Is the maker's rated PSU wattage the actual limit?

[BonzaiDuck]

I'll say that I've only occasionally posted in this forum. I just read through a sticky thread to prepare for this, so nobody would say I was somehow "amiss," deficient -- whatever.

I only learned enough about PSU's over the years to fill my need to buy good ones.

These last couple weeks, I've been playing with a 2x SLI GTX 970 configuration inserted in a computer build as an afterthought. The PSU was a relatively new 650W semi-modular Seasonic. The "estimations" I made before I deployed the second card seemed to be close to "measurements" I've taken after the deployment.

So I became concerned about how close I was coming to the PSU's limits. I've now determined that running Kombustor tweaked to load both GPUs with a mild overclock (+100 core, +200 memory) pushes the PSU power output to ~ 530W while loading the CPU to 42% of its maximum overclocked "package" wattage. To be succinct, that maximum CPU wattage for the OC was 140W, so the hypothetical scenario would add 80W to the 530W total, or ~610W.

I started looking at 750W and 850W models, knowing that I still had a choice to leave the GPU clocks at stock (subtract about 25W from the 610). It just looked a bit "close."

I stumbled across an XFX "Black Pro" 750W unit confirmed to be a rebadged Seasonic, similar to a Seasonic X "Gold" model. The review at Hardware Secrets shows the model was released around 2012. They give it stellar ratings after thorough testing, noting that the test unit was actually under-rated as "Gold," and passed "Platinum" criteria:

http://www.hardwaresecrets.com/arti...tion-Full-Modular-Power-Supply-Review/1695/11

You can browse back and forth through the 13 pages, but my interest was drawn to "Overload Tests." In the overload test, they pushed the power draw on the PSU to ~126%. The PSU was able to sustain wattages up to ~ 950W with 84% efficiency before the protection features kicked in and shut it down.

Without lab tests, we're probably in the dark about anything else, but certain conclusions or gambles on similar PSUs might -- (or might not!) -- prove reliable.

It would seem likely that the more respected manufacturers would be somewhat cautious in their ratings anyway, especially with 5, 7 -- even 10 year warranty (EVGA is promoting a 1000W unit with the 10-year -- rebadged Super Flower). Therefore we might feel confident -- optimistic -- that some slight overage in the power draw could be sustainable.

This might also address the issue someone recently raised about spikes or variations of power draw around the mean.

Thoughts about this?

 

[VirtualLarry]

Don't forget that Maxwell GPUs' peak power consumption is actually higher than what you would see as an averaged power consumption read by a Kill-A-Watt or UPS monitoring.

It's just that the peaks are so short-lived, they don't show up except under an O-scope. TomsHardware did just that though. I don't have the link handy.

 

[Deders]

Most decent PSU's will handle above their rated wattage, just most likely not within the rated efficiency band.

 

[BonzaiDuck]

Don't forget that Maxwell GPUs' peak power consumption is actually higher than what you would see as an averaged power consumption read by a Kill-A-Watt or UPS monitoring.

It's just that the peaks are so short-lived, they don't show up except under an O-scope. TomsHardware did just that though. I don't have the link handy.

I think the link or the graphs were also provided in a recent thread here.

I guess my original point is that people tend to look at the spec as an absolute limit, but as Deders noted in his post following yours, that's not the actual case.

I think I saw in the graphs which (I hope) we're both referencing, that the spikes could exceed the mean value by as much as 100W. If a good PSU is capable of handling wattage draws above its rated spec, it would diminish somewhat the importance of the variation.

Deders is correct that the watt overage provided by such a PSU would be below the rated efficiency, and the HW Secrets review of the XFX (Seasonic) measured that -- for the particular PSU.

I didn't get my Maxwells together -- likely it wasn't a matter of "need" for "performance" -- but it was a matter of my curiosity and a desire to tinker. So I started from the "wrong end." That is to say, if the original plan included 2x SLI, or if this were really a "project with a plan," I would've picked a PSU of higher wattage. I just happened to have this unit handy.

I've overclocked the 2 cards -- both core and memory -- so they're reaching ~ 105% of the allowable 110% wattage. I STILL keep coming up with a worst case scenario of maximum CPU OC wattage (140W) and what I measure with the cards, to project about 613W for that worst case.

As it stands, the more likely situation has the maximum wattage running Kombustor at closer to 500+W to include the CPU package value of ~40W. Either way, I'm thinking that a good 650W PSU is "good to go" with these cards. If the PSU is even "almost as good" as the XFX 750 in the review, the unmeasured spikes would at most be maybe 50W above the spec.

I'm worrying about the 650W PSU less and less the more I play with it and ponder. And truth is -- referring again to Deders' remark -- if the PSU's efficiency is 90% under moderate load and 87% nearest the rated spec, then my estimates are STILL excessive by a factor of 3%.

Yes . . . I like Maxwell . . .

 

[Deders]

Kombuster (the tessellation tests in particular) will load your GPU beyond any game and having something like prime 95 in the background will again load your CPU beyond any game.

If you want a realistic extreme scenario then Crysis 3's Welcome to the Jungle level with all the grass will load all your CPU cores and your GPU's. I've not seen any part of any game that draws more power than this.

 

[BonzaiDuck]

Kombuster (the tessellation tests in particular) will load your GPU beyond any game and having something like prime 95 in the background will again load your CPU beyond any game.

If you want a realistic extreme scenario then Crysis 3's Welcome to the Jungle level with all the grass will load all your CPU cores and your GPU's. I've not seen any part of any game that draws more power than this.

I guess it's the discrete nature of standards we impose for PC building and configuration. It's always been the case that OEMs choose their parts with a PSU of limited additional expansion capacity, although I've never had an expensive boutique OEM machine with SLI and the corresponding price-tag, so I wouldn't know how much Lebensraum Alienware or others provided.

Nothing wrong, really, with such a draconian test scenario for loading CPU and GPUs to the max, but maybe the analysis is "worth it" in terms of any related issues in the landscape like "variation, peak and mean GPU power consumption."

No reason there shouldn't be a "real-world software" or "gaming application" standard to compare.

So I'm thinking that the difference between my 650W seasonic and a 750 or 850W unit is "confidence:" "How confident would one be that running both Prime95 and Kombustor at the same time on your rig (SLI or otherwise) won't lead to unpleasant -- if totally recoverable -- results.)

Then it boils down to "estimation vs actual testing." Are you CONFIDENT enough about the estimates to totally abjure testing them?

At this point, it's like asking an enthusiast-consumer whether he's inclined to run an "Overload Test" on his PSU the way it was done in the review I linked pertaining to the XFX Black Pro 750 (and rebadged Seasonic 750X).

Who exactly on these forums do we know who ever ran such a load test, unless it was work-related or subsidized by a manufacturer who sends out samples to HardWare Secrets, TechReport, XBitLabs and others?

But an OEM would certainly do such a thing, especially if planning to make or buy parts in bulk for their well-bundled PCs.

So I'd conclude that enthusiast-consumers are more likely to buy excess capacity with their PSUs, even if tutored with "Extreme Outervision" -- on a statistical basis, anyway -- call it "on the average." And consumer wannabe-enthusiasts will be disappointed when they try to add a 4-disk HDD RAID array and controller card to their . . . OptiPlex. I'll bet just about everybody who ever explored the inside of the PC chassis has had that experience at one time or another -- more likely the longer they've been fiddling with the technology.

 

[jamesd168]

I guess it's the discrete nature of standards we impose for PC building and configuration. It's always been the case that OEMs choose their parts with a PSU of limited additional expansion capacity, although I've never had an expensive boutique OEM machine with SLI and the corresponding price-tag, so I wouldn't know how much Lebensraum Alienware or others provided.

Nothing wrong, really, with such a draconian test scenario for loading CPU and GPUs to the max, but maybe the analysis is "worth it" in terms of any related issues in the landscape like "variation, peak and mean GPU power consumption."

No reason there shouldn't be a "real-world software" or "gaming application" standard to compare.

So I'm thinking that the difference between my 650W seasonic and a 750 or 850W unit is "confidence:" "How confident would one be that running both Prime95 and Kombustor at the same time on your rig (SLI or otherwise) won't lead to unpleasant -- if totally recoverable -- results.)

Then it boils down to "estimation vs actual testing." Are you CONFIDENT enough about the estimates to totally abjure testing them?

At this point, it's like asking an enthusiast-consumer whether he's inclined to run an "Overload Test" on his PSU the way it was done in the review I linked pertaining to the XFX Black Pro 750 (and rebadged Seasonic 750X).

Who exactly on these forums do we know who ever ran such a load test, unless it was work-related or subsidized by a manufacturer who sends out samples to HardWare Secrets, TechReport, XBitLabs and others?

But an OEM would certainly do such a thing, especially if planning to make or buy parts in bulk for their well-bundled PCs.

So I'd conclude that enthusiast-consumers are more likely to buy excess capacity with their PSUs, even if tutored with "Extreme Outervision" -- on a statistical basis, anyway -- call it "on the average." And consumer wannabe-enthusiasts will be disappointed when they try to add a 4-disk HDD RAID array and controller card to their . . . OptiPlex. I'll bet just about everybody who ever explored the inside of the PC chassis has had that experience at one time or another -- more likely the longer they've been fiddling with the technology.

one thing that can happen also is if the operating environment changes, say temperature gets really high, the power supply efficiency will go down and will not be able to put out rated power. in that case if you have extra capacity, you will still be fine.

 

[PhIlLy ChEeSe]

OK,
I didn't read everything but got most of this. If your over clocking/benching then probably buy a bigger one, if your just gonna over clock and a run a few games but not leave the unit running 24/7 I see no issue. Newer PSU arent going to leave you in a blaze of glory while your gaming. Nor take out stuff if it does go bad, not to say it can't or wont. It will start to cause issue's, crashes ETC.

 

[BonzaiDuck]

OK,
I didn't read everything but got most of this. If your over clocking/benching then probably buy a bigger one, if your just gonna over clock and a run a few games but not leave the unit running 24/7 I see no issue. Newer PSU arent going to leave you in a blaze of glory while your gaming. Nor take out stuff if it does go bad, not to say it can't or wont. It will start to cause issue's, crashes ETC.

Well, situation so far:

2700K OC'd to 4.7Ghz
2x SLI GTX 970's core: 1,440; memory: 3,750 (7,500)

System sleeps after 2 hours no activity; hibernates an hour after that.
WHS-2011 wakes it up at 1AM and continues to successfully back up the boot disk.

No problems with any games: Under those circumstances, I think the actual draw on the PSU is less than 400W.

Heaven Benchmark mildly loads the GPUs to 98 or 99% and may draw 400W. I think I set it up with "normal" tessellation and 4xAA with 1920x1080.

No crashes, lockups, or Event Log yellows or reds that can't be accounted for in other ways (like I slipped and punched the PC power button by mistake with Event ID 41 after reboot.)

Seems rock-stable.

Then there's an anomaly about what appears to be the same version of Kombustor (September 2014) run either from its own icon in Programs, or through the "K" button on the AfterBurner tweaking window. Former shows AA 2x 1920/1080 and the GPU burn test loads one core to ~98% and the other to ~80%. This is showing about 430W at the wall. I suspect that through the Start-Programs path, there is one combination of settings applied, while the link in AfterBurner has another set.

The latter access to Kombustor shows 570W at the wall. I think the tests or the settings for the tests must be different, depending on which installation is loaded (but there should be only one installation). I may have tweaked "tessellation" or AA, and I think I had AA set to 4x with Heaven Benchmark -- which I already said draws 400 or less.

570 at the wall translates to ~ 513 actual output of the PSU. If I loaded both CPU and GPUs with Prime95 and Kombustor, CPU would add 90W measured through HWMonitor. This is close to all the other estimates I made, or 603W actual draw on the PSU.

So I'll AGREE that I might start looking for another PSU while I watch the numbers on this one -- its voltages against the first ones I ever recorded for it. And I'm looking at 750W and 850W units.

But the $100 to $150 in prospective outlay is not burning a hole in my wallet, either.

Also, let me say I found a 2013 analysis of power-supplies, OEM manufacturers, designers, and companies involved only in the labeling or "branding" of the retail product. There are ALL these interlocking connections! But there are maybe two "brands" who market extensively to other firms like PCP&C, XFX or Corsair, but who do all the manufacturing for their own label as well. And one of those is Seasonic. Seasonic doesn't buy PSUs from anybody else: it only makes them.

Wish I'd saved the link, but it had a table that ran through the alphabet of "brands" and was 10 pages long. Very interesting.

UPDATE/AFTERTHOUGHT: My most recent understanding of PSU specification practices is that the PSU is rated for the MEAN peak wattage. This would suggest that "power spikes" in new graphics cards like the GTX 970 are even less consequential in assessing the PSU wattage requirement. If the PSU is rated for the MEAN, it should also handle the transients.