[Dave Baumann]

Originally Posted by lanek

On reference card, or do you mean for non reference card ?

The GHz edition ASIC is explicitly designed to fit within the same thermal / electrical / physical footprint as the standard 7970 and as such, given that every one of the partners have transistioned to their own designs they will be using their own fansink designs. The "reference" design is not much more than our qualification mule now and is sampled for the purposes of performance testing.

[Alexko]

Originally Posted by UniversalTruth

http://www.techpowerup.com/reviews/A...Edition/4.html

Well that explains why it's such a small improvement, then. But it also raises the question: why didn't AMD respin it when they knew they could reduce leakage significantly?

Did they not want to divert resources away from Sea Islands? Or overshadow it with a super-clocked Tahiti?

[Dave Baumann]

Originally Posted by lanek

And Powertune Boost:

Note - fom Cat 12.7 Beta DTE will apply not just to boards with Boost, but generally to PowerTune.

Originally Posted by Alexko

But it also raises the question: why didn't AMD respin it when they knew they could reduce leakage significantly?

Sorry?

[Man from Atlantis]

Originally Posted by Alexko

Well that explains why it's such a small improvement, then. But it also raises the question: why didn't AMD respin it when they knew they could reduce leakage significantly?

Did they not want to divert resources away from Sea Islands? Or overshadow it with a super-clocked Tahiti?

They could but it would still have low yields and low amount of wafers till end of the year.. when things start to be prettier, there will be Sea Islands at the end anyway..

[UniversalTruth]

Originally Posted by Man from Atlantis

They could but it would still have low yields and low amount of wafers till end of the year.. when things start to be prettier, there will be Sea Islands at the end anyway..

It could have low yields in a case of unrealistically high targets- like 1150 MHz at relatively low power.
It was said Tahiti didn't need any further revisions. Why- it is beyond me.

[Gipsel]

Originally Posted by Dave Baumann

Sorry?

Or to phrase it differently: How do you (or AMD) know a respin would significantly reduce leakage? Is it realistic at all, assuming AMD didn't mess up the implementation in the first place? And it would require a full (base layer) respin anyway, a metal layer spin would not going to help.

[lanek]

Originally Posted by Dave Baumann

Note - fom Cat 12.7 Beta DTE will apply not just to boards with Boost, but generally to PowerTune.

Nice to know, as i use it allready.


Anyway what impact can have watercooling on DTE ? (for example both 7970 i use never reach 40°C full load in the worst case) .

Originally Posted by Dave Baumann

The GHz edition ASIC is explicitly designed to fit within the same thermal / electrical / physical footprint as the standard 7970 and as such, given that every one of the partners have transistioned to their own designs they will be using their own fansink designs. The "reference" design is not much more than our qualification mule now and is sampled for the purposes of performance testing.

Ok i see thanks, this is what i was thinking.

[Alexko]

Originally Posted by Dave Baumann

Note - fom Cat 12.7 Beta DTE will apply not just to boards with Boost, but generally to PowerTune.


Sorry?

Isn't that what you (or perhaps someone else from AMD) said? That Pitcairn and Cape Verde were built using feedback from Tahiti, and as a result had lower leakage? Proportionally, of course.

You even dropped a few hints that this knowledge might be fed back into a respin of Tahiti, which, if I understand correctly, was not done.

@Gipsel: yes, it would have had to be a full respin. In fact when you compare the 7970 GHz Edition (GE) with the 7870, you find that even though:

— they work at almost the same clocks,
— the 7870 is, in most respect, about 60~100% of the 7970 GE (CUs, die size, bus width, die size, front end, ROPs),
— the 7870 is a relatively higher-volume part, presumably binned somewhat less strictly,

in practice the 7870 only draws about 55% of the 7970 GE's power: http://www.hardware.fr/articles/869-...nces-watt.html
So there does seem to be something inherently better about Pitcairn, power-wise.

[Dave Baumann]

Not quite. Tahiti's parameters were set on earlier, pre-production wafers. Tahiti was the first GPU in 28nm production and subsequent GPU's benefited from this as that creates a lot of learning activites and changes. Tahiti XT2 is reseting the Tahiti parameters according to what the characteristics of the actual shipping products (effectively, bringing it back in line with where Verde and Pitcairn were set).

[lanek]

Anyway i have a question, but i dont think you can respond to it..

The voltage in some review is really high ( 1.256mv ), why so high? i can imagine even 1.2mv should be largely enough. all 7970 i have seen end stable at 1050mhz at their default voltage ( 1112-1175mv ( with fluctuation with digital pwm ), let say it look a big marge over it. Even when Boost is kick in.

[Alexko]

Originally Posted by Dave Baumann

Not quite. Tahiti's parameters were set on earlier, pre-production wafers. Tahiti was the first GPU in 28nm production and subsequent GPU's benefited from this as that creates a lot of learning activites and changes. Tahiti XT2 is reseting the Tahiti parameters according to what the characteristics of the actual shipping products (effectively, bringing it back in line with where Verde and Pitcairn were set).

I see, thanks. So Pitcairn's apparent proportionately lower power draw is due to something else? Is it just normal variability giving a false impression, or something specific to Tahiti increasing power draw? (ECC, DP, etc.)

[Kaotik]

Originally Posted by Dave Baumann

Not quite. Tahiti's parameters were set on earlier, pre-production wafers. Tahiti was the first GPU in 28nm production and subsequent GPU's benefited from this as that creates a lot of learning activites and changes. Tahiti XT2 is reseting the Tahiti parameters according to what the characteristics of the actual shipping products (effectively, bringing it back in line with where Verde and Pitcairn were set).

The "Tahiti XT2" is still chip-wise exact same as "Tahiti", or is it a new metal layer spin or did it actually get some silicon changes?

Can old 7970's get the Boost-feature, assuming the chip can take GHz Edition clocks, by simple BIOS-flash to GHz Edition BIOS?

[UniversalTruth]

Originally Posted by Dave Baumann

The GHz edition ASIC is explicitly designed to fit within the same thermal / electrical / physical footprint as the standard 7970.

I don't think the word "designed" is the most appropriate one. It simply doesn't fit in the whole sentence.
Why didn't you design all of the chips within wafers such?

[lanek]

Originally Posted by Jaaanosik

Do the reviewers mention this? They have charts with noise and guess what cooler is in those charts.

Some reviewers mention it too ..
http://techreport.com/articles.x/23150/1

lets be honest, AIB dont do 3-4 versions of the same cards just for the fun... this is the only way to differentiate them, and so to get competition..hopefully they dont just put a stickers on the cooler.
In general it take sometimes before release of the custom versions.. They are allready ready this time, so why release cards with reference cooler ? You will maybe see them in their site, but not in the shops.

[xDxD]

about 40 more watts in full and an acquisition cost greater than 7970, for performances that often do not exceed the gtx680 ... definitely fail.

http://www.hardocp.com/article/2012/..._card_review/1

[Dave Baumann]

Originally Posted by UniversalTruth

I don't think the word "designed" is the most appropriate one. It simply doesn't fit in the whole sentence.
Why didn't you design all of the chips within wafers such?

From a concept-to-implementation point of view that is accurate. The overall program was designed to limit partner churn and transition as much as possible.

[lanek]

removed, dont need to enter the discussion.

[xDxD]

Originally Posted by lanek

Seriously you will vote on 3 games ?

3 of the best on market for pc, actually? I think yes

obviously i don't watch only the performance itself, but also because of higher consumption resulting.

Look:

[lanek]

i dont understand this tdp too ... the voltage are really high, and i dont see why ... both 7970 i own need not even touch the stock voltage for be stable at this clock speed. ( so i have got 3 7970 on my hand so far, 2x 1112mv and 1 1175mv ) ... Ofc stability for 24/24 games for me, is not maybe stable and safe ( across board ) in term of qualification at an engineer level.

+ Power consumption differ so much from a game to another... make me think to some remark from some reviewer: they need test more games for tdp for each card if they have the time.. as it differ completely from a game to an other.

[ECH]

Pretty good results
CB review

[no-X]

Originally Posted by xDxD

about 40 more watts in full and an acquisition cost greater than 7970, for performances that often do not exceed the gtx680 ... definitely fail.

http://www.hardocp.com/article/2012/..._card_review/1

Well, there are also reviews, which show quite the opposite:

Originally Posted by TechReport

The Radeon HD 7970 GHz Edition has indeed recaptured the single-GPU performance title for AMD; it's even faster than Zotac's GTX 680 AMP! Edition. And at $499.99, the 7970 GHz Edition is unambiguously a better value than the stock-clocked GeForce GTX 680. Everything seems to be going AMD's way—even our power consumption results turned out to be closer than expected.

According to ComputerBase.de, GE (compared to GTX 680) drains 14 % more power while offering 15-20 % more performance with MSAA 8× enabled. Hardware.fr results show 10-12 % better performance than GTX 680.

Performance difference in many tests is very similar to HD 6970 - GTX 580 performance delta. The later one consumed ~50 watts more and nobody called it a fail.

[Alexko]

Originally Posted by no-X

Well, there are also reviews, which show quite the opposite:

According to ComputerBase.de, GE (compared to GTX 680) drains 14 % more power while offering 15-20 % more performance with MSAA 8× enabled. Hardware.fr results show 10-12 % better performance than GTX 680.

Performance difference in many tests is very similar to HD 6970 - GTX 580 performance delta. The later one consumed ~50 watts more and nobody called it a fail.

TechReport tested power consumption with Batman, which is much faster on Kepler. I suspect the latter reaches higher utilization in this game, thereby closing the power gap.

Still, the new 7970 is at the same 250W TDP as the old one, so it's probable that there were equally power hungry 7970s floating around already, just at 925MHz and with accordingly lower performance.

[hurleybird]

Originally Posted by Dave Baumann

Not sure why anyone cares about the reference cooler since it won't be used.

In that case, instead of dropping the ball on the cooler you dropped the ball on communication with reviewers.

If people would have understood that the reference cooler wasn't going to be used, nobody would care about the noise measurements. Because this wasn't communicated, people made reasonable assumptions. You have shipped similarly loud products in the past after all.

Hate to say it, but Nvidia wouldn't have made this mistake.

[Homeles]

Originally Posted by Alexko

TechReport tested power consumption with Batman, which is much faster on Kepler. I suspect the latter reaches higher utilization in this game, thereby closing the power gap.

Still, the new 7970 is at the same 250W TDP as the old one, so it's probable that there were equally power hungry 7970s floating around already, just at 925MHz and with accordingly lower performance.

Don't forget, TDP isn't the same as power consumption. They're related, but a 250W TDP doesn't mean the card draws 250W.

[Dave Baumann]

Originally Posted by hurleybird

In that case, instead of dropping the ball on the cooler you dropped the ball on communication with reviewers.

If people would have understood that the reference cooler wasn't going to be used, nobody would care about the noise measurements. Because this wasn't communicated, people made reasonable assumptions. You have shipped similarly loud products in the past after all.

Hate to say it, but Nvidia wouldn't have made this mistake.

It was communicated, hence why you saw many note it.