How to work out actual wattage GPU is drawing

[BakGat]

Hi Guys

If your GPU is drawing 423 Watts at 1.092 V , how would you work out the clean power the PSU has to provide in order to deliver reliable voltage?

I am sure there are some electrical gurus on this site who could inform my small mind how it all works.

Thanks.

 

[John Naylor]

Most people rely on tools that don't do what they think they do ... For example GPU-z, an application I have reverence for does not actually measure the full power draw on AMD cards

How Much Power Do GPUs Actually Consume?

Graphics cards, like CPUs, don't always communicate exactly how much power they use. The differences have implications for overall GPU efficiency comparisons.

www.extremetech.com

There are software tools for reporting GPU power consumption, but apps like GPU-Z still depend on the GPU telling the application how much power it’s using. Nvidia GPUs report total power consumption to the program fairly accurately, but AMD only reports its actual GPU core power consumption, not the impact of the rest of the board. This doesn’t impact system-level power measurements drawn at the wall, but it does hide the impact of GPU VRM and RAM power consumption from the GPU-Z application itself.

It also doen not measure GPU RAM Usage, it measures VRAM allocation

Is 4GB of VRAM enough? AMD's Fury X faces off with Nvidia's GTX 980 Ti, Titan X

Is 4GB enough for a high-end GPU? We investigated and tested 15 titles to find out.

www.extremetech.com

GPU-Z doesn’t actually report how much VRAM the GPU is actually using — instead, it reports the amount of VRAM that a game has requested. We spoke to Nvidia’s Brandon Bell on this topic, who told us the following: “None of the GPU tools on the market report memory usage correctly, whether it’s GPU-Z, Afterburner, Precision, etc. They all report the amount of memory requested by the GPU, not the actual memory usage. Cards will larger memory will request more memory, but that doesn’t mean that they actually use it. They simply request it because the memory is available.”

The way I look at this, lacking the expensive equipment and time necessary to draw applicable power readings is as described in the 1st link.

Nailing down how much power PC components consume is a surprisingly difficult question. It’s easy to answer how much power a system draws at the wall — a basic Kill-A-Watt meter doesn’t cost much, and while it doesn’t offer features like power consumption tracking over time, it’ll still provide good bulk information if you track it manually for several minutes during a consistent workload.

Tracking the exact power consumption of a given component, be it CPU, GPU, or something like RAM, requires a soldering iron and some skill in using it. ....

Like CPUs, GPus define a TDP rating. As with CPUs, that TDP rating is best thought of as a metric for GPU power dissipation (meaning, how much heat the cooler needs to be able to deal with) as opposed to an exact power consumption metric. Neither Nvidia nor AMD guarantee that a 150W GPU will draw exactly 150W, for example.

So the approach I follow is to look at the power at the wall and then, using the efficiency appropriate for the % load, calculate back to the DC wattage being supplied to the system. Then, one can look at respected tests reviews for the measured voltage variations and electrical noise at the % load in the ttest tabes in the review. If ya had this PSU and you were outputting 420 watts or so, based upon the results you see starting at the link below, id be a very happy camper.

Seasonic FOCUS Plus 550 Gold Power Supply – Page 3 – JonnyGURU.com

www.jonnyguru.com

Now is your particualr PSU up to the same performance as the one tested ? .. That's a whole nother story.

 

[BakGat]

Thanks John.
But that would be just for my GPU? As it draws 423 Watts at 1.092 V by its self, and
I have noticed that memory overclock ( plus 1350 Mhz on afterburner) when running benchmarks like superposition and 3DMark can really increase the power/voltage load. With a 750 watt EVGA PSU I continuously get VRel Perf cap in GPU_Z, but when I use the Seasonic 1000 Watt Gold PSU I do not. I am under the impression that you need a power supply that can provide "clean" power to your entire system in order to get the best reliability/stability out of a GPU. Some social media "experts" even suggesting powering a GPU with a seperate PSU when doing extreme OC's to get a reliable power delivery.

I am new to this, but I find it fascinating, but, expensive..lol

Hope you are keeping safe in Long Island.

 

[thebluebumblebee]

Do you guys know how GPU testing is done here at TPU?

 

[agent_x007]

Vrel = Highest point on V/F curve you can get to.
It means card is limited by voltage/max. frequency in vBIOS table, and not power from PSU.

Card draws power from PSU, what you see in GPU-z or HWInfo is power it thinks it's drawing from all 12V lines. Depending on vGPU at the moment, actual load, and 12V value that PSU is providing - it can fluctuate.

 

[BakGat]

Thanks Agent

Sorry for late reply..was sleep time here in Austrlalia. Why is it that at same temps, mem overclock and clock overclock does the 1 psu show Vrel as a prfcap, and the other does not, both are gold rated?
I know jack about how wattage and voltage effect power delivery. One thing I can not get my head around is how voltage effects power draw from the psu?
@ thebluebumblebee..have no idea! On that note, about 2 weeks ago I uploaded my GPU bios from GPU_Z to this site for the VGA bios records and it has not shown up in the list of cards and bios's. Is there a reason for that?

 

[EarthDog]

As it draws 423 Watts at 1.092 V by its self,

Doubtful. Only a very few video cards could ever do that...

Why is it that at same temps, mem overclock and clock overclock does the 1 psu ls how Vrel as a prfcap, and the other does not, both are gold rated?

that doesnt make sense... the vrel has nothing to do with the power supply.

One thing I can not get my head around is how voltage effects power draw from the psu?

indirectly. If you raise the voltage, you raise the amount of wattage used by the card hence more draw from the power supply.

 

[BakGat]

Well Mr Earthdog
Doubtful. Only a very few video cards can do that!
see below. Colorful RTX 2080 TI Kudan limited edition

 

[Decryptor009]

Well Mr Earthdog
Doubtful. Only a very few video cards can do that!
see below.

Not too big of a surprise, i managed to pull amost 500 watts with as GTX 480 pushing 950+mhz on the core.

 

[EarthDog]

Well Mr Earthdog
Doubtful. Only a very few video cards can do that!
see below. Colorful RTX 2080 TI Kudan limited edition

Well mr bakgat, that would be the first unmodified 2080ti I see that could reach 400w!! That card is a special edition limited run though...not remotely normal.

What are you asking, exactly? Your question or point is lost to me. Very simply, if your gpu is using 400W and your psu is 90% efficient, that means your power supply is pulling 440W from the wall to power the gpu.

Not too big of a surprise, i managed to pull amost 500 watts with as GTX 480 pushing 950+mhz on the core.

A gtx 480 isnt an rtx 2080ti. Also, I've seen system draw of 500W with a 480, never a 480 itself pulling 500W. Anyway, I've also acknowledged a couple cards can do this (was thinking of dual GPU cards, r9 295x2, etc)... but why are we bringing up an ancient card to prove a point I've conceded??? Very few cards will use 400w...I dont need examples, lol.

 

[Vya Domus]

I honestly doubt that card is actually drawing that much power, something seems funky.

Anyway I don't really understand what is it that you want to know, you already know the power this card draws. A 750W PSU is probably pushing it, I imagine it would have about 600W on the 12V rails so when you factor in the CPU and whatever else runs off 12V you're getting pretty close to what the PSU can provide.

Card draws power from PSU, what you see in GPU-z or HWInfo is power it thinks it's drawing from all 12V lines. Depending on vGPU at the moment, actual load, and 12V value that PSU is providing - it can fluctuate.

It's not really what the card thinks it's drawing, there are actual resistors on board that are used to measure power within margin of error. These cards measure board power coming from both the PCIe connectors and the PCIe slots, in fact that's how people remove power limits with shunt mods.

 

[Decryptor009]

Well mr bakgat, that would be the first unmodified 2080ti I see that could reach 400w!! That card is a special edition limited run though...not remotely normal.

What are you asking, exactly? Your question or point is lost to me. Very simply, if your gpu is using 400W and your psu is 90% efficient, that means your power supply is pulling 440W from the wall to power the gpu.

A gtx 480 isnt an rtx 2080ti. Also, I've seen system draw of 500W with a 480, never a 480 itself pulling 500W. Anyway, I've also acknowledged a couple cards can do this (was thinking of dual GPU cards, r9 295x2, etc)... but why are we bringing up an ancient card to prove a point I've conceded??? Very few cards will use 400w...I dont need examples, lol.

487 watts on it's own, the 480 pulled with 1.213MV applied.

990mhz core, 2000 memory.

His card is overclocked so will draw more power.

 

[EarthDog]

487 watts on it's own, the 480 pulled with 1.213MV applied.

990mhz core, 2000 memory.

His card is overclocked so will draw more power.

My guy... a 480 has nothing to do with it. His power draw is listed as 423W. I've answered the question...... but no idea what the OP even wants.

 

[Decryptor009]

My guy... a 480 has nothing to do with it. His power draw is listed as 423W. I've answered the question...... but no idea what the OP even wants.

I made a point that GPU's can go way out of spec once you start overclocking.

I don't see any definitive answering at all, just backing one's own self up.

 

[EarthDog]

I made a point that GPU's can go way out of spec once you start overclocking.

I don't see any definitive answering at all, just backing one's own self up.

A point that isnt lost. But again.. 250W gpu with 39% power limit doesnt equal close to 423W. Following? A 480, iirc, isnt limited like Turing and a couple generations prior.

I answered his question in the OP.... how much power will the psu use for his card? I used the value in gpuz (regardless if it is accurate or not) to give him that answer.

But again, seems odd considering what we know about his card in the first place. Overclocking or not, UNLESS THERE ARE MODS TO THE CARD/BIOS, it is limited to what the power limit lists

Everything I've stated is factual. I dont care about a 480 or need to be told overclocking adds more power. But ty.

 

[Decryptor009]

A point that isnt lost. But again.. 250W gpu with 39% power limit doesnt equal close to 423W. Following? A 480, iirc, isnt limited like Turing and a couple generations prior.

I answered his question in the OP.... how much power will the psu use for his card? I used the value in gpuz (regardless if it is accurate or not) to give him that answer.

But again, seems odd considering what we know about his card in the first place. Overclocking or not, UNLESS THERE ARE MODS TO THE CARD/BIOS, it is limited to what the power limit lists

Everything I've stated is factual. I dont care about a 480 or need to be told overclocking adds more power. But ty.

You rely on pure numbers when GPU's have transistors that add leakage with heat and voltage creating even more power draw, this is why the GTX 480 was so bad, not inherently the numbers.

So 39% will be closer to 45-50% under some scenarios.

 

[Vya Domus]

Except a 480 didn't have a power limit so it could technically draw as much power as it's physically possible. But a 2080ti does and it's atypical for it to draw that much power.

 

[Decryptor009]

Except a 480 didn't have a power limit so it could technically draw as much power as it's physically possible. But a 2080ti does and it's atypical for it to draw that much power.

Yes, but the card will still draw more than TDP, this is why we see typically higher numbers under the right scenario than the listed TDP spec.
It's no different for CPU's which do have power limit's, current draw and even motherboards providing power limits where speed throttling will occur.

The chips will still hit that high number, then the chip will throttle.

Maintaining high numbers like that, yes Fermi could do that as any chip before GF100 could.

 

[EarthDog]

You rely on pure numbers when GPU's have transistors that add leakage with heat and voltage creating even more power draw, this is why the GTX 480 was so bad, not inherently the numbers.

So 39% will be closer to 45-50% under some scenarios.

nobody in this thread gives two hoots about a 10 year old gpu without power limits like modern cards. Please, move on. You are only serving to confuse the op. This isnt helpful.

 

[Decryptor009]

nobody in this thread gives two hoots about a 10 year old gpu without power limits like modern cards. Please, move on. You are only serving to confuse the op. This isnt helpful.

I never thought anyone did give a hoot, the principle is there, the same base methodology for my point of view, i was using the experience of a different card for data, it does not apply to a 2080Ti, it is just the base method of how voltage and heat effect power consumption along with the ramp up in consumption from raw numbers.

I was not here to disprove anybody, i was just giving my own view.

I tend to do that to people.. they just love to argue with me.

 

[EarthDog]

I never thought anyone did give a hoot, the principle is there, the same base methodology for my point of view, i was using the experience of a different card for data, it does not apply to a 2080Ti, it is just the base method of how voltage and heat effect power consumption along with the ramp up in consumption from raw numbers.

I was not here to disprove anybody, i was just giving my own view.

I tend to do that to people.. they just love to argue with me.

Your own view may not be right is the thing. Again, I agree with the high level thought, but outside of that, the rest doesnt apply here.

If people are incessantly debating your talking points, maybe it is time to look in the mirror and reflect on why.

Anyway, enough of this tangent, please. Waiting for the OP to clarify wth he wants.

 

[Decryptor009]

Apples and oranges. Move along.

If people are incessantly debating your talking points, maybe it is time to look in the mirror and reflect on why.

Anyway, enough of this tangent, please.

How about you move along? you seem so full of yourself.

The base method applies to all transistors, they are not made equal, but it still applies.

The real reasoning is that my view is different and it goes against what you deem as right, well i am glad i can't hold an opinion and i can't share any type of experience, since apparently you know all.

 

[EarthDog]

Yawn....

People also hold the opinion the earth is flat...

...just because it is an opinion doesnt mean it shouldnt be scrutinized or cant be flat out wrong. Again, we get your high level point but the 480 comparison is not applicable.. and the inefficiencies from heat etc dont add another 75-100w. Come on man. Please move on.

PS - I'm full of myself because I happen to be correct here. 480 comparison = useless. Info about overclocking using more power = obvious and clearly considered in my response, inefficiencies of hardware when hot = also obvious but doesnt make up that 75-100W difference.

Nice chatting...but I'm done.

 

[londiste]

Most people rely on tools that don't do what they think they do ... For example GPU-z, an application I have reverence for does not actually measure the full power draw on AMD cards

Irrelevant here but this is mostly because AMD cards report a GPU-only consumption while Nvidia cards report the consumption for the entire card.

So the approach I follow is to look at the power at the wall and then, using the efficiency appropriate for the % load, calculate back to the DC wattage being supplied to the system. Then, one can look at respected tests reviews for the measured voltage variations and electrical noise at the % load in the ttest tabes in the review. If ya had this PSU and you were outputting 420 watts or so, based upon the results you see starting at the link below, id be a very happy camper.

Been there, done that. Nvidia cards match the reported power consumption to a very reasonable error margin. So do AMD cards with some leeway given to the non-GPU consumption on the card (mainly VRAM). At this point I would not hesitate saying monitoring tools report a pretty accurate power consumption number for GPUs.

I have noticed that memory overclock ( plus 1350 Mhz on afterburner) when running benchmarks like superposition and 3DMark can really increase the power/voltage load. With a 750 watt EVGA PSU I continuously get VRel Perf cap in GPU_Z, but when I use the Seasonic 1000 Watt Gold PSU I do not. I am under the impression that you need a power supply that can provide "clean" power to your entire system in order to get the best reliability/stability out of a GPU. Some social media "experts" even suggesting powering a GPU with a seperate PSU when doing extreme OC's to get a reliable power delivery.

VRel is one of the trickier limit reasons. It is possible that gets triggered by lack of power or sometimes the "cleanliness" of power but usually it should just mean the voltage limit has been hit. Nvidia cards are voltage limited and for Turing cards that limit should be 1.068V.

If your specs page is accurate and this is the Colorful IGame RTX 2080 TI Kudan limited edition, I can imagine it ignores the voltage limit a bit but it is still there albeit a smidge higher.
422.9W at 132.1% TDP is believable as well, 320W TDP is not unusual for the higher end or special edition cards to have.
The 400+W by itself is not completely unexpected - Nvidia GPUs start to consume large amounts of power just above the normal frequency/voltage.

What load are these numbers from and what frequencies are you running during that load?

 

[BakGat]

Well OP here
Why I was inquiring was because, I wanted to work out if I needed a good psu for my system, what type of wattage would I be looking at in order to ensure a clean power delivery to the GPU. I thought that power delivery efficiency was curved based, and how increased voltage on mem oc increased the wattage the card was drawing over and above the GPU-Z figure.

And yes, it is a special edition card..it weighs 2.5 kG and comes with a AIO cooler. It is packed with heat pipes. I have spent a reasonable amount of money, and want to make sure that the most expensive part of the build is not hobbled by a power supply that does not allow it to be pushed to its limits.