AMD Cayman info (or rumor)

[-Boris-]

TDP for 580gtx is 350w.
cayman likely is around 200w.
its 150w difference.
300w difference for 6990, now, that is why amd went value and small mm/W design.
they beat nvidia with efficiency and offers better price/performance ratio and now for 3 straight designs they just hammer nvidia.

amd sells with high profit, if Nvidia didnt have their professional market they be gone like 3dfx.

It might consume 350W. But it's rated TDP is far lower.

[highoctane]

TDP for 580gtx is 350w.

Thats not the tdp of the 580, show us where the 580 comes close to 350w during in game play.

Why make up the numbers when you know people on these forums might have a clue.

http://www.techpowerup.com/reviews/N...TX_580/25.html

[Heinz68]

Read that again, he said on par with the 58xx series, NOT 5xx series... He's comparing the launches between the 69xx and the 58xx from AMD, not performance.

True, but the 58** series launch can hardly be compared, since the HD 58** series did not have any competition for about half year and even longer with the mainstream card. Actually even the AMD flagship card is still the fastest Graphic Card on the market now.

I think he was maimly promoting good supply for launch day and competitive prices.

Good supply is good news since some people in this thread were predicting paper launch.

I think if the HD 6970 acording to rumors with about 20-25% smaller die can perform anywhere close to GTX 580 than the HD 6970 is very competitive.

The rumored 190 TDP on HD 6970 should also be AMD advantage when designing the HD 6990, I hope to buy one in january.

[Borden]

http://bbs.expreview.com/forum-26-1.html

[kaktus1907]

http://forum.beyond3d.com/showpost.p...postcount=5906

[Chickenfeed]

Well that info adds credibility to the whole not 1920sp thing. Now the question remains how aggresively AMD will price the 6970 in regards to the 570. IF they launch between 400-450, they should be a great card.

Now roughly eye balling those comparisons, on average we are looking at a 20% advantage for 6970 over the 570, which should place it in 580 territory in a few scenarios no doubt. It looks like they have a product which *should* use less power and hopefully make less noise than Nvidias most comparable product at a higher performance level for the money.


PS: Those graphs are evil.

[FischOderAal]

Well, I guess this shows that AMD is pitching HD 6970 against GTX 570 and not GTX 580. (If the slides are for realz, of course.)

[Dark-Energy]

Nvidia Confidential? Can't wait till this card is released. I'm sick of all these conflicting rumors.

Ontopic though, I really think AMD has something huge waiting for us other than performance. With the conflicting TDP and stuff it really makes me think that perhaps AMD has a switch (the rumor has been popping up everywhere) that can switch between perhaps 190 TDP and 240 TDP? Maybe the 190W is almost as fast as the 580, and the 240W is faster? Also, I don't get the shader count too. How come it's 1536 SP? I mean seriously? Or maybe it can switch between 1536 to 1920 too? That would be sweet.

I don't know what I am talking about.

[tajoh111]

i don't know why people expectations have changed from unrealistic performance to now unrealistic power usage. The card can be just a competitive card with good performance near, at or beating a gtx 580 with a relatively high power usage(still not as high as the gtx 580) which is completely believable and I think will happen. The 5870 has a power usage of 212 watts at peak in furmark and it's significantly lower in gaming, but if we are going to go by that method, so does the gtx 580. If we test during game sessions, the gtx 580 consumes around 225 watts as wizzards test show, while the 5870 consumes 144 watts. If by gaming tests, power consumptions goes up to 200 or 190 watts, this is completely believable for the 6970. However they don't look nearly as awe-inspiring while the gtx 580 is at 225. It's better power efficiency than the gtx 580, but not super incredible.

The only way your going to get close to that 150 watt number flopper(and you still won't reach it) is if you compare real world gaming scenarios running on the 6970, vs a gtx 580 running furmark which is hardly a fair comparison.

In a peak tests like furmark, there is no way the 6970 is staying under 200 watts. You guys have to remember the 2gb of slower ddr5 vs 1gb of slower ddr5 added 30+ watts to the power consumption of the 5870.

http://www.tweaktown.com/reviews/324...d/index17.html

The 6970's memory is significantly faster so it should add more than 30watts+. Lastly one of the most telling tales is the size of the chip and the frequency. This chip is larger than the 5870 and clocked higher. Ultimate if the 6970 has more efficient shader usage, power consumption should be going up considerably.

The reason why the 6870 is more efficient size wise compared to the 5870 and they have similar performance is because the 6870 is actually using its shaders, while a lot of the 5870 shaders are just idle. Hence when you run furmark which uses all the power a graphics card, the power consumption on the 5870 jumps from 144 to 212w or a 48% jump compared to the 6870 which jumps from 127 watts to 163 watts or a 28% jump.

They actually have similar performance per watt in gaming scenario's, furmark just shows how how efficiently those shaders are being used in games when you compare the two. The larger the difference between the two, the less efficient the shaders are being used.

If the 6970 has high performance, it's power usage should go up considerably if it is an efficient architecture because that performance is coming from shaders doing work rather than just idling. The cost of performance is energy and the 6970 is not all of a sudden going to outperform the 5870 in games by 30 or 40% and consume less power.

I noticed the r300 tag disappeared..

[zerazax]

Well that info adds credibility to the whole not 1920sp thing. Now the question remains how aggresively AMD will price the 6970 in regards to the 570. IF they launch between 400-450, they should be a great card.

Now roughly eye balling those comparisons, on average we are looking at a 20% advantage for 6970 over the 570, which should place it in 580 territory in a few scenarios no doubt. It looks like they have a product which *should* use less power and hopefully make less noise than Nvidias most comparable product at a higher performance level for the money.


PS: Those graphs are evil.

Well supposedly the 6970 consumes < 200 W, and might even be 6pin + 6 pin, in which case it might consume even less than the GTX 570

Interestingly enough, AMD has had a tendency to lowball their own card comparisons in terms of positioning - they had the 6850 against the GTX 460 768mb and the 6870 against the GTX 460 1GB when real world comparisons of reference models is more of 6850 against the 1GB and the 6870 against the GTX 470. And of course, there was the 4850 against the 8800GT and 4870 against the 9800GTX

And yeah, if that performance slide is accurate, eyeballing it, the 6970 is probably nipping on the heals of the 580

[hurleybird]

If 6970 performs worse than GTX 580 I will be seriously disapointed, and AMD just lost a sale.

[zerazax]

i don't know why people expectations have changed from unrealistic performance to now unrealistic power usage. The card can be just a competitive card with good performance near, at or beating a gtx 580 with a relatively high power usage(still not as high as the gtx 580) which is completely believable and I think will happen. The 5870 has a power usage of 212 watts at peak in furmark and it's significantly lower in gaming, but if we are going to go by that method, so does the gtx 580. If we test during game sessions, the gtx 580 consumes around 225 watts as wizzards test show, while the 5870 consumes 144 watts. If by gaming tests, power consumptions goes up to 200 or 190 watts, this is completely believable for the 6970. However they don't look nearly as awe-inspiring while the gtx 580 is at 225. It's better power efficiency than the gtx 580, but not super incredible.

Because that's where the latest rumors from credible sources are pointing its power consumption at - credible people aren't just making it up out of nowhere at this point, with all the leaks. They're posted in b3d and ocuk and guru3d and some other places now. Whether that's rated vs. real in game remains to be seen. And 144 vs. 225 is a big deal because the 5870 is rated at 188W (so the Furmark ~200W peak isn't out of the range) and Nvidia's 580 at 240W - but Nvidia rates their GPUs differently than AMD (I don't want to rehash that argument, but it was explained quite well in this thread itself) - if the 6970 is RATED at 190W, then its actual power consumption will be quite lower - heck, even if its rated at 225W, using AMD's terms, its a significant drop

The only way your going to get close to that 150 watt number flopper(and you still won't reach it) is if you compare real world gaming scenarios running on the 6970, vs a gtx 580 running furmark which is hardly a fair comparison.

In a peak tests like furmark, there is no way the 6970 is staying under 200 watts. You guys have to remember the 2gb of slower ddr5 vs 1gb of slower ddr5 added 30+ watts to the power consumption of the 5870.

http://www.tweaktown.com/reviews/324...d/index17.html

The 6970's memory is significantly faster so it should add more than 30watts+. Lastly one of the most telling tales is the size of the chip and the frequency. This chip is larger than the 5870 and clocked higher. Ultimate if the 6970 has more efficient shader usage, power consumption should be going up considerably.

Yes and no - we don't know what they did in terms of streamlining power. Keep in mind Barts managed to get 90% of the performance of the 5800's with even less power draw - its hard to believe they didn't change some things over to Cayman - not to mention the whole power regulation thing they're incorporating. It's not going to be a flat 30W+ more

Also, the GDDR5 definitely will consume more power going from 1GB -> 2GB faster stuff, then again, better binned GDDR5 isn't going to automatically scale linearly either

The reason why the 6870 is more efficient size wise compared to the 5870 and they have similar performance is because the 6870 is actually using its shaders, while a lot of the 5870 shaders are just idle. Hence when you run furmark which uses all the power a graphics card, the power consumption on the 5870 jumps from 144 to 212w or a 48% jump compared to the 6870 which jumps from 127 watts to 163 watts or a 28% jump.

They actually have similar performance per watt in gaming scenario's, furmark just shows how how efficiently those shaders are being used in games when you compare the two. The larger the difference between the two, the less efficient the shaders are being used.

If the 6970 has high performance, it's power usage should go up considerably if it is an efficient architecture because that performance is coming from shaders doing work rather than just idling. The cost of performance is energy and the 6970 is not all of a sudden going to outperform the 5870 in games by 30 or 40% and consume less power.

No one is making (well, I guess flopper is) that the 6970 is going to consume less power - it is, however, possibly rated at the same power.

Going with the same numbers you're using, if 6970 really does get rated at 190W like some are claiming, we might expect a similar Furmark peak (lets go with your 212W) but its more efficient, so using Barts efficiency #, we get 165W in game for Cayman. Lets say 170W.

Still significantly lower than the 225W or so the 580 pulls in game

And that's assuming Barts even fixed utilization of its shaders in the same way you're thinking - AFAIK, they're still using the same 5D as Cypress, just the scheduler and depth of each row was changed. Not to mention we have no clue how the change to 4-VLIW changes power from 5 etc.

edit: Taking a look at the sources, they say 190W in game, rated at 250W. Much closer to the 225W or so for the 580 (although we'd have to have w1z do a comparison so we're using the same equipment/methodology to be more accurate), so it remains to be seen what performance level its at - the same guy who said 190W said it more or less matches the 580, just a few % points under on the overall average

If 6970 performs worse than GTX 580 I will be seriously disapointed, and AMD just lost a sale.

Can't really say AMD is responsible for hyped expectations when they haven't said a peep

[hurrdurr]

From AMD's slides it looks like HD6970 will be closer to GTX 570 than GTX 580.

From "Nvidia confidential" it looks like HD6970 will be on par with GTX 580.

The world must be coming to an end...

[zerazax]

From AMD's slides it looks like HD6970 will be closer to GTX 570 than GTX 580.

From "Nvidia confidential" it looks like HD6970 will be on par with GTX 580.

The world must be coming to an end...

Apparently Nvidia confidential are slides from either an internal Nvidia benchmark or one done by one of the AIBs that dose Nvidia AND AMD

As mentioned though, AMD has had a history of placing their cards in their slides against lower competition. The 6850 vs. the 460 768mb, the 6870 vs. the 460 1GB when reference vs. reference, the 6870 is ~the 470 for example. And of course, 4850 vs. 8800GT and 4870 vs. 9800GTX wasn't even close to reality

Interesting thing I realized is that the 190W might also be from AMD's pdf of slides: they supposedly did the idle/typical/max breakdown as something like <30W/190W/250W. Going off Barts, 190W->250W is similar percentage-wise

That said, I wonder if AMD underrated/overrated power - which is up to the reviewers

[tajoh111]

The gtx 580 is not close to being a champion of power efficiency, so beating it should be no problem. AMD earns bragging rights if it signficantly beats its prior generation since it was actually good with power consumption. E.g barts xt uses 127 watts of power, if they increase speed by 50% and power consumption is 190 watts, they simply matched barts efficiency and have not surpassed it.

If the gtx 580 consumes 225 watts in gaming scenario's and the 6970 consumes 180 or 190 watts in gaming scenario's, which I can see happening, it's will be better efficiency than gf110 for sure but no better than barts(barts is great anyways at this as is cypress so no shame in this).

If the rated tdp of Cayman xt from AMD is 190 watts vs Cypress 188, I have a feeling AMD has become just more like Nvidia in its rating, because the memory addition alone should make it more than 2 watts greater than cypress. I think it has to increase more because increased efficiency, means less wasted shader because they are used more often, which translate into higher power consumption, which translate into higher performance. I have a feeling this is what is letting AMD increase the size of the chip only 20% but get more than 20% performance.

Basically what I am saying is there is no way AMD has increased performance over cypress 30-40%, has added 2gb of ddr3. That would have to mean Cayman xt consumes 30-40 watts less than cypress(which is incredibly efficient in this respect already) and performs 30-40% better. This chip is bigger and clocked higher. I am almost certain that this is impossible.

Zerazex, I believe 190watts could be the typical usage. This is completely in the realm of believability but I still think it will be a tiny bit higher.

[zerazax]

The gtx 580 is not close to being a champion of power efficiency, so beating it should be no problem. AMD earns bragging rights if it signficantly beats its prior generation since it was actually good with power consumption. E.g barts xt uses 127 watts of power, if they increase speed by 50% and power consumption is 190 watts, they simply matched barts efficiency and have not surpassed it.

If the gtx 580 consumes 225 watts in gaming scenario's and the 6970 consumes 180 or 190 watts in gaming scenario's, which I can see happening, it's will be better efficiency than gf110 for sure but no better than barts(barts is great anyways at this as is cypress so no shame in this).

If the rated tdp of Cayman xt from AMD is 190 watts vs Cypress 188, I have a feeling AMD has become just more like Nvidia in its rating, because the memory addition alone should make it more than 2 watts greater than cypress. I think it has to increase more because increased efficiency, means less wasted shader because they are used more often, which translate into higher power consumption, which translate into higher performance. I have a feeling this is what is letting AMD increase the size of the chip only 20% but get more than 20% performance.

Basically what I am saying is there is no way AMD has increased performance over cypress 30-40%, has added 2gb of ddr3. That would have to mean Cayman xt consumes 30-40 watts less than cypress(which is incredibly efficient in this respect already) and performs 30-40% better. This chip is bigger and clocked higher. I am almost certain that this is impossible.

Right, I'm pretty sure the 190W typical/250W max rating rumored is most likely - keeping Barts efficiency should be a priority anyways since that'd be silly to do that for Barts and not for its big brother. Of course, if they did that, the 2GB factor still must be considered... in which case, it's even more efficient

[Chrono Detector]

http://bbs.expreview.com/forum-26-1.html

If this slide is true then thats very disappointing.

[tajoh111]

Right, I'm pretty sure the 190W typical/250W max rating rumored is most likely - keeping Barts efficiency should be a priority anyways since that'd be silly to do that for Barts and not for its big brother. Of course, if they did that, the 2GB factor still must be considered... in which case, it's even more efficient

That depends, at least in wizzards case, the 5870 and 6870 consume their respective power amounts in 3dmark. How typical is 3dmark?

Something tells me those charts are fake. The Nvidia confidential things is weird combined when the slide has a title of an AMD marketing slide. They even have the trademark logo in the radeon name. Its just seems so wrong in so many ways because Nvidia wouldn't make such a slide in the first place and as a result, why would they need to put a Nvidia confidentiality or NDA on it. There really should be no NDA on such a slide unless it was an AMD one.

[..::G80::..]

WTF? No 1920SP??? ATI please make HD 6980 with 1920SP.

[zalbard]

Wow, have you seen Unigine Heaven? If that's true then this is some awesome tessellation performance boost coming from 5870.

WTF? No 1920SP??? ATI please make HD 6980 with 1920SP.

Wouldn't that be awesome?!

[hurrdurr]

You guys "being disappointed" with the product placement slide of AMD: I can't understand you, a month before there was the Analyst Day and there was another product placement slide which placed HD6970 significantly below HD5970... nobody took care of it then, but now people don't want to believe this.

[zerazax]

If the 28nm process really got pushed back to mid 2012 by TSMC, I can see a ~450mm^2 1920SP version of this architecture on 40nm as a single high end GPU

I guess AMD is fortunate that they still have a lot of room to get bigger

That depends, at least in wizzards case, the 5870 and 6870 consume their respective power amounts in 3dmark. How typical is 3dmark?

Probably above typical, since synthetics are good at stressing the card, but not at Furmark levels

If this slide is true then thats very disappointing.

What if that's where they're priced

[eric66]

[zalbard]

Why would anybody buy 6950 if "6950 = 5870"? Clearly 5870 is going to be much cheaper...
Also, interesting die size figures. Cayman has ~370mm^2 die size, GTX580 has ~530mm^2. I wouldn't say that Cayman is a chip with a big die, then.
And 1920sp for 6970 is back, heh...
Fecking rumours!

[eric66]

Wow, have you seen Unigine Heaven? If that's true then this is some awesome tessellation performance boost coming from 5870.


Wouldn't that be awesome?!

lol even nvidia admitted that 6970 is faster than 580 in unigine so i guess thats less talking about unigine for nvidia from now on