PSU tier list 2.0

[turkey3_scratch]

Yes the 750 G2 is good.

 

[Darkbreeze]

The XFX TS unit is Haswell/Intel low power state compatible, the S12II 520 and 620 are not. They can still be used if you turn off the C6/C7 states in the bios but the XFX unit is probably a better choice since you won't need to do that.

Sorry, didn't understand you well can you explain it in another way?I'm using a skylake cpu

I really don't think I can make it any more simple than that. I can make it a lot more complex, but that's about as simple as it gets. The XFX unit supports the C6/C7 CPU low power states used by Haswell and newer Intel CPUs. The S12II 520w and 620w do not. Your CPU is newer than Haswell, and uses the C6/C7 states, so unless you want to go into the BIOS and turn off the C6/C7 states under power management, which you CAN do, then I'd go with the XFX unit so you won't have to or need to do that.

 

[Darkbreeze]

Group regulated power supplies like the Seasonic S12ii cannot maintain stable voltages when crossloading. The sleep states make the CPU use virtually no energy, so you're looking at about 1W on that 12V rail (which powers things like the GPU and GPU) being utilized, while the 3.3V is still working because of RAM, and the 5V rail... not sure.

But anyway it's this really low 12V load that causes the 12V to go whopping high or the 3.3V and 5V rails to go whopping low. Units that do not have a group regulated design don't have this problem, and are compatible with those super low power sleep states Haswell and Skylake CPUs have.

He wanted it simplified, not made exponentially more complicated.

 

[turkey3_scratch]

The 850 P2 tests well for PWR_OK, while the Superflower Leadex Platinum 550, while not tested for PWR_OK, has low hold-up time and can be assumed to fail PWR_OK testing if it was done. So the PWR_OK problem might very well only be a problem with lower-wattage G2 and P2 models. For sure the 550 G2. Maybe the 650 G2 and 650 P2. But I have a feeling the higher wattage models are safe, if anybody has those. Again, all speculation, but we've been discussing this on Jonnyguru and this is sort of where we're maybe heading. One guy on there thinks it may only be an all-around G2 problem, but I think it's just a low-wattage 550 G2 problem, and maybe the 650 G2 and 650 P2.

 

[Darkbreeze]

I've requested Aris to retest this on some other samples if and when he has them or has the opportunity, so it can be determined if it was just only a sample of one, or is recurrent. Don't know if he will do it or not, but I asked anyhow. Have you seen examples of this in other tests not done by Aris on the G2 models? If so, can you link those reviews?

 

[turkey3_scratch]

As far as I know Aris is the only one who has the equipment to test it since supposedly it costs a lot of money.

 

[Someone Somewhere]

I think he means other reviewers, like JG.

 

[turkey3_scratch]

I think he means other reviewers, like JG.

Exactly, the other reviewers are incapable of performing these tests.

Technically, though, it's the Superflower Leadex Gold 550 that was tested for this flaw. So, perhaps the G2s are fine and EVGA requested for this to be changed on their own units? Possible, but is it likely?

Hmm. It was primarily the people on JG who talked about this, but then again they are kind of anti-EVGA, so...

 

[turkey3_scratch]

Now wait a second, now I'm a little confused. In the Tomshardware review of the 550 G2 here, before the AC_LOSS to PWR_OK testing was around, Aris tested hold-up time, which is defined as the time of AC_LOSS to when voltage goes out of spec.

If the yellow line is voltage, I don't quite understand why we don't see that voltage drop on the graph? Here is the Superflower Leadex Gold 550:

In this graph, green is the PWR_OK signal, and yellow is still voltage, but this time you actually get to see voltage. I'm guessing that the old testing equipment for some reason would not show the yellow line voltage go down? Hope I worded that right.

Here is a review of the 750 P2 http://www.hardwareinsights.com/wp/super-flower-leadex-platinum-750-w-review/6/ which does not have this flaw, so I don't think any of the P2 series has this issue, because Aris tested the 850 P2 and that was also good.

For all we know, the G2 might not even have it, if Superflower really changed it. But seeing as the hold-up time tested for the 550 G2 was 11ms, I wouldn't be surprised if it had the same behavior as the Leadex Gold 550 (I sure hope not, though, I own that PSU!).

At the present moment, what we do know:

1) Superflower Leadex Gold 550 drops the PWR_OK signal at 10.8V
2) EVGA 550 G2 most likely has this behavior, but there is no proof
3) EVGA T2 series is safe
4) EVGA P2 units that have been tested so far are perfectly safe, and likely the whole series is
5) The rest of the G2 lineup, we don't know. For all we know the 550 G2 could be safe, but it's doubtful.

I did get a little carried away, I apologize. The guys on JG acted as if they were certain all G2 units have this problem, and it got to me. But as of now only the 550 G2 is to suspect. I should be more cautious jumping forums, on JG they tend to have a thing against EVGA, so I need to be aware of that bias. Now that I'm looking into it a lot deeper, I personally think the rest of the G2 lineup is safe.

If Aris is to test any unit for this behavior again, going back to the 550 G2 and testing that would be great.

 

[Someone Somewhere]

I'm pretty sure that on the first one, the yellow line is PWR_OK. They were testing how long it took POWER_OK to drop, not how long it took the voltage to drop.

 

[turkey3_scratch]

Uh oh, I take all that back. Now this is very concerning. On Aris' review of the 650 G2 on Techpowerup: https://www.techpowerup.com/reviews/EVGA/SuperNOVA_G2_650/5.html

Hold-up time easily surpassed the 16 ms mark the ATX specification sets. Please note that although the unit uses the exact same bulk cap as the 550 G2 that failed this test, it somehow managed to produce a very long hold-up time. Given the 550 G2 has 100 W less capacity, it should register an even lower hold-up time than its small brother. The assumption is that significant improvements that somehow affect the hold-up time were made, or that the bulk cap of our 550 G2 sample wasn't at its best.

Somehow the 650 G2 got 21ms hold-up time. Guess how you think Superflower did that? Yep, setting that PWR_OK value to a really low one. There's no way the 650 G2 could possibly remain on longer during AC loss than the 550 G2 when it uses the exact same bulk capacitors!

Then there is Aris' review of the 750 G2: https://www.techpowerup.com/reviews/EVGA/SuperNOVA_G2_750/5.html

Although the G2-750 uses the exact same platform and components as the Leadex Gold 750 unit, performance in this test was very different to the Leadex unit. The Leadex Gold 750 did not reach the minimum allowed time, but the GS-750 easily passed 16 ms and even reached 22 ms! We actually made a point of desoldering both bulk caps inside the G2-750 to shed some light on this mystery, and capacity measurements confirmed that their combined capacity is in line with the official specification. We could not do the same to the Leadex unit since it wasn't at our disposal anymore and can only speculate that its bulk caps were not in their best state, although something else might have severally affected the unit's hold-up time. This is the first time we have come across such a phenomenon, but are pretty sure that it is due to the Leadex's bulk caps.

Seems awfully suspicious to me. They can't just increase that timer without setting the PWR_OK value to a lower one. Unfortunately, it seems to me like the 750 G2 also most likely has this problem. The 850 G2 might be alright: https://www.techpowerup.com/reviews/EVGA/SuperNOVA_G2_850/6.html

Ah, it just struck me! In the old tests, the yellow line is the PWR_OK signal, not the voltages! But in the new tests the yellow is the voltage line, and the green is the PWR_OK signal. Which actually would mean that in the 550 G2 is probably the safe one. Oh, I have just been enlightened. Hear me out. It's such a confusing matter that I must use a spoiler.

Spoiler

Do you know how I'm very nit-picky when it comes to terminology? Well, this is a case of that! It's because before the new testing equipment hold-up time tests didn't test hold-up time; they tested AC_LOSS to PWR_OK time, and were wrongly called hold-up time tests by reviewers. This is why I have been so confused. So in the 550 G2, even though it says hold-up time is 11ms, that is wrong/ AC_LOSS to PWR_OK is 11ms, not hold-up time. The new testing equipment doesn't test AC_LOSS to PWR_OK; it tests hold-up time! It's all backwards, this is why it's so confusing.

With that knowledge, the 550 G2 is actually likely safe. The 650 G2 is probably unsafe and so is the 750G2. The 850 G2, not sure. See if, I wasn't a man who questions things, I never would have come to this conclusion.

 

[turkey3_scratch]

I'm pretty sure that on the first one, the yellow line is PWR_OK. They were testing how long it took POWER_OK to drop, not how long it took the voltage to drop.

Yep, I just realized that. How come these tests were falsely called hold-up time tests? Isn't it confusing to call tests hold-up time tests when they're not testing hold-up time but are testing AC_LOSS to PWR_OK time?

Gaaah! It's so late and I'm going nuts!

 

[Darkbreeze]

I don't know about saying "safe" and "unsafe". Chances are good that most users will never encounter the circumstances that MIGHT call these scenarios into question anyhow. And if they did, it still doesn't mean that it WOULD cause an issue, only that it could.

 

[Someone Somewhere]

Because for any compliant PSU, hold up time is greater than the time from AC power loss to the PWR_OK signal dropping.

 

[CTurbo]

Exactly, the other reviewers are incapable of performing these tests.

Turkey, this is the exact reason why you shouldn't pee your pants and form ultimate conclusive opinion facts over an entire psu line from just one review.

I still think you put WAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAY too much stock in a psu review.

 

[dimitili]

Now i got confused by all this....Is finally the G2 750 evga good?....should i buy? or should i choose another psu?

 

[turkey3_scratch]

Because for any compliant PSU, hold up time is the time from AC power loss to the PWR_OK signal dropping.

That can't be, because in the latest Aris reviews holdup time and AC_LOSS to PWR_OK are two different things. Holdup time in this case is AC_LOSS to DC_LOSS (out of spec voltage):

Also, to quote what Orion told me:

Wording this is important - if the voltages are out of spec and the PSU still didn't shut down, then this period is not included in hold-up time at all.
Hold-up time is strictly defined as time in which the voltages are kept in spec with no AC power - so if your PSU maintains voltages in spec for 17ms, then they drop out of spec for 3ms, and then PWR_OK shuts it down late, you can't say "hold-up time is 20ms, but 3ms of that are ouf of spec". You can say "hold up time is 17ms, but the PSU keeps working with out-of-spec voltages for 3 more milliseconds".
So, I would rephrase it as "hold-up time will appear to be longer if detected by measuring PWR_OK only, but will turn out to be shorter with the remaining time being out of spec".

Also, on Jonnyguru I seem to have concurrence. I said:

Actually, we had it all wrong. The EVGA 550 G2 is most likely a safe power supply.

Here is the Tomshardware review for the 550 G2: http://www.tomshardware.com/reviews/...ly,4244-4.html

It says 11ms is the hold-up time, but that is not true! The old testing equipment did not test hold-up time, it instead tested AC_LOSS to PWR_OK, even though they were falsely called hold-up time tests. So the G2 has a much lower AC_LOSS to PWR_OK value than the Leadex Gold 550, which means the 550 G2 should be 100% safe.

Aris' new testing equipment doesn't add in AC_LOSS to PWR_OK testing; it simply renames the old test, which was called "hold-up time tests", to "AC_LOSS to PWR_OK". Which means the new testing equipment is used to test real hold-up time.

The 650 G2 manages to have twice as high of an AC_LOSS to PWR_OK time than the 550 G2 and it uses the same bulk capacitor. It is the 650 G2 that is the problem, not the 550 G2. The 750 G2 also has too high of AC_LOSS to PWR_OK time, which means that that unit is also a problem.

The 850 G2 is at 17.1ms, so it may or may not be a problem.

And sith'ari seems to agree with me completely:

Great logical assumptions all of them!!
Unfortunately, we are talking about a PSU, a critical component, whose proper function is essential for the longevity & stability for the rest of the hardware!
(*and the great failure of SF Leadex Gold550 has been a real shock to me, that had impact for the entire Leadex platform )
So now i need reviews (data) to verify logic.
P.S.:Regardless to what i said, quite possibly, all the assumptions that you made are correct! Great work!

 

[turkey3_scratch]

Now i got confused by all this....Is finally the G2 750 evga good?....should i buy? or should i choose another psu?

Right now it's all speculation, but I am well confident so far that:

1) The 550 G2 does not have this problem
2) The 650 and 750 G2 do have this problem

@Cturbo: I did just come up with a conclusion from PSU reviews! The 650 G2 drops the PWR_OK at an unsafe voltage. I am 99.9% sure of this. It's impossible for the unit to honestly have double the AC_LOSS to PWR_OK time without setting the voltage value lower when it has the same bulk cap as the 550 G2. If those reviews were not available, I wouldn't have solved this.

 

[Someone Somewhere]

Note that I said 'compliant'. Therefore, those PSUs are not compliant, and my point stands.

The full requirement of the Holy PSU Bible (pg 23) is that PWR_OK shall drop at least 1ms before any rails drop below 95% of their nominal value.

The definition of "AC loss to PWR_OK hold-up time" (which is what must be >16ms) is exactly that. Time from AC loss to DC loss is not measured, but by simple addition of time from AC loss to PWR_OK inactive, and PWR_OK inactive to DC loss, we can conclude that it must be at least 17ms.

There's a lot of other interesting stuff in there I didn't know, like this:

NOTES:
1. At +12V1DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to
±10%.
2. At +12V2DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to
±10%.
3. Voltage tolerance is required at main connector and SATA connector (if used).

Not sure exactly what they mean by that - whether it's just under crossload, or at any point. If so, many PSUs will have been failed for regulation when they were within spec.

Also, 3V3 sensing is required, not optional:

Remote Sensing - REQUIRED
The +3.3 VDC output should have provisions for remote sensing to compensate for
excessive cable drops. The default sense should be connected to pin 13 of the main
power connector (Figure 7). The power supply should draw no more than 10 mA
through the remote sense line to keep DC offset voltages to a minimum.

 

[turkey3_scratch]

I'm guessing peak loading would be something like a GPU that, when running Furmark, gets that huge brief power spike that goes far beyond the PSUs continuous power rating, meaning the 12V rail could go from 11.0V to 13.0V in that moment? Sounds almost like a transient response. Also, what do you mean the time from AC_LOSS to DC_LOSS is not measured? In the past it wasn't, but now it is. Look at that vertical line that cuts through the yellow line once it reaches 11.4V.

But the thing is, relying on addition to calculate AC_LOSS to DC loss is no good when companies drop the PWR_OK signal after there is DC loss. Because then it makes it look like DC loss is much later when it is, when in reality, PWR_OK to DC loss is a negative value since DC loss came first.

 

[Someone Somewhere]

Yeah, transient response seems likely.

I'm saying that the ATX v2.31 spec I linked to does not name the time between AC loss and DC power loss/out-of-spec, nor does it provide any requirements for its value.

Go read the spec, look at the pretty pictures, and read my post again.

 

[turkey3_scratch]

Yes it does:

PWR_OK to DC_LOSS delay is supposed to be >1ms, so when that number is negative it's out of spec.

 

[Someone Somewhere]

That is the time from PWR_OK dropping to DC loss. As I said, hold up time is defined as the time from AC loss to PWR_OK dropping. They are different.

I have never argued that having voltages go out of spec before PWR_OK drops is compliant. I argued that, assuming an otherwise compliant PSU, measuring hold-up time how it is measured in the spec is perfectly acceptable.

Measuring the difference between PWR_OK dropping and voltages going out of spec should actually be easier than measuring the full hold-up time, though - less need to measure AC, or accurately disconnect incoming power.

It's impossible for the unit to honestly have double the AC_LOSS to PWR_OK time without setting the voltage value lower when it has the same bulk cap as the 550 G2. If those reviews were not available, I wouldn't have solved this.

The controller on the smaller unit could (I am not saying it was, merely that it could have been and would lead to this) have been configured to not fully discharge the bulk cap. Don't know why you'd do it, but it's certainly possible.

 

[turkey3_scratch]

So then you are saying what Orion told me is incorrect, and on Aris' new reviews his definition of holdup time is incorrect? I have different people telling me different things.

Also, if measuring the difference between PWR_OK to DC loss is easier than measuring full holdup time, why is it that Aris needed new testing equipment to test PWR_OK to DC loss while his old equipment tested AC_LOSS to PWR_OK, or as you call it holdup time?

 

[Someone Somewhere]

I am saying that according to the intel docs, that is their definition of hold up time for ATX PSUs. Not necessarily the definition used for generic switchmode supplies, which don't usually have a 'PWR_OK' signal.

For any reasonably designed PSU, PWR_OK to DC loss shouldn't be something you have to worry about - it's not going to get cost-downed, and it's usually just the marketers that lie, not the engineers involved in designing the actual product. Slapping a new label on the PSU won't change the voltage at which PWR_OK drops, and nor will telling the factory to install a cheaper smaller cap.

Using a 2-channel scope is enough to measure two of:

• ■The point at which the AC voltage is cut off,
  ■The point at which PWR_OK drops, and
  ■One rail of the DC output.

If you want to measure all three simultaneously, you will need a scope with more channels. But you can measure any two with a two-channel scope.