How to determine needed PSU wattage

[Wheelzz]

It does. http://www.techpowerup.com/vgabios/127271/zotac-gtx690-2048-120421.html

How does that even work? Is the 690 not two downclocked 760s?

[Faa]

How does that even work? Is the 690 not two downclocked 760s?

Not at all just 2x unlocked GK104's aka 680. Theyre optimized as I've said, if you look at the mars 760 that card pulls so much more power than a 690. http://www.guru3d.com/articles_pages/asus_geforce_gtx_760_mars_review,7.html

[Wheelzz]

Not at all just 2x unlocked GK104's aka 680. Theyre optimized as I've said, if you look at the mars 760 that card pulls so much more power than a 690. http://www.guru3d.com/articles_pages/asus_geforce_gtx_760_mars_review,7.html

So why are all other cards optimized so terribly? I don't get where they get off making a single card put out much higher watts than a card that is essentially two lower clocked versions of the same card on one PCB. 

[Faa]

So why are all other cards optimized so terribly? I don't get where they get off making a single card put out much higher watts than a card that is essentially two lower clocked versions of the same card on one PCB. 

Because they come with beasty coolers? A 690's cooler is nothing against them so they have to tweak it or you get a 7990 scenario. 

[Wheelzz]

Because they come with beasty coolers? A 690's cooler is nothing against them so they have to tweak it or you get a 7990 scenario. 

but if a single 770 is using over 200w and a card that has two 680s is (which are fairly similar to the a 770 performance wise) use around 40w less while pumping out way more performance. I just don't get it. If they were downclocked that much, the performance difference would be little.

[Faa]

but if a single 770 is using over 200w and a card that has two 680s is (which are fairly similar to the a 770 performance wise) use around 40w less while pumping out way more performance. I just don't get it. If they were downclocked that much, the performance difference would be little.

Some chips are just more efficient, requiring less vcore and power etc depends on chip so it's pretty much a binning process.

[Wheelzz]

Some chips are just more efficient, requiring less vcore and power etc depends on chip so it's pretty much a binning process.

But the power target is less on the 690, I don't get it...how can two of the same GPU run at less power consumption than one? Regardless of downclock. The GTX 770 came out later so if anything it should be better.

 

Also, if the board power limit is 185W, how is their card using 213W at max load?

 

Please note I am not trying to argue, learning is the goal. You clearly know more than me.

[Faa]

But the power target is less on the 690, I don't get it...how can two of the same GPU run at less power consumption than one? Regardless of downclock. The GTX 770 came out later so if anything it should be better.

 

Also, if the board power limit is 185W, how is their card using 213W at max load?

 

Please note I am not trying to argue, learning is the goal. You clearly know more than me.

I've answered that question in my previous post. To give you an idea I had a 680 DC2T before that was by default boosting to 1254MHz and did 1300MHz quite easily, the TDP just never went above 80% with the stock power target settings which was 225W at 100% thats 180W. Lots of other cards were sitting in the 250W range or even more with those clocks. I got your point now yes it doesnt sit at 170W maybe it's just per GPU who knows but still it uses much less power than 2 680's in sli.

[MG2R]

But the power target is less on the 690, I don't get it...how can two of the same GPU run at less power consumption than one? Regardless of downclock. The GTX 770 came out later so if anything it should be better.

Allow me to chimme in here.

 

The reason it's more efficient is because P = C * f * V². This formula says that the power an IC consumes equals to a constant © mutliplied by the frequency multiplied by the squared supplied voltage.

 

Say you have a chip that consumes runs at 1 Hz and need 10V to run at that 1 Hz. Say C=1 for easy calculation. This IC consumes 100W. Now if you take two of these chips and run them in parallel, you get twice the performance (assuming the workload is able to run completely in parallel). If we want the same performance as one chip at full power, we can half the frequency of both chip. Both work half as fast as one, so combined they're just as powerful. The power draw, if we only change the frequency will also half, so combine they use just as much power as one chip at full power. However, IC's generally need less voltage to stay stable when we lower the frequency. Say we half the voltage on each chip as well (how much we can lower the voltage, is dependant on the architecture, amongst other things). Suddenly, each IC only uses an eight of it's original power. That means that the total combined power raw of both chips is only a quarter of one chip running at full speed (25 W = 2*(0.5 Hz * 5²V)).

 

It's a bit abstract, but the above example shows how much more efficient a two-core setup can be.

 

PS: I have no idea about the actual power requirements of any of the GPU's discussed above, this post is only to show that power consumption can be drastically lower on two-core/chip cards.

[s3ns3]

but if a single 770 is using over 200w

a gtx 770 uses 180W . the 690 272W 

[Wheelzz]

I've answered that question in my previous post. To give you an idea I had a 680 DC2T before that was by default boosting to 1254MHz and did 1300MHz quite easily, the TDP just never went above 80% with the stock power target settings which was 225W at 100% thats 180W. Lots of other cards were sitting in the 250W range or even more with those clocks. I got your point now yes it doesnt sit at 170W maybe it's just per GPU who knows but still it uses much less power than 2 680's in sli.

Yeah I think there was a miss-communication there, but I think we're on same page now. As this TPU Review shows, the 690 4GB seems to use about 274W at game load, the 680 2GB 186W, and the SLI 680s about 360w - which makes a lot more sense. I am awfully surprised at how little the performance difference is for the wattage difference, I guess slight downclocks reduce power usage a lot more than performance? Or is there other major tweaks they do to harness the power at a reduced power? I assume this is all done to stop the boards from frying themselves?

 

That figure for card power usage makes a lot more sense than 170w when you consider that the Hexus review picture you linked showed the system using 456w at game load with the single 690.

 

What is TDP actually? Is it how much power the card requires? 

 

Allow me to chimme in here.

 

The reason it's more efficient is because P = C * f * V². This formula says that the power an IC consumes equals to a constant © mutliplied by the frequency multiplied by the squared supplied voltage.

 

Say you have a chip that consumes runs at 1 Hz and need 10V to run at that 1 Hz. Say C=1 for easy calculation. This IC consumes 100W. Now if you take two of these chips and run them in parallel, you get twice the performance (assuming the workload is able to run completely in parallel). If we want the same performance as one chip at full power, we can half the frequency of both chip. Both work half as fast as one, so combined they're just as powerful. The power draw, if we only change the frequency will also half, so combine they use just as much power as one chip at full power. However, IC's generally need less voltage to stay stable when we lower the frequency. Say we half the voltage on each chip as well (how much we can lower the voltage, is dependant on the architecture, amongst other things). Suddenly, each IC only uses an eight of it's original power. That means that the total combined power raw of both chips is only a quarter of one chip running at full speed (25 W = 2*(0.5 Hz * 5²V)).

 

It's a bit abstract, but the above example shows how much more efficient a two-core setup can be.

 

PS: I have no idea about the actual power requirements of any of the GPU's discussed above, this post is only to show that power consumption can be drastically lower on two-core/chip cards.

So reducing the frequency reduces the power usage a lot more than the performance? Hence how you can have the 680 SLI use much more power than a 680 SLI setup while only being a little bit slower?

[MG2R]

What is TDP actually? Is it how much power the card requires?

TDP, or Thermal Design Point, is a directive given by a chip vendor that indicates to system or board builders what their cooling system should be capable of dissipating continuously. Read more here.

 

 

So reducing the frequency reduces the power usage a lot more than the performance? Hence how you can have the 680 SLI use much more power than a 680 SLI setup while only being a little bit slower?

It's reducing the VOLTAGE that drastically reduces the power draw. It's the same when overclocking: it's only when you start upping the voltage, that your chip starts to put out insane amounts of heat.

 

I'm probably understanding the highlighted part wrong, but 680 SLI setup is the same as running two chips at full speed in parallel. SLI doesn't suddenly lower the voltage/frequency of the cards.

[Faa]

Yeah I think there was a miss-communication there, but I think we're on same page now. As this TPU Review shows, the 690 4GB seems to use about 274W at game load, the 680 2GB 186W, and the SLI 680s about 360w - which makes a lot more sense. I am awfully surprised at how little the performance difference is for the wattage difference, I guess slight downclocks reduce power usage a lot more than performance? Or is there other major tweaks they do to harness the power at a reduced power? I assume this is all done to stop the boards from frying themselves?

 

That figure for card power usage makes a lot more sense than 170w when you consider that the Hexus review picture you linked showed the system using 456w at game load with the single 690.

 

What is TDP actually? Is it how much power the card requires? 

 

So reducing the frequency reduces the power usage a lot more than the performance? Hence how you can have the 680 SLI use much more power than a 680 SLI setup while only being a little bit slower?

 

 

People use TDP term wrong it's not the max power; It's just a cooling spec that's all. Using that formula just gives me a 14% difference between a 680 & 1 core from 690 with a 50MHz difference, scales nearly perfectly in the wattage results.

He just means 1000MHz will be twice as fast as 500MHz so -> 1x680@1000MHz performance & total power consumption will be equal to 2 680's@500MHz. 

[Wheelzz]

People use TDP term wrong it's not the max power; It's just a cooling spec that's all. Using that formula just gives me a 14% difference between a 680 & 1 core from 690 with a 50MHz difference, scales nearly perfectly in the wattage results.

He just means 1000MHz will be twice as fast as 500MHz so -> 1x680@1000MHz performance & total power consumption will be equal to 2 680's@500MHz. 

TDP is usually given in watts right? How does this relate to cooling?

 

TDP, or Thermal Design Point, is a directive given by a chip vendor that indicates to system or board builders what their cooling system should be capable of dissipating continuously. Read more here.

 

 

It's reducing the VOLTAGE that drastically reduces the power draw. It's the same when overclocking: it's only when you start upping the voltage, that your chip starts to put out insane amounts of heat.

 

I'm probably understanding the highlighted part wrong, but 680 SLI setup is the same as running two chips at full speed in parallel. SLI doesn't suddenly lower the voltage/frequency of the cards.

So the 690 runs at so much lower power usage because they have the chips running with a lot less voltage? Is there also I bit of downclock? I assume so.

[MG2R]

TDP is usually given in watts right? How does this relate to cooling?

Heat transfer is measured in the amount of energy per second that is transfered. Energy (Joule) per second = Watt. Seriously, check out the link I gave you.

 

 

So the 690 runs at so much lower power usage because they have the chips running with a lot less voltage? Is there also I bit of downclock? I assume so.

As I stated before: I don't know any specific details from the top of my head. But yes, the 690 is basically just two downclocked 680's on one PCB. The downclock probably allows for a small downvolt as well.

 

Again, I'm not a GPU expert, I just know a thing or two about integrated circuit design.

[Faa]

TDP is usually given in watts right? How does this relate to cooling?

 

Intel defines TDP as follows: The upper point of the thermal profile consists of the Thermal Design 

Power (TDP) and the associated Tcase value. Thermal Design Power (TDP) should be used for 

processor thermal solution design targets. TDP is not the maximum power that the processor can 

dissipate. TDP is measured at maximum TCASE.1

Tcase is a temp. It's not measured with stresstesters that push the CPU to its max power but with normal applications, theres a minimum of heat disappation in watts required. A cooler with 40W TDP will cause a 95W TDP cpu to overheat you get the point.