Wait, so does the title of the Webcast imply the next gen won't be out until 2014?
AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
Wait, so does the title of the Webcast imply the next gen won't be out until 2014?
hoom
Veteran
I thought part of the point was also that making a new chip on a mature process you already built several on, you know the tricks & can build a relatively big die expecting solid yield.
You also have opportunity to clean up the design: correct mistakes, remove some unneeded redundancy, fix some clock &/or power issues.
Its what catapulted ATI into mainstream with 9700Pro & got them back competitive with RV770 (was that 4870?)
I'm all good for a high efficiency, good price/performance relatively low power chip in the NZ$650-700 range.
I don't give a damn if it isn't as fast as a NZ$1,000 more expensive chip.
You also have opportunity to clean up the design: correct mistakes, remove some unneeded redundancy, fix some clock &/or power issues.
Its what catapulted ATI into mainstream with 9700Pro & got them back competitive with RV770 (was that 4870?)
I'm all good for a high efficiency, good price/performance relatively low power chip in the NZ$650-700 range.
I don't give a damn if it isn't as fast as a NZ$1,000 more expensive chip.
RV770 was something of an outlier, since R600's physical implementation left something to be desired in terms of density. That was something blamed on tools not being ready.
There's been no complaint about where the current GPUs sit in this regard.
There's been no complaint about where the current GPUs sit in this regard.
Then again, Tahiti's density does leave something to be desired compared to Pitcairn, Bonaire, etc.
Whether that's because of improvements to the tools, the use of the HPL process instead of HP, a different mixture of logic and I/O, or a combination of the above, I don't know.
I think they all use the 28HP process so far. Isn't TSMC claiming higher performance (or lower power, or both to some extent) for HPM compared to HP? If it isn't that much more expensive than HP, could it be an option?
Damien thinks every GCN chip is built on HPL, except for Tahiti, which uses HP: http://www.hardware.fr/news/13356/gpu-hawaii-20-plus-petit-que-gk110.html
He's usually careful about making such statements, so I think he can be trusted on this one, but I'm not really basing this on anything else.
The large high-speed DDR interface is a very big feature in the die photo. A big chunk of that deficit goes away if that's not counted.
If that would be true, I would expect some more significant differences in the clock vs voltage behaviour between Tahiti and the rest of the line. After all, the nominal voltage for HPL is 0.15V higher than for HP and offers a lower transistor speed.
Per 128b chunk the Pitcairn PHY is about half the size and Tahiti has 3 of them.
That much smaller, eh? Is it just because it's slower?
http://www.anandtech.com/show/7342/amd-teases-next-enthusiast-gpu-still-28nm-23-smaller-than-gk110
hopefully the performance is there. I might wait another generation for the 20nm stuff.
hopefully the performance is there. I might wait another generation for the 20nm stuff.
I wish i had that luxury, i already sold my 6850; I need to see something good(apart from Hawaii) otherwise i will by a cheap card to hold out for a year.
I hear you. There isn't much I want this year. I figure if EQ3 is next year I can easily hold off till then. Mabye buy a rift plus 20nm chip next year.
Bouncing Zabaglione Bros.
Legend
28nm and 20nm are at different points of their life-cycles, and so you're not really comparing like-for-like.
You're talking about a well understood and proven 28nm process against a brand new 20nm that is barely ready for risk production (unless you're Intel).
Right now, AMD would have to play it safe with a protoype 20nm product, whereas they can push 28nm to the limit.
so a 512bit "Pitcairn-Style" @ 1200MHz interface would be smaller (2/3 the area) than a 384bit "Tahiti-style" interface @ 1500MHz....
but 512bit @ 1200MHz > 384bit @ 1500MHz
so why did you choose the 384bit interface for Tahiti?
Fewer memory chips and an easier layout of the package and the board save costs.
====================================
edit:
I don't know if you can say anything about it, but one could get the impression that the 6+GBps PHYs of Tahiti are a bit overengineered or (to put it maybe better) set up relatively conservatively. One can probably understand that (high frequency interface with the first chip on a new process), but if we draw the comparison to nVidia (which tended to trail a bit in the memory interface area), their 7GBps PHYs appear to be not larger or even smaller than the ones of Tahiti. Is that an area we may see an improvement in the next generation? Or got Bonaire already a new version (I think there is no die shot around to see it)?
Last edited by a moderator:
What does that even mean?