AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

43% isn't boring but I tend to fixate on moments like R580/G71 to G80. For all we know these companies are developing new ground-breaking architectures that will do that again someday and make current hardware look much less cool than it currently appears.

Cypress to Cayman was ~30%? 6970 started at $370 though. GCN has been a rip off and still is not so well priced. But I can't blame AMD for returning to R420-R580 pricing and seeing if people will bite.
 
Current gen pricing is the result of a few things that both AMD and Nvidia are keen to avoid and to be frank it's here to stay, or going to get worse.

The next gen won't look so overpriced at launch because the current gen cards have held their prices very well. By this stage of a cycle outgoing AMD cards would generally be dirt cheap and the incoming new cards would look too expensive in comparison. That's another side benefit of the gaming bundle - it continues to give the perception of the AMD cards holding their price which is good (for them) in a lot of ways.

Anyway, I wouldn't hold back on buying a card now in the expectation of faster+cheaper to come in the near future.
 
That's true, I got my 7950 for slightly cheaper than they are available today, but the bundled games were nowhere near as good.

You can easily get a 7950 for $300 and a 7970 for $400 here. Those are pretty decent prices for what you get in my mind, much more so considering the game bundles.
 
Not happening as it will be 28nm again, also i believe the next node is 20nm(TSMC).

Late 2014/early 2015 will be the 20nm cards aka Maxwell/800 series and Volcanic Islands/9000 series.

Huh? 20nm is scheduled for risk production late 2013 and mass in early 2014, late 2014 is already 16nm FinFETs

If AMD and/or NVIDIA follows what AMD did with HD7900, it should be able to come for sale around new year 2013/2014, just like 7900 came around new year 2011/2012
 
look like the reason, both AMD and Nvidia have decided to skip the new series to end of 2013 -2014 no ? ( understand, 28nm cant provide enough impreovement for a new series ? )

With nvidia having a bit more luck by having the GK110 ready for tesla and they could do something with that.
 
look like the reason, both AMD and Nvidia have decided to skip the new series to end of 2013 -2014 no ? ( understand, 28nm cant provide enough impreovement for a new series ? )

With nvidia having a bit more luck by having the GK110 ready for tesla and they could do something with that.

NVIDIA's plans are unclear, but as far as AMD is concerned, that's evidently not the case. They've already released Oland, which is admittedly based on GCN 1.0, but also Bonaire based on GCN 1.1, and now they're about to launch Saturn and Neptune:

http://wccftech.com/amd-radeon-hd-8...s-gpu-performance-unveiled-msi-gx70-notebook/
 
Ignoring pricing... G80 seemed so much faster in part because it was a lot bigger chip than R580/G71.


GK110, GF100, and GT200 have been relatively large. But of course they are after the phantom GPGPU money maker now so that's where the transistor budget goes.
 
Not happening as it will be 28nm again, also i believe the next node is 20nm(TSMC).

Late 2014/early 2015 will be the 20nm cards aka Maxwell/800 series and Volcanic Islands/9000 series.
Imho AMD should not need to jump on a significantly revamped architecture and a new process to put more pressure on Nvidia. AMD still has a performance/mm^2 advantage on Nvidia and it is still unwilling to go with more "exotic" bus width outside of their high end cards.

For me it is a mistake, they could have a line of "bartized" products that Nvidia would have a tougher time competing against:
Instead of the HD7790 (labelled H8xxx by OEM) they should have introduced two cards, the HD 8770 &50: 196 bit bus, 1.5 GB of RAM, 12/14 CU
They should also introduce something in between Pitcairn and Tahiti, based on 320 bit bus and name it HD 88xx (26-22 CU, 32 ROps).

In both that card may still be tinier (even though slightly) than the Nvidia card they are competing with (same applies for the bus width as I expect Nvidia to somehow introduce salvage GTX 690 at some point with a bus width somewhere in between 256 and 512)

The high end would not get refreshed till the next node.
 
Last edited by a moderator:
Huh? 20nm is scheduled for risk production late 2013 and mass in early 2014, late 2014 is already 16nm FinFETs

If AMD and/or NVIDIA follows what AMD did with HD7900, it should be able to come for sale around new year 2013/2014, just like 7900 came around new year 2011/2012

Mass production starting in early 2014 usually means actual products in late 2014.
 
ugh another year ? man wtf is going on with the pc industry. I feel like its crawling to a complete stop.

That's the reality of it. Moore's law is slowing, soon to stop. Each node is getting harder and more expensive than the last. Most programming doesn't even need more processing power. Software needs to catch up and find something interesting to do with more power, otherwise, why do you need more power? New resolutions and tri-monitor set-ups were added because they had excess power that was essentially useless until someone does something new that actually uses it. New consoles might do something about this, but we'll see, might be a general overall trend.
 
Hasn't AMD said pretty clearly that in the future they wont be as likely to jump into new nodes as quickly? it's getting pretty hard to fight for the first wafers against companies like Apple and Qualcomn.
 
Yup, that's why I would expect this round NV first with new pipe cleaners...

Moore's law is slowing, soon to stop

Actually it is not a law and I doubt it will stop. If it stops all the companies will go sleeping forever.

Will we see in our lifetimes a 30-inch Retina display with life-like graphics, for example on Crysis N?
 
Hasn't AMD said pretty clearly that in the future they wont be as likely to jump into new nodes as quickly? it's getting pretty hard to fight for the first wafers against companies like Apple and Qualcomn.

In reality, todays, Apple is still use the FABs of Samsung ( The giant Austin Texas FAB ( who is surely the biggest fab never made) , and Qualcomm dont use anymore so much TSMC and use too Samsung FAB now.
( Its unclear for Qualcomm, if all the production have been moved outside TSMC or not )

Back to Topic, isnt some source was claim AMD have tapped out all the new architectures allready ? and just delay the mass production ?
 
Last edited by a moderator:
Back
Top