AMD Cayman info (or rumor)

[Calmatory]

Also I believe Techpowerup had a review (of GTX 480 or 470?) that compared two exactly same clocked Fermi's, one with better custom cooling and one with stock cooling. The one with better custom cooling (and temperatures) consumed a lot less power, in spite of the fact that the two GPU's were identical.

If they compared two different cards (in attempt to figure out about the power consumption) to each other and the only difference was the cooling solution, that was a wrong way to go. The right way would've been to test both cooling solutions in both cards and measure. Twice. Most probably most of the difference was due to better bin GPU and better bin VRAM. But sure temperature does play a role in power consumption of a chip, there are many other factors too, which vary from card to card.

It would be quite interesting to see accurate measurements though.

[hurrdurr]

If they compared two different cards (in attempt to figure out about the power consumption) to each other and the only difference was the cooling solution, that was a wrong way to go. The right way would've been to test both cooling solutions in both cards and measure. Twice. Most probably most of the difference was due to better bin GPU and better bin VRAM. But sure temperature does play a role in power consumption of a chip, there are many other factors too, which vary from card to card.

It would be quite interesting to see accurate measurements though.

I don't think so. After seeing that the overclocked, custom cooled 480 (which was Zotac's version I believe) actually consumed less power than the stock 480, the guys did another test: They changed the card's temperature (I think by controlling the fan speed) and recorded the temperatures. Results were clear, higher temperature meant more power drawn; and that was the source of the original differences in the temperatures of the stock and Zotac cards.

[Manicdan]

If they compared two different cards (in attempt to figure out about the power consumption) to each other and the only difference was the cooling solution, that was a wrong way to go. The right way would've been to test both cooling solutions in both cards and measure. Twice. Most probably most of the difference was due to better bin GPU and better bin VRAM. But sure temperature does play a role in power consumption of a chip, there are many other factors too, which vary from card to card.

It would be quite interesting to see accurate measurements though.

he did that. he took the card with better temps and better power draw and simply turned down the fan speed and watched it go right back up. he then tested it every few C and got us a curve. about every 1.5C reduced power by 1W

it shouldnt be too hard to simply search all of W1ssard's posts in here and look for a fermi or 480 thread

[Sadasius]

That would make sense since heat is the leakage and would need more power to compensate. The higher the temps the more leakage will happen and cause you to use more power for the same purpose or job. I think I made sense...

[LightSpeed]

he did that. he took the card with better temps and better power draw and simply turned down the fan speed and watched it go right back up. he then tested it every few C and got us a curve. about every 1.5C reduced power by 1W

it shouldnt be too hard to simply search all of W1ssard's posts in here and look for a fermi or 480 thread

The power draw/temperature relation is quite substantial due to the excessive power usage of the GTX480, but yeah power consumption definitely does affect power draw.

[Manicdan]

found it:

there is the data, linear fit is y = 1.1975x + 215.29

basically for every °C that the card runs hotter it needs 1.2W more power to handle the exact same load

[Iconyu]

Yeah the chip was big enough that each centigrade changed the internal resistance enough to register on a watt meter. That's seriously impressive, it gives you a idea of how huge Fermi is. 1 watt/C is pretty shocking.

[Jowy Atreides]

2 connectors. Rumour says it's Antilles.

[Xoulz]

that looks different.. enlarged photos anywhere?

[Dimitriman]

that looks too small for antilles given thats a zalman vf3000 cooler there meant for 5870 and should be pin compatible with 6970

[GoldenTiger]

2 connectors. Rumour says it's Antilles.

It's not. http://www.xtremesystems.org/forums/...postcount=1013

[Dimitriman]

It's not. http://www.xtremesystems.org/forums/...postcount=1013

thats a killer 6850!
and o correct myself cayman is not confirmed compatible with 5870 coolers. rather 6870 is.

[spursindonesia]

Antilles is a dual Cayman card, PERIOD. How can AMD achieve that, what kinda crippling that needs to be done to do that ? I think they know better than the rest of us. We can speculate, but Antilles as a dual Cayman card is a FACT known from HD 68xx series launch presentation slide.

IIRC, GTX 275 is a 215 w TDP card, while GTX 295 (dual GTX 275 card) is a 296 w TDP card, so it CAN be done, and i can see this kinda analogy in work for Antilles (speed & voltage binning, reduced clock & voltage).

[zerazax]

Antilles is a dual Cayman card, PERIOD. How can AMD achieve that, what kinda crippling that needs to be done to do that ? I think they know better than the rest of us. We can speculate, but Antilles as a dual Cayman card is a FACT known from HD 68xx series launch presentation slide.

IIRC, GTX 275 is a 215 w TDP card, while GTX 295 (dual GTX 275 card) is a 296 w TDP card, so it CAN be done, and i can see this kinda analogy in work for Antilles (speed & voltage binning, reduced clock & voltage).

As mentioned.. TDP != power consumption, particularly with AMD's ratings

[Chumbucket843]

Yeah the chip was big enough that each centigrade changed the internal resistance enough to register on a watt meter. That's seriously impressive, it gives you a idea of how huge Fermi is. 1 watt/C is pretty shocking.

temperature dependence of transistor properties is more complex than that. many different properties change with temperature from leakage, to velocity saturation, and different voltages may reverse these effects.

[Dimitriman]

dont take ratings for granted. amd said 150 watts and >150 watts for 6850 and 6870 but reality is around 120w/145w respectively.

so if they say 300w for antilles im more inclined to believe 270w real draw and 225w for cayman xt

[hurrdurr]

As mentioned.. TDP != power consumption, particularly with AMD's ratings

Still AMD has strived to keep its own definition of TDP, whatever that is, under 300W. If Cayman is 255W, the amount of cutting down AMD has to do, if it still wants to stay under 300W, is insane.

[spursindonesia]

Still AMD has strived to keep its own definition of TDP, whatever that is, under 300W. If Cayman is 255W, the amount of cutting down AMD has to do, if it still wants to stay under 300W, is insane.

Well, i don't think Cayman XT's TDP will be far north of 200 w level (210-220 w seems probable). Considering the suggested architectural change compared to Bart & die size compared to Cypress, i certainly hope that AMD will successfully pull a better performance/watt in this new flagship, highend single chip of them.

Bart XT needs to be clocked so high to be rated @151 w TDP, yet against HD 5850 that's rated the same, the actual power usage is slightly smaller. If Cayman actually delivers a new, more efficient architecture compared to what was applied in the current products, can be clocked lower & given lower voltage, i think they can achieve that, and make Antilles a worthwhile card valuewise (not just king of e peen, faster but not good enough to warrant its price tag).

[Sn0wm@n]

Still AMD has strived to keep its own definition of TDP, whatever that is, under 300W. If Cayman is 255W, the amount of cutting down AMD has to do, if it still wants to stay under 300W, is insane.

its pci express compliant so they can put pci express stickers on the boxes .. and it leaves non refference design headroom for oc version ...

or oc it yourself and break the 300watt tdp that amd had to obey

[hurrdurr]

its pci express compliant so they can put pci express stickers on the boxes .. and it leaves non refference design headroom for oc version ...

or oc it yourself and break the 300watt tdp that amd had to obey

AMD cannot release a severely cut down card that will barely be faster than its single GPU version (if 6970's really 255W).

[Sn0wm@n]

AMD cannot release a severely cut down card that will barely be faster than its single GPU version (if 6970's really 255W).

i know it wouldnt make sense .. but even if its only 40% faster then a 6970 its still a good card ... and it might leave us with oc refference or non refference design ...


ill wait after the nda ends to see if its a worthwile purchase or should i get 2 6850

[hurrdurr]

HD5870's TDP was 180W, 5970's was 300W. And yet 5970 was only 40% percent faster than 5870 on average.

If the case is 255W vs 300W, even with improved CF scaling I don't think a 40% figure is possible.

[Sn0wm@n]

HD5870's TDP was 180W, 5970's was 300W. And yet 5970 was only 40% percent faster than 5870 on average.

If the case is 255W vs 300W, even with improved CF scaling I don't think a 40% figure is possible.

they said they had some sort of buffer chip on the boards and other stuff so it might be possible to have such performance .. but i guess we will have to wait for proper reviews

[zerazax]

If its 255W is a big question. The slides only said > 225W, which can mean anything from 226W up to 300W of course.

That said, the 6850's and 6870s are getting a lot of scaling in CF, so who knows if some hardware changes were made to allow for this. If Cayman is a superset of Barts' features, then the scaling may be there, even with downclocked/cut-down cores.

That said, it seems like downclocking might be the biggest difference in power draw. For instance, they can set the 6970 to be clocked very high at stock, say 900 or 950 MHz, and put Antilles at 725 with the same overall core. It depends greatly on how much they want the 6970 to perform as a single GPU. They might be pulling out all stops to claim the single GPU crown, and so Cayman XT might be far higher clocked than it might otherwise need be.

[kuroikenshi]

If its 255W is a big question. The slides only said > 225W, which can mean anything from 226W up to 300W of course.

That said, the 6850's and 6870s are getting a lot of scaling in CF, so who knows if some hardware changes were made to allow for this. If Cayman is a superset of Barts' features, then the scaling may be there, even with downclocked/cut-down cores.

That said, it seems like downclocking might be the biggest difference in power draw. For instance, they can set the 6970 to be clocked very high at stock, say 900 or 950 MHz, and put Antilles at 725 with the same overall core. It depends greatly on how much they want the 6970 to perform as a single GPU. They might be pulling out all stops to claim the single GPU crown, and so Cayman XT might be far higher clocked than it might otherwise need be.

that's exactly what they did with 5970, down-clock 2 5870 cores to meet tdp's