The Math Behind GPU Power Consumption And PSUs

[beetlejuicegr]

Ohh and another thing, i bet its related.
My R9 290 is way more stable (almost not black screens or game crashes) if i "Force constant voltage" in the card from MSI afterburner. I wonder if that's the case this article describes

 

[Johan Kryger Haglert]

bettlejuicegr: Nah. I don't think there ever was any talk about 2 PSUs.

Just how much power a cable of 6 or 8 wires was designed for - And that having an extra connector on the same cable doesn't (necessarily?) mean it was designed to carry twice the power.

So if it's things which require little power maybe it's ok to use the double connectors at the other end.

However if it's a power demanding card or two of them and they would be close to the limit what they mean is to connect two PCI-expres power cables rather than using one cable and the two connectors on it.

I don't remember what was said above but this:
http://www.playtool.com/pages/psuconnectors/connectors.html#pciexpress8
Say 150 watt for an 8-pin PCI-expres cable.

Like this GTX 970 have two connectors 8 + 6 wire:
http://cdn.videocardz.com/1/2014/09/Galaxy-GeForce-GTX-970-GC-4GB-1.jpg

The normal TDP for a GTX 970 is 145 watt. Whatever that mean it's ok to use a 8 wire PCI-express cable with two connectors because the cable was designed for 150 watt I don't know.

Or whatever there's a reason it has two connectors and one should really use two cables because maybe draw occasionally can get higher (or for overclocking purposes?), though I guess it would be good if someone went in here and told whatever one cable with two connectors on the other end will mean it's designed for 300 watt or not.

If nothing else you shouldn't split it even more so you take one PCI-express cable and split each connector and hook up a 295x2 on each.

I shouldn't be answering this I feel =P, but I think it was there in the text and it was about using two cables when so was desired rather than splitting things in whatever way it was done. It was never about using multiple PSUs (multiple rails possibly, )

 

[FormatC]

The normal TDP for a GTX 970 is 145 watt. Whatever that mean it's ok to use a 8 wire PCI-express cable with two connectors because the cable was designed for 150 watt I don't know.

The TDP of 145 Watts for the GTX 970 is Nvidias reference number, nothing else. The Gigabyte GTX 979 Gaming G1 has for example a power target of 250 (280) Watts to keep the boost as high as possible! But this is far away from reference and needs other connectors!

If wrote a few weeks ago an in-depth analysis about the different power targets and it is very funny to see, how the efiiciency goes to hell with such high power targets. Again a marketing horror:

 

[Johan Kryger Haglert]

Care to explain in short:

TDP - It's the maximum amount of heat one have to lift off from the product? Shall I assume all energy become heat and hence power demand is the same? It's say the GPU or the whole card?
Can power demands increase higher than that in short enough time periods? Is it an "average" on as high as possible load? Or is it the maximum load/power ever in any time period?

I assume number of connectors on the card is due to the standard for how much power each card is supposed to be able to deliver?

What is the current designed for maximum power for 6 and 8 wire PCI-express power cables?

If the PSU end of the cable have 8 wires do that always mean it's just designed to power one 8 wire PCI-express device really? Or do two connectors on the other side mean it should be able to power two?

Some (all?) of that may have already been answered in the article.

 

[skit75]

I know this could be a noob question but i will do it. So, is it better to use 2 PSUs of 500watts, giving one 8pin PCI-E cable each to a hungry GPU each and splitting the rest of the PC power needs than having a 1000watt of same quality PSU? Is that it?

The motherboards I have seen that allow for two PSU's to be connected are more for redundancy, not capacity.

 

[webbwbb]

This brings up a good point. GPU manufacturers could smooth out power spikes as seen by the PSU by including their own caps and therefore make their cards more compatible with standard power supplies.

 

[anxiousinfusion]

I really hope Tom's continues to produce these in depth technical articles. Rarely do I find myself able to read through several pages on such a subject.

 

[06yfz450ridr]

good read thanks

 

[gondor]

What is this nonsense about "3330" microfarad capacitors? There are 3300 uF, there is no such thin gas 3330 uF.

 

[FormatC]

Call it (a) typo or (b) nitpicking

(this text was translated / moved and is always correct in German original)

There are more other translation problems. I wrote about the slower regulation of this digital PSUs, not the lower switching clock rate etc... I hope that we are able to correct this soon.

 

[rizdhan]

wow, this is a very serious article, i have to read it 3x on each pages to get to know... excellent! now i'm more aware when picking up GPU & CPU configurations.

 

[Ninjawithagun]

Ninja, I didn't quite get the same take. A lot of people are indeed inclined to buy the minimum-rated PSU, and the article explains why that may not be the best idea. Still, minimum rating wouldn't be so bad (if the quality is otherwise excellent), as long as they have enough filtering capacity to overcome those momentary spikes.
My PCs are mostly unremarkable, but I think I put my SG-650 in my will. It's "only" 80+ bronze, but I'll probably croak before it does.

...famous last words before your power supply goes "siiiizzzzzllllle - POP!" in the next few months...lol

 

[Ninjawithagun]

Call it (a) typo or (b) nitpicking

(this text was translated / moved and is always correct in German original)

There are more other translation problems. I wrote about the slower regulation of this digital PSUs, not the lower switching clock rate etc... I hope that we are able to correct this soon.

The newer generation all-digital power supplies do not exhibit these characteristics that are outlined in this article. For example, the new Corsair AX1500i is an all-digital unit and does not exhibit any of the negative affects stated in the article. Older first generation power supply units that had both analog and digital conversion did exhibit some of the drawbacks listed in the article, which leads me to believe the article is outdated and thus invalid in some areas.

 

[FormatC]

One of the AX860i died with the R9 295X2 during the tests for the launch article, the 2nd runs not stable and was everytime switched off. The link software is slow and not really stable. And to be honest: it is not the first generation. We had longer discussions about this with some OEMs. FSP for example cancelled their Dyna series because it is always problematic to get the same fast regulation as the good old analog way.

 

[Eggz]

Isn't this digital vs analog discussion part of a theme in computing anyway? I don't know of any digital process that can fully replicate the best of analog in terms of quality, and it seems that a digital process would theoretically never exactly match an analog process. "Analog" really stands for things that naturally occur in the world, and a digital process isn't a natural occurrence.

That makes sense because doing something digitally entails an emulation process - forcing a machine to copy what occurs naturally. Whether it comes to power flow in a PSU, digital sensors vs film, digital vs analog sound recordings, or whatever you like, you face a real challenge. Namely, the laws of physics are the ultimate "real-time," and creating a parallel process digitally introduces complexity, processing requirements, and lag. That's not to say a digital process can't come very close, but that's about as good as things will get - very close.

In a very abstract way, it reminds me of a convergence problem, where summing a series to a finite point gives you an approximate value based on many rectangular areas rather than a single smoothed area, whereas getting an accurate sum can call for summing to infinity.

 

[FormatC]

You are right, the main problem is always the granularity. Ok, you can get a very close result (for example in Audio) but the expense is very high. Nothing for PSUs

All this things like the "real-time" monitoring software of this PSUs is pure marketing and says simply nothing. The numbers you see are something else but not exact or real. This is only an estimated value in too large intervals, not more.

 

[Lucie2A]

Perfect post ! Keep going !

 

[damric]

Good job on the capacitor explanation, but there is still some room for more information there.

Bottom line:

STOP TRYING TO POWER YOUR GRAPHICS WITH CORSAIR CX TIME BOMBS.

 

[damric]

Should also have shown how cheapo caps degrade much faster in higher temperature environments. This is a real problem for passive/semi-passive and hot running PSUs in general.

 

[Onus]

^That is why Corsair CX (and GS) PSUs ultimately choke and croak (and sometimes, if YouTube is to be believed, smoke), although they don't kill attached parts like cheap PSUs are known to do.

 

[juanjostorreshernndez]

Wow.
Great Job.