular said:

But still, connecting a single GPU with two connectors should lower current requirements for OCP to quite mundane ~25A in transient for even the most power-hungry GPUs out there so my point is still valid.

 

2 minutes ago, Juular said:

maybe it should be in ATX spec that per-rail OCP on multi-rail units shouldn't be higher than 30A

@Juular everything you have said appears to be under the assumption that each connector on the PSU has its own dedicated 12V rail.
Different connectors does not necessarily mean different rails. PSUs don't have individual rails for every connector. You could have 2 VGA ports on the PSU but they could come from the same 12V rail with a 40A limit. Whether you're getting that power through one cable or two won't make any difference for OCP if they're on the same rail.

For example the Bitfenix Whisper M 850W has 4 VGA* ports on the PSU with 2x 12V 40A rails for the VGA ports. 2 ports share a 40A rail. If you have 2x RTX2080Tis installed then each card is going to be drawing power through the same rail, despite having 2 separate cables plugged in to the PSU.

Same thing goes for Peripheral connectors, which seems to be what started this discussion with @Mezoxin linking to posts where the peripheral connectors have melted as they exceeded the 8A rating for the connector despite the units having multi rail 12V OCP. In most power supplies the Peripheral connectors are all going to share the same rail. There might be 5 peripheral ports on the PSU that share a rail rated for 25A or 30A. You can't limit the rail to something low like 8A to protect the connector of one cable from melting because once you plug in the second cable that shares the same rail you could be tripping OCP unnecessarily, and people will be returning their units thinking they're faulty.

The only way what you're proposing would work is if PSUs monitored each connector individually, and that's going to be cost prohibitive to implement.

[Mezoxin]

As far as I know for the rmi 850 Each pcie\eps  cable connector on the back of the psu has a 12vocp of 40A which supplies every 2x8pin pcie connectors   , and the 24 pin and the peripherals have a shared 12v ocp @40A.

[Juular]

13 minutes ago, Spotty said:

everything you have said appears to be under the assumption that each connector on the PSU has its own dedicated 12V rail.

Damn, say no more. So that means that the only units that can simultaneously power very power hungry GPUs from single rail and have relatively low OCP limit (20-25A) to keep the advantage of having multi-rail in the first place with lower power draw GPUs are digital ones ... That's until PSU manufacturers start to use more sophisticated OCP hardware that will allow setting lower long-window current limit while keeping high-enough short-window limit to accommodate for transients.

[Spotty]

In my opinion Tier S and Tier A+ as they currently are should be removed.

There's way too much weighting put in to whether or not a PSU has multi rail, and it's ridiculous that the top two rankings are reserved for it. By reserving the top two tiers to multi rail units you're effectively saying that multi-rail is the sole most important determining factor in a power supply, more important than other factors such as performance, which simply isn't true.

 

From Aris' comments regarding this tier list:

Quote

Protections are important factors of a PSU so I agree there. Multi rail is not, especially since it can create problem if users don't pay attention on how they connect the cables and it can be substituted, in a degree at least, by OPP and UVP.

After protections, THE MOST IMPORTANT performance factors, with that order, are ripple suppression, transient response AND Load Regulation and then goes all the rest.

 

You could still have a separate tier above Tier A for the highest quality premium units like the Corsair AX1600i just to set them apart from the rest, but it should be based on the overall quality of the unit and not based just on whether or not it has multi rail. A lot of the PSUs that are good enough to reach that level are probably going to have selectable multi-rail feature anyway, but multi rail shouldn't be the deciding factor and if there's a single rail PSU that is absolutely fantastic in every other way then it should still qualify for that ranking.

[Vejnemojnen]

Let's assume, that I would buy an SFX PSU, which is roughly on par with the Bitfenix WhisperM in quality and performance. I was looking at these: Evga GM, Corsair SF, Enermax Revolution SFX, SeaSonic SGX

 

I don't really like SS units, but, 120months warranty seems fine to me. Though the Corsair has seven years, which is also good. Unfortunately, the EVGA units only have 1-2 years in-store warranties.

 

I'm kind of in favour of the Corsair SF, it's not expensive, looks good, seems quality. Any other decentish SFX units I should consider? : ) Preferably up to 100euros at most. Thanks in advance for any input, responses. : )

[LukeSavenije]

2 hours ago, Spotty said:

n my opinion Tier S and Tier A+ as they currently are should be removed.

There's way too much weighting put in to whether or not a PSU has multi rail, and it's ridiculous that the top two rankings are reserved for it.

i can limit it to one, start on a score based system that'll still include my 60a/rail max and separate the psus if higher from the lower wattage models, add a lot of other requirements to it. i've actually thought about this, made up scoring systems but never 100% figured it out. this would make it possible to even factor in noise with a tier a - b - c - d system for quality and + and - subtiers that factor in noise

 

if you have any suggestions to it, let me know

 

as an example here, this is a start i came up with at the time (forum handles graphs poorly, so sorry in advance for that)

	Total	best	worst
Good OCP	10	10
Late OCP	-10
Lacking OCP	-20		-20
Good OPP	20	20
Late OPP	-20
Lacking OPP	-30		-30
Good OTP	10	10
Late OTP	-10
Lacking OTP	-20		-20
Good UVP	20	20
Late UVP	-10
Lacking UVP	-20		-20
Good OVP	20	20
Late OVP	-10
Lacking OVP	-20		-20
Inrush Current Limiting	10	10
Meets C6/C7	10	10
Fails C6/C7	-10		-10
indipendent regulation	10	10
DC-DC	10	10
Group regulation	-20		-20
Double Forward	-10		-10
ACRF	0
Half/full bridge	10	10
<70a on 1 rail	10	10
<100a on 1 rail	-20
>100a on 1 rail	-30		-30
Shutdown issues	-10		-10
<50 mV Ripple (12v)	20	20
>50 mV Ripple (12v)	-10
Fails ATX Ripple (12v)	-30		-30
Fails overload ripple (12v)	-20
<30 mV Ripple (5v)	10	10
>30 mV Ripple (5v)	-10
Fails ATX Ripple (5v)	-30		-30
Fails overload ripple (5v)	-20
<30 mV Ripple (3.3v)	10	10
>30 mV Ripple (3.3v)	-10
Fails ATX Ripple (3.3v)	-30		-30
Fails overload ripple (3.3v)	-20
Output noise <30 dBA	10	10
Output noise <50 dBA	0
Output noise >50 dBA	-10		-10
Temperature rating <40c	-10		-10
Temperature rating <50c	0
Temperature rating 50c	10	10
		200	-300

 

[Mezoxin]

@LukeSavenije
 

why 60A per 12v rail  ? How did you come up with that value ?
 

There is no one that tests ovp or uvp during reviews not even Aris , for this reason 

 

Quote

• Over/Under Voltage Protection (OVP/UVP): These protections only kick in when voltages surpass or go below a specific level. Given that the corresponding ATX limits are practically dangerous for the system’s health, and the fact that there is no safe way to test these protections, we decided not to deal with them, for the moment at least.

 

[Juular]

2 hours ago, Spotty said:

In my opinion Tier S and Tier A+ as they currently are should be removed.

Respectfully i disagree, but you probably already seen my point with all that discussion above, keep multi-rail requirement but make a requirement that it should be configurable to less than 30A per rail, i.e digital. This will limit the number of units that qualify for it to mainly Corsair units but why not, if they have multi rail that simultaneously fool-proof and still offers advantage in protection if you configure it lower than stock value.

2 hours ago, Spotty said:

A lot of the PSUs that are good enough to reach that level are probably going to have selectable multi-rail feature anyway

Except Seasonic

2 hours ago, Spotty said:

but multi rail shouldn't be the deciding factor and if there's a single rail PSU that is absolutely fantastic in every other way then it should still qualify for that ranking.

If i shop for high-wattage PSU that would end-up in tier A\S quality wise anyway, i'd rather get multi-rail & digital one if price difference isn't very big. Regardless, just the fact that there are two tiers present that are higher that tier A doesn't mean that tier A units aren't recommended to buy, this is the whole point of this tier list, to show how all these hundreds of PSUs are different for the uninitiated, from the tiers description it should be clear that anything from tier A and up are really very good quality and you only get some perks like better build quality, and somewhat better performance but you shouldn't pay too much for that if tier A units are way cheaper.

[Spotty]

4 minutes ago, LukeSavenije said:

i can limit it to one, start on a score based system that'll still include my 60a/rail max and separate the psus if higher from the lower wattage models, add a lot of other requirements to it.

Not good enough!
Even by reserving one tier higher than the others for multi rail units you're still saying that Multi-Rail is the most important factor when it comes to a power supply and is more important than every other factor including overall performance.

 

6 minutes ago, LukeSavenije said:

Good OCP	10	10
Late OCP	-10
Lacking OCP	-20		-20

 

<70a on 1 rail	10	10
<100a on 1 rail	-20
>100a on 1 rail	-30		-30

 

<50 mV Ripple (12v)	20	20
>50 mV Ripple (12v)	-10
Fails ATX Ripple (12v)	-30		-30

You're kidding, right?
-20/-30 points just for a power supply being single rail?  It's unclear from the table, but does it also get another -20 points stacked for failing to have OCP on 12V since it uses OPP instead to serve the same purpose?

In your mind a power supply being single rail (up to -30 points or -50pts if you stack not having OCP on 12V) is just as bad or worse than a power supply failing to meet ATX ripple specifications (-30 points)?

 

Let's take a look at a quick hypothetical situation (and I won't go through all the aspects you listed, but for arguments sake let's assume anything not mentioned is the same)

• First PSU is 1200W single rail, performs well in all other aspects. Straight away -30 points for being single rail >100A. -20 points for not having OCP on 12V rail. +20 points for having 10mV ripple on 12V.
  • Total: -30pts
• Second PSU is a multi rail (+10 points), has OCP (+10 points), but fails to meet ATX ripple spec and shows 250mV ripple on 12V at 100% load (-30pts).
  • Total: -10pts

According to your ranking system in that situation despite the single rail PSU performing much better it is considered worse than the multi rail PSU that failed to meet ripple spec.

Heck, a PSU with "Shutdown issues" is only -10 points, because somehow having your PSU shut down unexpectedly due to a design fault is much better than your PSU being single rail?

 

Can't you see how flawed this is putting so much emphasis on one thing like multi rail/single rail?

Multi rail PSUs should not be ranked higher than equivalent or better performing single rail units just because they are multi rail. If you want to denote multi rail PSUs as special then just highlight multi rail units as a different font colour in the list and people can decide for themselves how much value they want to put in to it being multi rail.

[LukeSavenije]

34 minutes ago, Spotty said:

Even by reserving one tier higher than the others for multi rail units you're still saying that Multi-Rail is the most important factor when it comes to a power supply and is more important than every other factor including overall performance.

maybe it's a very important thing at 1200w because i don't want people to burn through their connectors?

 

give that a theoratical thought, ocp can't stop it, opp can't stop it because they're both ALWAYS too late in the event that uvp would fail

[Mezoxin]

  

1 hour ago, Spotty said:

In your mind a power supply being single rail (up to -30 points or -50pts if you stack not having OCP on 12V) is just as bad or worse than a power supply failing to meet ATX ripple specifications (-30 points)?

1 hour ago, LukeSavenije said:

maybe it's a very important thing at 1200w because i don't want people to burn through their connectors?

 

give that a theoratical thought, ocp can't stop it, opp can't stop it because they're both ALWAYS too late in the event that uvp would fail

 

when you hear a thunder , the first thing that goes through your mind is that it may rain or that you may get struck by lightning ?

out of spec ripple at normal operating conditions is far more dangerous than absence  multirail 12v OCP 

Multirail 12v OCP is good to have but the way it's currently implemented protects against a specific very low probability scenario of pinched wires or broken connectors that may form a resistive load , add to that you will only encounter it during the first time you start that psu that had faulty connectors or wirings or finishing a new build and you pinched one of the cables in the side panel for example  ( that's another conditional probability that will make the total probability of the scenario much lower )

 

[Juular]

I don't think that score based tiering is a good idea, it would need a lot of fine tuning and for the uninitiated it would be very hard to tell what all that means, maybe as a side means of ranking it would do, at least until it would be tuned to acceptable level but not as main means of tiering, it should be kept simple as it's now.

[Beerus]

2 hours ago, Juular said:

Respectfully i disagree, but you probably already seen my point with all that discussion above, keep multi-rail requirement but make a requirement that it should be configurable to less than 30A per rail, i.e digital. This will limit the number of units that qualify for it to mainly Corsair units but why not, if they have multi rail that simultaneously fool-proof and still offers advantage in protection if you configure it lower than stock value.

Except Seasonic

If i shop for high-wattage PSU that would end-up in tier A\S quality wise anyway, i'd rather get multi-rail & digital one if price difference isn't very big. Regardless, just the fact that there are two tiers present that are higher that tier A doesn't mean that tier A units aren't recommended to buy, this is the whole point of this tier list, to show how all these hundreds of PSUs are different for the uninitiated, from the tiers description it should be clear that anything from tier A and up are really very good quality and you only get some perks like better build quality, and somewhat better performance but you shouldn't pay too much for that if tier A units are way cheaper.

However the  S tier CM unit has inferior build quality that lesser tier ones. 

1 hour ago, LukeSavenije said:

maybe it's a very important thing at 1200w because i don't want people to burn through their connectors?

 

give that a theoratical thought, ocp can't stop it, opp can't stop it because they're both ALWAYS too late in the event that uvp would fail

1200W is overkill for most. Only a hanful or professionals running multi gpu and multi cpu setups need something that large. An pro overclockers potentially. Nowdays you dont need 1200watts even if you run a 32 core TR 3970x. Build quality>>>multi rail to my book. 

[LukeSavenije]

15 minutes ago, Beerus said:

the  S tier CM unit has inferior build quality that lesser tier ones. 

i wish you a lot of luck destroying a delta

[LukeSavenije]

22 minutes ago, Juular said:

it would need a lot of fine tuning

that's something i can do internally before i release it... it's not like i do a week on a full rewrite, i still sleep

 

but i'll ask everyone here the exact question i've asked aris, JG and many others: what are the downsides of a well implemented multirail?

because i can only name two:
cost

people stupid enough to not read the manual

Edited January 24, 2020 by LukeSavenije

[Juular]

9 hours ago, Beerus said:

1200W is overkill for most. Only a hanful or professionals running multi gpu and multi cpu setups need something that large. An pro overclockers potentially. Nowdays you dont need 1200watts even if you run a 32 core TR 3970x. Build quality>>>multi rail to my book. 

Why not get both multi-rail and the best build quality ? Tier S. Need only the best build quality disregarding multi-rail ? Well, some of tier A\A+ units qualify for this. Maybe Luke should've tiered multi-rail units to say tier A++ and units that otherwise would clarify for tier S if they were multi-rail in tier A+. I honestly don't see the point in tiering units just based on build quality, it's not a performance metric and most tier A units aren't very far behind in the first place.

9 hours ago, LukeSavenije said:

but i'll ask everyone here the exact question i've asked aris, JG and many others: what are the downsides of a well implemented multirail?

What you consider well implemented in the first place ?

[LukeSavenije]

15 minutes ago, Juular said:

What you consider well implemented in the first place ?

60a or lower/rail

no early shutdowns

 

so basically not too much, but also not too little current.

 

there are likely other parts that come to the exact implementation, but i can't name them at this moment