AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

I think a lot of people are forgetting that Cypress/RV870 was initially planned to be ~GT200b sized and the original design of Cayman on 40nm would have been +400mm2, 420-480mm2.
 
You or someone else is making stuff up.

http://www.anandtech.com/show/2937/3

planned870.png
 
That's the size it would have been on 40nm... which it was initially never planned to be produced on. With 32nm, clearly it would have been smaller. In which case it doesn't have much relation to Hawaii being stuck on 28nm other than to say that the die size is likely to only be a little bigger than Tahiti going off a past history of 1 event - which goes against the rumours.

EDIT: My bad, I forgot that RV870 was Cypress, not Cayman. Those old code names are hard to keep track of these days :p
 
There's no actual law saying AMD can't make big GPU's, and they can't allow themselves to be haunted by the ghost of R600 forever.

The timing right now is basically perfect for it. They will not get a better shot at beating Nvidia at their own game, and they won't get a better time either with BF4 coming out and the console prestige. This is one they really, really need to win.
 
There's no actual law saying AMD can't make big GPU's, and they can't allow themselves to be haunted by the ghost of R600 forever.

The timing right now is basically perfect for it. They will not get a better shot at beating Nvidia at their own game, and they won't get a better time either with BF4 coming out and the console prestige. This is one they really, really need to win.

yes. the increase in Hawaii die size is definitely going to be more than from HD 5870 (334 sq mm) to HD 6970(389 sq mm). After 21 months and a very mature TSMC node AMD can afford to go big die. With the impending console launch and BF4 being a AMD GE title this is the right time for AMD to launch a big die 470 - 480 sq mm flagship and take the GPU crown. The launch of next gen consoles with AMD tech and to have the PC GPU leadership would have a massive effect on AMD's brand value. Also AMD has high ambitions for the HPC space. A Firepro product which does well against a Quadro GK110 would mean improved market share and proft margins.

TSMC 20nm is going to start volume production in Q2 2014 with the first 20nm GPU products launching in late June or early July 2014. But its not going to be available in any serious volume till late Q3 2014. So with hawaii launching in Oct 2013, AMD can keep the GPU crown for 9 - 12 months. Also this allows AMD not to rush to 20nm. AMD can wait till initial yields are good and 20nm volume is significant to launch their 20nm flagship Pirate Islands GPU in Sep 2014. This way AMD would hit the crucial holiday season for both 2013 and 2014 with competitive products.
 
That account is (quite pronouncedly) incorrect. But it sounds neat and heroic and all, so that probably explains why the Internetz loves it.
Could you provide something to back that statement, Anandtech ain't some random site.
 
How so? What was wrong about that statement?
From my perspective neither Cypress nor Cayman was planned to be that large. Thanks to Jimbo's link I see where it came from. I thought this was one of those "common knowledge" things that didn't have a real source.

It turns out there are multiple viewpoints inside a company of what goes on and that article has one viewpoint. A lot of ideas were batted around and configurations do change over time, but by the time it got to design engineering it was not >= Pele in size.
 
From Digitimes:

AMD Hawaii-based graphics cards to mass ship in October

As AMD is set to announce its next-generation high-end GPU codenamed Hawaii, graphics card players including Asustek Computer, Micro-Star International (MSI) and PowerColor are expected to start mass shipping related products in October, according to sources from the upstream supply chain.

To welcome its new high-end products, AMD has started cutting prices for its Radeon HD 7000 series products including the HD 7990 recently.

As for Nvidia, since the GPU maker is seeing weaker-than-expected profits, the company is unlikely to launch a price cut immediately, but will observe Hawaii's performance before making a move, the sources noted.
 
I think there is going to be alot of overestimates on die size for Hawaii, just as everyone expected a much bigger Cayman in 2010. My bet for the chip would be around 400 sq mm - slightly larger than Cayman on 40nm. If they can hit titan like speeds with something only 10% larger than Tahiti, and retain all the compute goodies, they have done a good job.

On the flip side, nv probably has a full GK110 sku with 7.0Gbps memory ready to still claim fastest gpu when AMD launches Hawaii. This smells like GF110 vs Cayman all over again.

Completely agree and GK110 has a lot more in the tank. The Gk110 is completely capable of running at the same clocks as a 7970. Yet we see them being clocked at 900mhz.

There are already pre-overclocked cards in the 1050-1100 mhz range. It costed about 30 more watts to get there, but this is completely fair.

So take a fully enabled gk110 with ddr5(7.0ghz) which should add 10 percent already, while increase the clocks to 1100 with(a 26 percent increase in clocks vs titan(873)) which should add another 20 percent or so. So 30 percent faster in total than a gtx Titan. It would be a bit tough to do this with a blower fan configuration, but with an open air like AMD card makers has been doing with its ghz cards, it should be easy.

And at this point with GK110 manufacturing being stable, I could see Nvidia being able to make Titan like cards at gtx 780 volume. Some of the titan cards I have seen are using chips with a 60 percent asic which is piss. And I have seen gtx 780's with better asics in the mid 70s range.
 
Why would AMD do another Cayman? What would the point be? What would be achieved by another 15% faster, 15% bigger die?

Absolutely fuck all, is the answer. This part was (supposedly) cancelled over a year ago, they aren't doing it any more. If it was pointless then it's even more so now.
 
Last edited by a moderator:
Completely agree and GK110 has a lot more in the tank. The Gk110 is completely capable of running at the same clocks as a 7970. Yet we see them being clocked at 900mhz.

There are already pre-overclocked cards in the 1050-1100 mhz range. It costed about 30 more watts to get there, but this is completely fair.

So take a fully enabled gk110 with ddr5(7.0ghz) which should add 10 percent already, while increase the clocks to 1100 with(a 26 percent increase in clocks vs titan(873)) which should add another 20 percent or so. So 30 percent faster in total than a gtx Titan. It would be a bit tough to do this with a blower fan configuration, but with an open air like AMD card makers has been doing with its ghz cards, it should be easy.

And at this point with GK110 manufacturing being stable, I could see Nvidia being able to make Titan like cards at gtx 780 volume. Some of the titan cards I have seen are using chips with a 60 percent asic which is piss. And I have seen gtx 780's with better asics in the mid 70s range.

A full GK110 would have 15 SMXs over Titan's 14. That's a 7% increase, assuming ideal scaling.

And in practice, Titan frequently Boosts up to 1006MHz or a little above, so at 1100MHz it's a 10% additional increase. In total, such a card would be about 18% faster, tops.

GK110 has some more potential (albeit at a cost in power consumption, obviously) but not 30%.

The SOURCE(s) stated profits not revenues. Nvidia's GPU Profits, Margins and Revenues were all up Q-Q and Y-Y.

So the source(s) statement that Nvidia can't adjust prices because "profits were lower than expected" is complete nonsense.

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MTk2NzE2fENoaWxkSUQ9LTF8VHlwZT0z&t=1

NVIDIA's net income was $120 million in Q2'FY2013, and $96 million in Q2'FY2014 (their calendar is weird). So yes, profits were down. I don't really think it has much bearing on their pricing policy, however.

If a price cut makes financial sense, it makes financial sense. Profits from the previous quarter or a year ago shouldn't really influence that decision.
 
Why would AMD do another Cayman? What would the point be? What would be achieved by another 15% faster, 15% bigger die?

Absolutely fuck all, is the answer. This part was (supposedly) cancelled over a year ago, they aren't doing it any more. If it was pointless then it's even more so now.


We shall see. But those expecting a 500mm sq Radeon monster sku might be setting themselves up for disappointment.

With increased efficiency GCN 2.0 architecture, Pitcairn transistor density, who is to say they couldnt out perform GK110 with a more modest 400-420mm sq die....
 
A full GK110 would have 15 SMXs over Titan's 14. That's a 7% increase, assuming ideal scaling.

And in practice, Titan frequently Boosts up to 1006MHz or a little above, so at 1100MHz it's a 10% additional increase. In total, such a card would be about 18% faster, tops.

GK110 has some more potential (albeit at a cost in power consumption, obviously) but not 30%.



NVIDIA's net income was $120 million in Q2'FY2013, and $96 million in Q2'FY2014 (their calendar is weird). So yes, profits were down. I don't really think it has much bearing on their pricing policy, however.

If a price cut makes financial sense, it makes financial sense. Profits from the previous quarter or a year ago shouldn't really influence that decision.

I also mentioned the 7 ghz memory, which I where I am getting that last 3% from.

http://www.hardocp.com/article/2013/08/14/galaxy_geforce_gtx_780_hof_edition_review#.UhVc3D-etaY

This card is clocked 1100mhz from the factory but manages to be 20% faster than a standard gtx 780. Perhaps its boosting to more than 1100 mhz. But it is a 1100mhz gtx 780 that is 20 percent faster performance wise than a gtx 780. Also remember the gtx 780 has a slight clock advantage over titan.

This overclock wasn't even stretching the card by any means as they overclocked the card to 1300mhz.

This card at those clocks were almost 10 percent faster than titan as is already. So lets keep those same clocks, but add in 25 percent more shaders(2304->2880) and add ddr5 at 1750 mhz and you got a real shot at another 20 percent.
 
Back
Top