hx850i shows 965W with a 3080ti

[Juular]

3 minutes ago, jonnyGURU said:

What it looks like OP did was run iCUE in the background while doing a benchmark.  The spike occured, the graph resized.  Then he exited the benchmark, the spike was gone, but the graph was still resized to the measured spike.

I dunno, from the numerous screenshots OP provided any sensible person would assume that the scale is the scale, the top value on it is the maximum possible value on the graph, corresponding to the top of it. Actual graph goes just about to the 60% of that. Why keep the scale resized if the peak value is out of the window ? Why not render the peak value if the scale is actually resized to fit it, not just resized to keep the maximum observed value to about 60% of the graph scale for whatever reason as it seems to be the case from looking at the graph ? I digress, i don't have a monitoring-enabled Corsair PSU so i can't see for myself what's actually happening, and OP seems to be unable to take screenshots while running stress tests so we'll actually see what's the actual value of observed load during it and whether it actually matches the top value at the scale. A nice feature, for sure, but Corsair needs some competent UI/UX guys to make use of it.

[Guest]

13 minutes ago, Juular said:

I dunno, from the numerous screenshots OP provided any sensible person would assume that the scale is the scale, the top value on it is the maximum possible value on the graph, corresponding to the top of it. Actual graph goes just about to the 60% of that. Why keep the scale resized if the peak value is out of the window ? Why not render the peak value if the scale is actually resized to fit it, not just resized to keep the maximum observed value to about 60% of the graph scale for whatever reason as it seems to be the case from looking at the graph ? I digress, i don't have a monitoring-enabled Corsair PSU so i can't see for myself what's actually happening, and OP seems to be unable to take screenshots while running stress tests so we'll actually see what's the actual value of observed load during it and whether it actually matches the top value at the scale. A nice feature, for sure, but Corsair needs some competent UI/UX guys to make use of it.

Well.. I guess it's hard for you guys to understand if you've never used iCUE.

 

Maybe I'm not explaining myself well.  Because it works perfectly fine and none of you guys seem to have any idea what you're talking about. 

 

Obviously, no PC has a static load.  It's VERY dynamic. m The scale that the OP has shown moves.  It's not static.  It moves in real time.  The PHOTO is static.

 

If the user chooses a five minute time sample, the wattage scale is going to adjust for a five minute period. If the load stays at 200W max, the graph has a max of 200W and the line is anything between 0 and 200W.  Once that load exceeds 200W, the graph expands beyond that and the line rescales to fit within that range.  If there's a 1000W spike, then the scale resizes to that spike and the line rescales to the correct scale.

 

Listen to what you guys are asking for.... There's a 1000W spike.  Keep the scale at 500W?   The spike is only for a micro second.  Only keep the scale at the 1000W spike for one micro second?  Seriously... you guys need to use the software before you criticize it.

 

And as for the "make 1008W look like 1100W".  The iCue software is already bloated.  Have any of you done programming?  You're asking for EXTRA subroutines to be added that say "if W is great than 1000, but less than 1100W, then display 1100W."  Really?

[Jeppes]

42 minutes ago, jonnyGURU said:

Listen to what you guys are asking for.... There's a 1000W spike.  Keep the scale at 500W?   The spike is only for a micro second.  Only keep the scale at the 1000W spike for one micro second?  Seriously... you guys need to use the software before you criticize it.

 

And as for the "make 1008W look like 1100W".  The iCue software is already bloated.  Have any of you done programming?  You're asking for EXTRA subroutines to be added that say "if W is great than 1000, but less than 1100W, then display 1100W."  Really?

Maybe just show the area that the psu in question can handle. And the most common problem is that people purchasing are generally not electrical engineers. Should one worry about 1000W peaks that last for 1 micro second? How fast will opp trigger on a psu?

[akio123008]

2 hours ago, jonnyGURU said:

Because it works perfectly fine and none of you guys seem to have any idea what you're talking about. 

I understand, after you mentioned the 1008W is the peak value, and the graph also rescaled to that, it makes sense.

 

2 hours ago, jonnyGURU said:

Obviously, no PC has a static load.  It's VERY dynamic. m The scale that the OP has shown moves.  It's not static.  It moves in real time.  The PHOTO is static.

Yes I get that. That's why I asked: 

Quote

Is it too long ago?

 

But as a user opening this up for the first time, (or me seeing this for the first time on a forum post) not getting this graph isn't a surprise. Without the extra context you provided this graph is an absolute mess to read. 

 

For instance, it says "max 1008W" because that's the actual max, but then why does it also say "min 0W"? I'm pretty sure 0W isn't the actual measured minimum, unless it somehow measured the power draw when the computer was unplugged.

 

So when I see that, combined with the fact that the graph never actually reaches 1008W in the image, I start to suspect 1008 is just the max on the scale of the graph.

 

The fact that it combines axis label numbers, with statistics (peak value) is rather confusing.

 

 

2 hours ago, jonnyGURU said:

Listen to what you guys are asking for.... There's a 1000W spike.  Keep the scale at 500W?   The spike is only for a micro second.  Only keep the scale at the 1000W spike for one micro second?  Seriously... you guys need to use the software before you criticize it.

Rescaling a graph to the max value is not optimal. If that power supply uses 1008W, I'd much rather have it rescaled to 1100 or 1050. For 276W, rescale to 300. Something like that would be far better, combined with a separate spot for the peak value. I'm not saying I don't want the graph to rescale at all.

 

[Juular]

3 hours ago, jonnyGURU said:

If the user chooses a five minute time sample, the wattage scale is going to adjust for a five minute period. If the load stays at 200W max, the graph has a max of 200W and the line is anything between 0 and 200W.  Once that load exceeds 200W, the graph expands beyond that and the line rescales to fit within that range.  If there's a 1000W spike, then the scale resizes to that spike and the line rescales to the correct scale.

It's almost like we're talking about different things.

3 hours ago, jonnyGURU said:

And as for the "make 1008W look like 1100W".  The iCue software is already bloated.  Have any of you done programming?  You're asking for EXTRA subroutines to be added that say "if W is great than 1000, but less than 1100W, then display 1100W."  Really?

GraphScaleMax = MathCeil(GraphMaxValueInWindow() / 100) * 100;

[electropical]

2 hours ago, Jeppes said:

Maybe just show the area that the psu in question can handle.

That's the point : the HX850i can handle such transient spikes.
 

2 hours ago, Jeppes said:

And the most common problem is that people purchasing are generally not electrical engineers.

Why would anyone buy a PSU with monitoring if they don't understand jack about PSUs and power draw?
I am no electrical engineer, and I understood what was going on even before JG's explanations.

What happened here is that people rushed to conclusions.
I'm pretty sure you guys could have figured it out had you spent more time analyzing OP's screenshots.
 

2 hours ago, Jeppes said:

Should one worry about 1000W peaks that last for 1 micro second?

Yes, absolutely.
We've been talking about PSUs shutting down because of transient spikes quite a lot since the release of Ampere GPUs, actually.
 

3 hours ago, jonnyGURU said:

And as for the "make 1008W look like 1100W".  The iCue software is already bloated.  Have any of you done programming?  You're asking for EXTRA subroutines to be added that say "if W is great than 1000, but less than 1100W, then display 1100W."  Really?

That graph behavior is pragmatically perfect as far as I'm concerned, don't let them change it.
 

53 minutes ago, Juular said:

It's almost like we're talking about different things.

GraphScaleMax = MathCeil(GraphMaxValueInWindow() / 100) * 100;

I understand your logic, but we're missing the point of monitoring altogether : knowing precisely how much power was drawn with a precise figure.

The text isn't saying : "992w", it's saying "Max 992w".
Graphs have max values, but axes don't. It wouldn't make sense to right "Max 1100 watts" either.

By righting "Max 992W", we can deduce that such a value (=that isn't a multiple of 10) is an actual measurement.

And even if we had "1100W" at the top, it wouldn't help anyway.
Having 1100 divided by eight would still give us a scale that doesn't use multiples of 10.

I'm not a huge fan of Corsair in general, but I have to admit this implementation of monitoring is very pertinent.
It allows users to see what's going on in proper context without having to zoom in and out to get a complete picture.

The one thing we need to complain about right now is not being able to lower fan speed to lower RPMs, THAT is obnoxious.
(And that janky website, obviously )

[Juular]

5 minutes ago, electropical said:

Having 1100 divided by eight would still give us a scale that doesn't use multiples of 10.

Sure, don't do that, divide by multiples of 10 as in every other good UI. And let the user see what's was the value on arbitrary point of the graph, it looks like iCUE doesn't have that either. It's not about Corsair in general, it's about their software \ UI team. iCUE is miles better than anything MSI \ Gigabyte \ ASUS \ NZXT came with but it still has huge and obvious issues, and it ain't getting better as evident by the lastest 'redesign' of it which only made it worse in the areas i actually use (no idea if they fixed the graphs).

[akio123008]

7 minutes ago, electropical said:

Graphs have max values, but axes don't. It wouldn't make sense to right "Max 1100 watts" either.

And even if we had "1100W" at the top, it wouldn't help anyway.
Having 1100 divided by eight would still give us a scale that doesn't use multiples of 10.

You're completely missing my point.

 

The reason we set the graph to 1100 rather than 1008, is not to give multiples of 10 or whatever. The reason to use 1100 is to make clear it is not a measured value and distinguish it from the actual peak value. Using a round number like this makes it more obvious that we're looking at a scale number, not at a measured value 

 

(which I know wouldn't work if the peak happened to be exactly 1100W, but that's unlikely to happen) 

 

Because my main problem with this is how the axis label serves the double purpose of both displaying the peak value and being the max value on the axis.

THAT is the confusing part.

 

The graph should be somewhat like this, with the measured data (peak, avg, etc) in a separate place:

 

The same way you'd see it on the display of an oscilloscope.

 

 

 

 

[akio123008]

56 minutes ago, Juular said:

It's almost like we're talking about different things.

I think the 200W was just an example.

In this case the 992W would have been hit at some earlier point in time, setting the max to 992.

 

[Juular]

9 minutes ago, akio123008 said:

I think the 200W was just an example.

In this case the 992W would have been hit at some earlier point in time, setting the max to 992. Then in the image, the max is about 558. Because the time duration of the graph is 1 minute, but the "rescale frequency" is 5 minutes apparently, so 992W was hit say 3 minutes ago and not visible in the image.

That's the only other explanation other than they deliberately set the scale to 2x of the maximum observed value. They set the scale to the maximum measurement window possible, regardless of what's actually displayed on the graph, but combined with the fact that they don't actually show the user what was the value on the arbitrary point of the graph by mouse-over and the fact that the scale is for some reason divided by 8 (how the hell they even came up with this value), not 10 - that's just lazy programming and bad UI/UX. Which leads to us, trying to guess for the whole 2 pages whether the top value on the scale is actually a fucking scale or the maximum value on the graph.

[akio123008]

Just now, Juular said:

That's the only other explanation other than they deliberately set the scale to 2x of the maximum observed value. They set the scale to the maximum measurement window possible, regardless of what's actually displayed on the graph, but combined with the fact that they don't actually show the user what was the value on the arbitrary point of the graph by mouse-over and the fact that the scale is for some reason divided by 8 (how the hell they even came up with this value), not 10 - that's just lazy programming and bad UI/UX

I could say once again I find this graph confusing but I think I'm starting to repeat myself haha. 

[Jeppes]

37 minutes ago, electropical said:

That's the point : the HX850i can handle such transient spikes.
 

Why would anyone buy a PSU with monitoring if they don't understand jack about PSUs and power draw?
I am no electrical engineer, and I understood what was going on even before JG's explanations.
 

So how fast is the opp? Will it trigger a shut down on micro second spikes? Because there are tests with for example Corsair rm650x and it runs 3090 just fine. Those should have opp well under the spikes to 1000W:s. That confidense is most likely just Dunning Kruger effect.

[RejZoR]

Where's the issue? PSU's graph scales to peak power draw exactly. It doesn't have any reserved extra space on the graph above peak/max usage. Which in a way makes sense. If line is all the way at the top, that's the max it was. It could scale with 20% graph space above any peak power draw, but why complicate things when this works just fine and also serves as easy way to read what the max was.

 

Seeing 100W or even 150W spikes above what PSU is rated at is nothing unusual. Most PSU's can actually run way above rated power even for extended periods of time. It's not recommended, but most can. A 850W PSU can usually run at 1000W without much issues even for 30 minutes or something. It's just not recommended to do so regularly or constantly because it's not designed for such loads. If such loads only happen for a second, it's a non issue.

[electropical]

@Juular @akio123008 There's so much stuff to unpack in what you guys said, Corsair isn't paying me to defend them, and english isn't my native language.

To put it simply, my guesses are that :
- they're using the max measured value as top boundary to allow for maximum graph resolution (= zooming in as much as possible) while keeping things in context
- they won't use a static scale because they need to zoom in as much as possible in all configurations
- the point of the graph is not to give the user a tool to precisely read power draw at any given point in time, there's the logging feature for that
- the "Max W" value at the top is there to reduce clutter, as akio said it gives both the scale and the max power draw measured
- they're dividing by 8 and not by 10 to reduce clutter
- now that you guys know how that graph works, I'm pretty sure you would be able to use it without any sort of confusion

@Jeppes There's no way the RM650x can handle the transient spikes of a 3090 with a power hungry CPU.
And it isn't OPP that turns off the PSU, it's OCP.
When there isn't enough capacitance on the secondary side, the transient spikes draw too much juice, that's what's making the protection kick in.

[Juular]

Just now, electropical said:

To put it simply, my guesses are that :
- they're using the max measured value as top boundary to allow for maximum graph resolution (= zooming in as much as possible) while keeping things in context
- they won't use a static scale because they need to zoom in as much as possible in all configurations
- the point of the graph is not to give the user a tool to precisely read power draw at any given point in time, there's the logging feature for that
- the "Max W" value at the top is there to reduce clutter, as akio it gives both the scale and the max power draw measured
- they're dividing by 8 and not by 10 to reduce clutter
- now that you guys know how that graph works, I'm pretty sure you would be able to use it without any sort of confusion

Which is all, just a lazy, bad UI design, not an attempt to 'reduce clutter'. 10 dividers instead of 8 barely add any clutter, if anything resize the graph area to be a little bit larger vertically. The top value on the scale is supposed to signify the maximum value possible to be shown on this graph, that's what's the scale there is literally for. If you're going to display the maximum value on the graph, do it separately, not on the scale, otherwise it's confusing. And i would be having better time reading the graph if it had actual scale value at the scale instead of maximum graph value and being divided by 10 so i can easily interpolate graph's values (since there's no mouse-over). I don't want to guess what that top value on the scale means and to exercise in dividing numbers by 8. Now, we're all really just flaming here, as much as i'd want for Corsair software team to do a better job, that's probably not going to happen soon, definitely not because of all we've wrote here so i'll just recede.

[Jeppes]

52 minutes ago, electropical said:

There's no way the RM650x can handle the transient spikes of a 3090 with a power hungry CPU.
And it isn't OPP that turns off the PSU, it's OCP.
When there isn't enough capacitance on the secondary side, the transient spikes draw too much juice, that's what's making the protection kick in.

Yes way. And can you guess the next question? Will the ocp kick on micro second spikes if 1ms average stays well below triggering point? Badly worded but you can measure from multiple points. 1000W for 1μs in some wire is not necessarily 1000W for 1 μs when measured elsewhere behind capacitors.

[akio123008]

57 minutes ago, electropical said:

now that you guys know how that graph works, I'm pretty sure you would be able to use it without any sort of confusion

That's fair.

 

What started this discussion for me was someone mentioning the OP might have problems reading graphs, which I thought was a bit unfair given the slightly confusing nature of this graph. Basically what I'm saying is, I see how it works now, but I'm not surprised a first time user of the software would get it wrong.

[Mason Peterson]

2 hours ago, Jeppes said:

Yes way. And can you guess the next question? Will the ocp kick on micro second spikes if 1ms average stays well below triggering point? Badly worded but you can measure from multiple points. 1000W for 1μs in some wire is not necessarily 1000W for 1 μs when measured elsewhere behind capacitors.

If there is not a high enough spike in current to reach back to the resonant/SLS controller measuring across shunt resistors or the I(12V) pin on the supervisor IC, then typically, it's not going to turn off from OCP.

 

However, that's a single spike and not indicative of the real world (the time varies as well). As you spike a lot, the regulation primary side responsible for feeding the power through doesn't see anything and keeps whatever its sending constant while there is massive voltage drops and secondary side capacitor discharging to compensate for it.

 

If the spikes drain the capacitors enough, then voltage on the output drops. This signal is still sent to the supervisor IC or compared to another wire trace and might cause UVP to shut down the unit. This is seen with Seasonic Primes on Vega sometimes (on top of the problem I'll allude to) and with Corsair SF750 on 3090s/spiky 3080s. 

 

But capacitors aren't just some battery source that magically covers everything. The frequency in which they're seeing matters too. Sorta of like how past a certain frequency, the human body feels excessive warming rather than an electric shock, capacitors are only going to work at their specific capacitance thresholds within a certain range.

 

You could spike or send too much noise induced from GPU back into the system, and the capacitors all of a sudden have to provide 1 Farads of capacitance because the RLC circuit wasn't designed that robust (why some people say Seasonic doesn't have the greatest engineers). That would show excessive currents or even trip SCP because a controller thrown out of sync affects the switching winding. If things start turning on when they shouldn't be, you're going to phase fault.

 

[Guest]

15 hours ago, akio123008 said:

You're completely missing my point.

 

The reason we set the graph to 1100 rather than 1008, is not to give multiples of 10 or whatever. The reason to use 1100 is to make clear it is not a measured value and distinguish it from the actual peak value. Using a round number like this makes it more obvious that we're looking at a scale number, not at a measured value 

But it IS a measured value.  The PSU DID report a spike that high.

 

15 hours ago, Jeppes said:

So how fast is the opp? Will it trigger a shut down on micro second spikes? Because there are tests with for example Corsair rm650x and it runs 3090 just fine. Those should have opp well under the spikes to 1000W:s. That confidense is most likely just Dunning Kruger effect.

OPP on the analog ICs is in milliseconds, not microseconds.  Add to that that there's enough capacitance in the secondary capacitors to withstand a few microseconds of a spike, you can realize that seeing a 1000W+ spike on a 850W PSU is a non-issue.

[IIIIIIIIII]

1 hour ago, jonnyGURU said:

OPP on the analog ICs is in milliseconds, not microseconds.  Add to that that there's enough capacitance in the secondary capacitors to withstand a few microseconds of a spike, you can realize that seeing a 1000W+ spike on a 850W PSU is a non-issue.

In other words, the claim that the OCP is working and the PSU is shutting down because of a terrible spike on the Ampere GPU is 99% a scam?

 

At one manufacturer claims "From the products produced thereafter, the time constant of the capacitor (probably MLCC) in the OPP integration circuit was increased to adjust the timing at which the power supply responds to instantaneous peak power consumption."...

[Guest]

41 minutes ago, IIIIIIIIII said:

In other words, the claim that the OCP is working and the PSU is shutting down because of a terrible spike on the Ampere GPU is 99% a scam?

 

At one manufacturer claims "From the products produced thereafter, the time constant of the capacitor (probably MLCC) in the OPP integration circuit was increased to adjust the timing at which the power supply responds to instantaneous peak power consumption."...

Not sure I understand what you're saying.  Context?

 

 

[IIIIIIIIII]

42 minutes ago, jonnyGURU said:

Not sure I understand what you're saying.  Context?

Sorry. English is not my first language

 

In units produced after 2018, they are claimed that the sensitivity of the OPP was adjusted by adjusting the capacitor in the OPP circuit, so there is no problem with the Ampere GPU.

 

Edit : 

They also claim that the reason for the shutdown with Ampere and some older AMD GPUs was that the OPP was too sensitive.

[Guest]

47 minutes ago, IIIIIIIIII said:

Sorry. English is not my first language

 

In units produced after 2018, they are claimed that the sensitivity of the OPP was adjusted by adjusting the capacitor in the OPP circuit, so there is no problem with the Ampere GPU.

 

Edit : 

They also claim that the reason for the shutdown with Ampere and some older AMD GPUs was that the OPP was too sensitive.

Who is "they"?

 

[IIIIIIIIII]

12 minutes ago, jonnyGURU said:

Who is "they"?

Seasonic distributor.

[Guest]

13 minutes ago, IIIIIIIIII said:

Seasonic distributor.

Ok.. So.. The current "tripping" with Seasonic PSUs is not OCP.  It's noise on the +12V sense causing the supervisor IC to freak out and shut down.

 

Thought this was discussed already a few times.