Are PSU calculator sites accurate?

[Danyune]

I currently have a 700w. My system:

4790k @ 4.4
16gb 2400mhz
R9 270
5 fans, 3 high performance fans (total of 8)
1 SSD
1 HDD

I ordered a GTX 980 classified to replace the ATI and the calculator told me that it recommends me a 534w. I was concerned about my PSU not being strong enough.

It's a coolermaster PSU so it's not a generic brand, but there is no 80+ cert on it (RS700)

Think it'll be ok?

 

[lehtv]

Why are you concerned if your current PSU is labelled 166W more powerful than what the calculator recommends? In terms of specifications, 700W should be easily more than enough to run your system.

However, what you should be concerned about is the quality of your unit. Yes, Cooler Master is not a 'generic brand', but unfortanately for the customer, their units vary from disgusting to mediocre all the way to top tier, making it impossible to determine the quality of a particular unit based only on the brand name.

The PSU review database lists 11 units where the model number has "RS-700" in it, and of those, these are the ones lacking 80 Plus certification:

RS-700-ASAB-L3 (GX Lite, built by ATNG)
- an OK unit, but nowhere near the quality of your other hardware

RS-700-PCAA-E3 (Extreme Power Plus, built by Seventeam)
- Seventeam isn't regarded high tier and warranty is only 3 years which implies low quality components, but performance could be relatively good

RS-700-ACAB-D3 (Thunder Series, built by Enhance)
- only in Asian markets, I think

RS-700-ACAB-D3 (B series, built by Enhance)
- impressive 55A single +12V rail, and Enhance is generally not bad... but still, nowhere near the quality of the rest of your hardware

Overall, while your PC would probably continue to work fine with the existing unit, I'd much rather not leave any doubt. Powering well over $1000 worth of hardware with a unit that leaves doubts as to its reliability is just not a good idea. Upgrade to a high quality unit, and sell the RS-700 or keep it as an emergency backup.

 

[Valantar]

OMG, NO, you need to get at least a 1000W PSU!!!!!!!

All joking aside, it depends on what you consider accurate.

Under no circumstances will your entire system be running under full load. However, 100% GPU load + a high CPU load is not uncommon, for example when gaming - and this might cause a drive or some other parts to consume some power too. Still, if you take a look at AnandTech's GTX 980 review, their system (with a higher TDP CPU than yours) was using just 304W running Crysis 3 (or just 294 in FurMark!). The GTX 980 is ridiculously efficient. With a 700W PSU, as long as it isn't complete crap and the rating is continous rather than peak, you could easily run two GTX 980s in SLI.

My rule of thumb has become this: add the TDP of your CPU and GPU (the two main power draws, in your case 84W+165W), add 70-80 watts for other components (unless you have more than one 3.5" HDD, if so, add ~30W for each), and multiply by 1.5. This way, you'll end up around the sweet spot of both noise and efficiency of most PSUs, keep well away from overloading it. By applying my calculations to your setup, you get 478.5W. This should give you plenty of room for overclocking on a 700W PSU (if you pushed your CPU to 120W and your GPU to 250W, which are quite unrealistic numbers for your components, you're still not reaching 700W with my calculations). In fact, I would say your PSU is quite overpowered for this build, as you would only achieve peak efficiency (usually around 50% load) if your PC was running at 100% CPU and GPU.

TL;DR: Don't worry, be happy, keep your PSU, it's more powerful than you need.

 

[lehtv]

Under no circumstances will your entire system be running under full load. However, 100% GPU load + a high CPU load is not uncommon, for example when gaming - and this might cause a drive or some other parts to consume some power too. Still, if you take a look at AnandTech's GTX 980 review, their system (with a higher TDP CPU than yours) was using just 304W running Crysis 3 (or just 294 in FurMark!). The GTX 980 is ridiculously efficient. With a 700W PSU, as long as it isn't complete crap and the rating is continous rather than peak, you could easily run two GTX 980s in SLI.

*Ahem*

With Maxwell architecture, average wattage during load (such as Crysis 3) is not a good metric for determining required PSU wattage. This is because a Maxwell GPU's power consumption is incredibly variable. When you look closely, as Tom's Hardware did, power consumption can peak up to 350W for the stock clocked GPU alone, even though average power consumption is congruent with Anandtech's results:

Because of these power spikes, the PSU is often and repeatedly supplying currents way above the average current that the card is using. The PSU should be chosen according to the average peak wattage of the GPU, which looks to be around 300W judging by the graph posted above.

Thus, no, you should never run GTX 980 SLI on a 700W unit; the cards themselves can alone draw 700W worth of amperes simultaneously. It is not going to end well unless the unit is dramatically underrated.

 

[Danyune]

Well I thank you all for your answers, but I seemed to have answered my own question but physically looking at the card that arrived lol, I should have checked the simplest things first.

My 700w doesn't have the 8pin connectors for PCI-E, only two 6-pin ones, so I went ahead and put in a HX850 and it runs everything fine.

Thank you all for doing research though!

 

[MagnusTheBrewer]

Well I thank you all for your answers, but I seemed to have answered my own question but physically looking at the card that arrived lol, I should have checked the simplest things first.

My 700w doesn't have the 8pin connectors for PCI-E, only two 6-pin ones, so I went ahead and put in a HX850 and it runs everything fine.

Thank you all for doing research though!

Buying a PSU for the connectors it has is like buying a car because it's blue. All of the components in your computer are rated for the power they consume. You can simply go for a bigger power supply as you did however, the efficiency of a unit is not uniform across it's output. By buying too large a PSU, you're simply inflating your power bill.

 

[SOFTengCOMPelec]

By buying too large a PSU, you're simply inflating your power bill.

Modern PC (switching) Power supplies only use as much power, as the computer needs (plus a small power % loss, of around 10% or so, on a modern psu), so too big is fine, as regards power consumption.
Except that a bigger than you need power supply, may be slightly/fractionally less efficient, but usually by such a small amount, that it can be neglected. It may actually be more efficient (slightly) in practice, depending on the power supplies efficiency curve versus %of max output power, supplied.

 

[lehtv]

Well I thank you all for your answers, but I seemed to have answered my own question but physically looking at the card that arrived lol, I should have checked the simplest things first.

My 700w doesn't have the 8pin connectors for PCI-E, only two 6-pin ones, so I went ahead and put in a HX850 and it runs everything fine.

Thank you all for doing research though!

HX850? So you bought the first unit that you happened upon which had 8-pin connectors? HX850 has four such connectors, making it a GTX 970 SLI compatible unit. A decent quality 550W-650W unit with two 8-pin connectors would've been fine, and saved you quite a lot of money.

 

[lehtv]

Buying a PSU for the connectors it has is like buying a car because it's blue. All of the components in your computer are rated for the power they consume. You can simply go for a bigger power supply as you did however, the efficiency of a unit is not uniform across it's output. By buying too large a PSU, you're simply inflating your power bill.

I think this argument is a bit ... academic. In practice, the efficiency curve of a PSU is not dramatic enough for electricity bills to be noticeably affected by the size of the PSU. Perhaps you'll pay $1 or $2 more per year for running a 850W unit than a 550W one; possibly not even that, as efficiency curves vary between units. The fact that a 850W unit costs much more to begin with is a much more significant consideration, to the point that the efficiency curve doesn't even need to be mentioned.

 

[Valantar]

I think this argument is a bit ... academic. In practice, the efficiency curve of a PSU is not dramatic enough for electricity bills to be noticeably affected by the size of the PSU. Perhaps you'll pay $1 or $2 more per year for running a 850W unit than a 550W one; possibly not even that, as efficiency curves vary between units. The fact that a 850W unit costs much more to begin with is a much more significant consideration, to the point that the efficiency curve doesn't even need to be mentioned.

Quite true. One should also factor in other, more tangential concerns, such as the increased environmental impact of producing higher wattage PSUs (even if the impact of making one 850w unit compared to a 550w unit is miniscule, that changes when you're talking thousands of units - and yes, every unit makes an impact).

Also, Danyune, if you factor in time spent in or near idle (which with modern race-to-idle computing is quite a lot outside of gaming or other heavy continuous workloads), you're concievably operating at less than 10% output (85w) from the PSU for extended periods of time - an area where the efficiency of all but the best units is quite atrocious. In the AnandTech GTX 980 review, in a system with an overclocked Ivy Bridge-E i7-4960X (quite a bit less power efficient than a Haswell i7-4790, even at stock), the system idled at 73w - measured at the wall! Yours should thus idle at even lower wattages. Again, we're not talking many watts total, but still a completely unnecessary waste of electricity.

Even if you consider lehtv's (in my opinion quite pessimistic) power draw requirements, 850w is beyond overkill. You're looking at a system with a peak (torture/power virus) power draw of perhaps slightly more than 400w - never more than 450w. So even in sustained, completely unrealistic scenarios, you're just barely approaching 50% load - which is the area where the PSU operates most efficiently. This is just plain silly.

lehtv: You're right that what I first said here is a bit optimistic - I should probably have said theoretically rather than easily - but I still think it's true given a high quality PSU. Mind you, I wouldn't recommend it, obviously (see my "calculator" in my first reply), but it should be possible. The chance of these 350w peaks coinciding on two cards for a simultaneous power draw of 700w from two GTX 980s seems quite pie-in-the-sky, given that they last less than a second, and appear (in this graph) 2-3 times a minute (given that SLI ideally divides successive frames between the cards, this should also help avoid simulatneous peak loads when combined with the incredible adjustment speed of PowerTune 3/GPU Boost 2). Still, this would of course require a quality PSU with a peak output well beyond 700w for safety and longevity's sake. Also, given that GTX 970s and 980s aren't killing PSUs in a large scale (or at all, that I've heard of), it seems safe to assume that most PSUs are fully capable of keeping up with these rapid fluctuations. And yes, even though there are "sustained" peaks of ~300w in the graph, at the same time there are just as "sustained" troughs of ~100w. The average power draw is still the average power draw, which is what the PSU has to cope with over time. So while these peaks are of course worth factoring into a PSU purchase decision (i.e. don't run a GTX 980+84W Haswell on a 350W PSU, which "works" when only thinking of TDP) considering them the actual power draw of the card seems overly pessimistic in my opinion.

 

[magomago]

*Ahem*

With Maxwell architecture, average wattage during load (such as Crysis 3) is not a good metric for determining required PSU wattage. This is because a Maxwell GPU's power consumption is incredibly variable. When you look closely, as Tom's Hardware did, power consumption can peak up to 350W for the stock clocked GPU alone, even though average power consumption is congruent with Anandtech's results:

Because of these power spikes, the PSU is often and repeatedly supplying currents way above the average current that the card is using. The PSU should be chosen according to the average peak wattage of the GPU, which looks to be around 300W judging by the graph posted above.

Thus, no, you should never run GTX 980 SLI on a 700W unit; the cards themselves can alone draw 700W worth of amperes simultaneously. It is not going to end well unless the unit is dramatically underrated.

That is a terrible method to select a PSU. I have no idea what rate the data was measured at, but its clear that it only breaks 350W for fractions of a millisecond. It only seems to break 300W for just a couple milliseconds as well.
Concepts like Duty Cycle exist for a reason - and PSUs can handle such an intermittent spike. When designing all forms of hareware systems, duty cycle and rms power matters a lot because you can get by with less.
On a PC, when analyzing game performance, we aren't 'designing' for anything, but we can make the very smart judgement that a 700W is more than adequate for the system listed above.

 

[lehtv]

That is a terrible method to select a PSU. I have no idea what rate the data was measured at, but its clear that it only breaks 350W for fractions of a millisecond. It only seems to break 300W for just a couple milliseconds as well.

The resolution of the oscilloscope that Tom's used for the 60 second graph I posted was 1 millisecond, not fractions of a millisecond.

But yes, the power spikes last only for a short time - yet they occur all the time repeatedly. Over a longer period of time, the PSU will be spending a lot of its time handling 300-350W loads from the graphics card, which isn't a conclusion you can arrive at just by looking at the card's average power consumption.

And how is it a terrible method? The conclusion I would personally draw from Tom's Hardware's data is that NVIDIA's 500W minimum recommendation is correct and makes sense. Having 500W as a minimum leaves enough room for the spikes to occur without ever pushing the PSU too close to what it's capable of, assuming no overclocking. This isn't the case with Valantar's assertion that 700W is easily enough for 980's in SLI. A stock clocked system with 980 SLI can regularly, though for only short periods at a time, exceed what the +12V rail spec of a 700W power supply. If 500W is correct for a card that can spike up to 350W, then for two such cards you should be looking at 850W unit as a minimum.

Concepts like Duty Cycle exist for a reason - and PSUs can handle such an intermittent spike. When designing all forms of hareware systems, duty cycle and rms power matters a lot because you can get by with less.

Can you explain to me what duty cycle has to do with any of this? I understand the concept but I don't understand how it's relevant to Tom's Hardware's results.

but we can make the very smart judgement that a 700W is more than adequate for the system listed above.

I never said 700W wasn't adequate for a single GPU. I was responding to the claim that 700W would be adequate for 980 SLI.

 

[ElFenix]

*Ahem*

With Maxwell architecture, average wattage during load (such as Crysis 3) is not a good metric for determining required PSU wattage. This is because a Maxwell GPU's power consumption is incredibly variable. When you look closely, as Tom's Hardware did, power consumption can peak up to 350W for the stock clocked GPU alone, even though average power consumption is congruent with Anandtech's results:

iirc there was some discussion as to whether or not electronic components show that sort of peaking anyway and it's just that now someone decided to look at what power draw is at that fine a resolution. iirc many amateurs were alarmed but the electronics engineer types thought it was standard behavior.

not sure what was eventually determined on that front.

 

[lehtv]

iirc there was some discussion as to whether or not electronic components show that sort of peaking anyway and it's just that now someone decided to look at what power draw is at that fine a resolution. iirc many amateurs were alarmed but the electronics engineer types thought it was standard behavior.

not sure what was eventually determined on that front.

Is it this one? http://forums.anandtech.com/showthread.php?t=2400497

I'm gonna read it tomorrow...

 

[richaron]

iirc there was some discussion as to whether or not electronic components show that sort of peaking anyway and it's just that now someone decided to look at what power draw is at that fine a resolution. iirc many amateurs were alarmed but the electronics engineer types thought it was standard behavior.

not sure what was eventually determined on that front.

I suspect whether people are alarmed or if they consider these spikes 'standard behavior', they will spec out their PSU in the exact same way. It's those who are blind to the possibility of spikes who could have problems.

I remember another Tom's article about PSUs for the 295X2, and how there were problems with certain units due to these spikes (at least that's what I got out of it, I'll try link it). But of course it's not just the wattage of the units that matters, and I'd bet those electronics engineer types have a few tricks up their sleeve; my PSU shipped with big capacitors in the PCIe power cables.

 

[BonzaiDuck]

A stock clocked system with 980 SLI can regularly, though for only short periods at a time, exceed what the +12V rail spec of a 700W power supply. If 500W is correct for a card that can spike up to 350W, then for two such cards you should be looking at 850W unit as a minimum.


Can you explain to me what duty cycle has to do with any of this? I understand the concept but I don't understand how it's relevant to Tom's Hardware's results.


I never said 700W wasn't adequate for a single GPU. I was responding to the claim that 700W would be adequate for 980 SLI.

Interesting discussion, and big in my mind right now. I had posted my own thread about this -- almost with the same title -- less than a month ago.

Sustained power consumption readings are one thing; spikes are another.

I built a second system similar to my sig-rig in September/October last year, and it was an afterthought. the Sig-rig was -- at least -- well-planned. The second system (with a 2700K) was only planned according to expectations of the sig-rig.

As "available" parts for #2, I had a 650W Seasonic "semi-modular" unit, practically brand-new. I may have anticipated only a single GTX 970, but I put an SLI 2x configuration into the Extreme Outervision choices. EO showed 750+W, but I was still skeptical. I did some "calculations" of my own, and concluded that 650W would be adequate. THEN -- I ordered a second GTX 970 last week.

Figure maximum possible CPU power at the existing OC is around 140W. The GTX 970 specs say "145W;" 110% of that is ~160W; and double that is 320W. I was in for a surprise, though . . .

With 2x GTX 970 and the features enabled so that Kombustor pushes both cards to around 98% power-consumption, the draw on my UPS software shows about 510W usage at peak. But this is under conditions where the CPU package power peaks at 50W -- not 140W. In a mild gaming situation like GRID2, the UPS monitor shows about 420W.

My previously optimistic view of this admitted a comfort level for OC'ing the cards. At this point, I'm not sure that's a good idea -- yet.

Obviously, my method of gauging peak power-consumption is . . . well . . . "less than primitive." Or -- it depends on what one thinks is adequate. After all, there are some other minor devices powered off the UPS being monitored.

I may spring for a 750W unit soon, but it's no emergency. Of course, some could try and convince me that it is . . .

 

[TemjinGold]

Note that your UPS is showing you power draw from the wall, not the power those components are using.

 

[BonzaiDuck]

Note that your UPS is showing you power draw from the wall, not the power those components are using.

If that's the case -- for an 80+ "Gold" unit -- it would mean that actual power-consumption of the components is LOWER?!? And I'm thinking that your focus with that is PSU efficiency?

 

[Z15CAM]

I rely more on Johnny Guru and Hardware Secretes for PSU Reviews when building a Platform - You know God Damn well - Don't skimp on the PSU.

Personally. I've had no issues with XFX BEFX Pro Black Edition PSU's when it comes to Money Vs Quality, Performance and being Quiet.

 

[BonzaiDuck]

I rely more on Johnny Guru and Hardware Secretes for PSU Reviews when building a Platform - You know God Damn well - Don't skimp on the PSU.

Personally. I've had no issues with XFX BEFX Pro Black PSU's when it comes to Money Vs Quality and Performance.

Well, there are a lot of good PSU's "out there" these days, and it's also easy to become comfortable with a "proven" brand. That being said, I've seen more customer-reviews of Seasonic units written by unhappy enthusiasts. But also true-- those reviews are biased toward a greater number of DOA and other problems as a fraction of the total purchased: the dissatisfied are more likely to post a review than those without problems.

I don't doubt the quality of my Seasonics. It would make a difference, if -- as someone noted -- that "calculators" attempt to accommodate worst cases. For me at the moment, the question is whether my fairly new Seasonic 650W is adequate. Let me emphasize that I didn't buy it in anticipation of SLI, nor did I originally intend to use it for this system.

There are even "mystery" factors, such as how or whether the PSU builders made their units capable of dealing with the "spikes" under discussion.

You'd think that the upper half of the wattage range shown in the graphs of this thread is the meaningful part -- the difference between the mean power consumption and the highest spikes. For the most part, that looks like about 100W. Are we wrong to ignore the "outlier" spikes in that distribution?

So between you and TemjinGold, and between everybody here and posts I've seen at Tom's dealing with the very same issue -- the question for me is "Do I really need to replace the 650W with a 750 or 850 unit?"

Or conversely, now that I've explored the single-GPU clock range for the GTX 970, should I bump it up at least half as far?

Just today I saw a really heated discussion of some few months ago -- a "back and forth" of "No-- your 660W isn't adequate" and "Yes it is for 970-SLI."

EPILOGUE/AFTERTHOUGHT: Just did a search on those XFX BEFX models. Re-badged/re-branded Seasonics!

http://www.hardwaresecrets.com/arti...Edition-Full-Modular-Power-Supply-Review/1716

Nice to know, I guess . . . But like anything else, it's easy to get comfortable with an established brand that won lots of kudos. Good idea to refresh with recent comparison reviews and details of "ripple" and other factors.

Ultimately, the 5-year or 7-year warranties mean something up front. . .

 

[BonzaiDuck]

Still relevant to the OP, I suppose . . .

I just ran some more tests on my own "predicament" and did some arithmetic. Again -- reading the wattage "at the wall" with the PSU software. Idle: 171W. OCCT:Linpack boosts that to ~310W, so figure the cost of OC'ing is ~140W -- the original stress-load package-power. Under Afterburner, I said the loaded 970 graphics cards in SLI push it to 510W, so 510 - 171 = ~340W.

The worst-case scenario would therefore be 171W + 140W + 340W = 651W.

So a 650W is definitely "on the edge." A 700W or 750W would be "adequate" with wiggle-room. If the OP isn't overclocking, the 700W PSU is fine for a single GTX 980. Awfully close for two, and then -- who knows? I'd spring for an 800 or 850W PSU.

Two aspects to this: a QUALITY PSU -- many brands and models to choose from; and an ADEQUATELY POWERED system. Whether your real-world system is going to load up the OC'd CPU and graphics cards all at once -- maybe not likely. But it's probably worth it to give yourself an extra 100W+.

That still leaves the issue about "mean power" versus "power spikes" sort of hanging, but at least the additional 100W would leave it less of an issue. . . .

[Mmm . . . . Interesting . . . a 100 Mhz increase in SLI'd 970 clock speeds with no voltage adjustment means another 30W of power consumption. . . ]

 

[TemjinGold]

Now multiply that 651w by the efficiency percentage to find out how much your system is actually using. Remember, a 700w unit DELIVERS 700w. It pulls a lot more than 700w to do that.

 

[BonzaiDuck]

Now multiply that 651w by the efficiency percentage to find out how much your system is actually using. Remember, a 700w unit DELIVERS 700w. It pulls a lot more than 700w to do that.

So if the PSU is rated 90% efficient, and since I'm measuring power at the UPS or the wall, whichever -- than my system would actually be using 585W if my determinations of the 171W-idle, 140W of full CPU load, and 340W due to full graphics load were based on readings at the UPS?

Are we saying that I actually have 65W of capacity left on a 650W PSU?

Is that what you're saying and what I think you're saying?!?

It really seems to make sense. Makes even more sense for my Quicken accounts -- hah!

I'm surprised I didn't think of it that way!

 

[TemjinGold]

Yes that's what I'm saying.

 

[Smoblikat]

Buying a PSU for the connectors it has is like buying a car because it's blue. All of the components in your computer are rated for the power they consume. You can simply go for a bigger power supply as you did however, the efficiency of a unit is not uniform across it's output. By buying too large a PSU, you're simply inflating your power bill.

This is completely wrong. A PSU doesnt use the total amount of wattage it can supply, it only uses as much as is requested by the parts attached to it.