AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

With 25% more shaders and 20% more clocks over the R7 370, the 370X will have a big performance boost! But even assuming that AMD doesn't need to boost voltage to get to that high clock rate, it's going to use ~40% more wattage than the 370, putting it at ~155 Watts compared to GTX 950's ~90 Watts.
The GPU-Z screenshot is probably from a factory overclocked model (e.g. a rebrand of http://www.powercolor.com/au/products_features.asp?id=490).
There are already factory overclocked 1150mhz R7 370s and 1180mhz R9 270Xs. The stock clock jump is probably more conservative.
 
I wonder why AMD keeps making so many ASICs at TSMC when they have so much trouble fulfilling their wafer supply agreement with GloFo.
They probably don't have the money and/or engineering resources to spare to port any of their existing designs to GloFo's process.
 
They probably don't have the money and/or engineering resources to spare to port any of their existing designs to GloFo's process.

I suppose that's possible, but Kaveri and Carrizo are manufactured by GloFo, so GCN is already ported to GloFo's process, with the possible exception of some minor blocks that may be absent from APUs—the memory controllers in those, for example, are probably slightly different.
 
For AMD or overall? The overall record is held by G92, at least if you count the 65nm and 55nm versions as a single product.

The G92 was introduced in October 2007 with the 8800GT 512MB and by October 2010 nVidia's whole line-up had been replaced by Fermi cards (mobile included). That's about 3 years of service.
Pitcairn was introduced in March 2012 and 3.5 years later, in August 2015, there are still new cards being introduced with it.

Pitcairn has already surpassed G92's record and it'll probably do so for an entire year or more. And it didn't go through a die shrink.

It's interesting that Trinidad XT (i.e., Pitcairn) can easily clock this high compared to Fiji. It may have to do with the small micro-architectural differences, but it seems more likely to be related to design differences, and maybe even a process difference.

1180MHz is just a 11.8% overclock compared to the 3.5 year-old 7870 GHz Edition.
Pitcairn is about 1/3rd the size of Fiji. You expected both GPUs to clock the same?

Interesting, such a quick update, but this is where the money is.

And you know someone's been doing a terrible job when a 3.5 year-old product fills a "where the money is" category.
 
The G92 was introduced in October 2007 with the 8800GT 512MB and by October 2010 nVidia's whole line-up had been replaced by Fermi cards (mobile included). That's about 3 years of service.
Pitcairn was introduced in March 2012 and 3.5 years later, in August 2015, there are still new cards being introduced with it.

Pitcairn has already surpassed G92's record and it'll probably do so for an entire year or more. And it didn't go through a die shrink.

You'll probably find that Cape Verde puts up similar longevity (Caicos also - intro'd Jan. 2011 (HD 64x0M) and Feb. 2011 (HD 6450), EoL'd May 2015 with the arrival of Oland) - started life with the HD 7750 (Feb. 2012), and continues with the newly launched R9 M370X.
 
They probably don't have the money and/or engineering resources to spare to port any of their existing designs to GloFo's process.

Perhaps there isn't a good business case for a lateral node move at the tail end of a generation just to build the same product.
That there are segments that have stagnated this way may point out that refreshing them is where the money isn't.
 
For AMD or overall? The overall record is held by G92, at least if you count the 65nm and 55nm versions as a single product.

Overall. I wouldn't put G92 in the same boat. Has anything changed at all from the original Pitcairn silicon?

Not that there's anything necessarily wrong with that. It's obviously still doing well as it makes way down the pricing ladder.
 
To my knowledge, no AMD GPUs are made at GloFo. I don't have an Antigua, but since it's just Tonga, it would be TSMC.
I'll refer you to public/SEC filings:

http://www.amd.com/en-us/press-releases/Pages/amd-amends-wafer-2014apr1.aspx

These purchases contemplate AMD’s current PC market expectations and the manufacturing of certain Graphics Processor Units (GPUs) and semi-custom game console products at GLOBALFOUNDRIES in 2014

I suppose that's possible, but Kaveri and Carrizo are manufactured by GloFo, so GCN is already ported to GloFo's process, with the possible exception of some minor blocks that may be absent from APUs—the memory controllers in those, for example, are probably slightly different.

Also note the Carrizo materials pointing to using a more "GPU oriented" process path for increased density.

http://www.anandtech.com/show/9319/...leap-of-efficiency-and-architecture-updates/3
 
Last edited:
You'll probably find that Cape Verde puts up similar longevity (Caicos also - intro'd Jan. 2011 (HD 64x0M) and Feb. 2011 (HD 6450), EoL'd May 2015 with the arrival of Oland) - started life with the HD 7750 (Feb. 2012), and continues with the newly launched R9 M370X.

m370x is not cape verde, it's at least GCN2.
 
Perhaps there isn't a good business case for a lateral node move at the tail end of a generation just to build the same product.
You might well be very right of course, I'm not knowledgeable about such things, but if you have contract obligations there might be penalties associated with not fulfilling them... Maybe smaller than the cost of moving product to GloFo tho, what do I know? :)

Also, if GF wants future AMD contracts, maybe it's not good business sense penalising them...? *shrug* :LOL:
 
Back
Top