AMD: Sea Islands R1100 (8*** series) Speculation/ Rumour Thread

I thought as much, but is there some kind of source? What about the other ones? Is Volcanic Islands supposed to be a mix of 2014 GPUs, then and not just a certain architecture?

Dave said so about Sea Islands, probably in this thread somewhere.

There's nothing official about any of the other ones, but Volcanic Islands seems to be the next architecture, which I suppose we might call GCN 2.0.

I'm not sure there's such a thing as Central Islands or Canary Islands in AMD's pipeline.
 
That's a sooner than expected. Not bad. Nvidia had better come up with a real successor to the 600 series quickly.

AMD's graphics division is going to do really well in Q3 and Q4, relatively speaking. That Apple Mac Pro win was quite big. I don't know what segment consoles will fall under, but their IP is still making it into a larger percentage of products. Then they'll have a "real" refreshed GPU lineup, and not just a small silicon/metal layer revision with boosted clocks and a bios update.

Not that I think Nvidia made a bad move; they did exactly what they should have done, but they need a follow-up.
 
Wouldn't AMD be gunning for a 1H 20nm series as well? Or perhaps the "not pursuing new processes aggressively" thing is still in effect.
 

Just for the record, posted by AnarchX yesterday too ...
http://forum.beyond3d.com/showpost.php?p=1761468&postcount=78

But anyway, i dont know, we was waiting a release around october november ( maybe more november, december ), and the rest of the story is not really suprising and dont bring much info.. not called HD8000 (with OEM 8000 cards, this seems obvious for dont mix arch and name ), could be called HD9000 (obviously ), and for the rest i think many have some hope AMD will maybe do a special move with the anniversary of chips like the 9500's and 9700's.

If AMD will intend to increase their gamer market, they will need something special here. But lets wait and see.
 
Last edited by a moderator:
Well that would be an interesting turn of events. Nvidia's usually launches products on new nodes later than AMD, at least for the last two nodes.
 
Wouldn't AMD be gunning for a 1H 20nm series as well? Or perhaps the "not pursuing new processes aggressively" thing is still in effect.

Well this rumor mentions October 2012, and I doubt AMD would make a brand new architecture on 28nm, but who knows?
 
Still the possibility for AMD to release the next series on 2 times:

- 9550 like, gamers enthusiast parts on 28nm.. just here for take the lead on Nvidia and so follow the direction they have speak so much about lately ( taking back share on gaming market )

- followed by an professional, computing, high end gaming parts on 20nm, later in Q1-Q2 2014. ( let say a "9700Pro like" )

The situation is special, but its not like they have not do it by the past when the situation was special too: x1800xt PE 512 who was need some modification and fix and who have end something like 3 months then by the release of the X1900 series. Bart released 3 months before Cayman and who was not really based on the same NI arch. ( again due to a modification of their plan for the 32nm)

And i really doubt with all the computing features describes on their roadmap for next chips they could release a "pure" Gaming killer card and a computing aimed card on the same chip. And its not so much different of what have done Nvidia with GK104 and 110. In addition, on a strategy point of view, this will be certainly excellent looking at the release timing of actual Nvidia parts ( 700 series ) and future maxwell parts.
 
Last edited by a moderator:
20nm doesn't make sense to me -- seems too early. I know TSMC accelerated their ramp, but October...?

This is how it is in my eyes:

HD 6000 : HD 5000 :: HD 9000 : HD 7000

Minimal performance gains on the same node, as a result of larger dies and architecture tweaks. AMD may be calling the new architecture GCN 2.0, but I seriously doubt there will be any major changes.

It will likely be GCN 1.0 with some minor to moderate efficiency improvments (I think ROPs could use some work, particularly with blending? Nvidia seemingly has better blend rate/bandwidth), substantially higher memory bus frequencies with a good yield (at least 6.0 -- I'm not counting the 7970 GE) and a top-to-bottom, more robust turbo boost.

That right there would bring substantial performance improvements for most of the lineup, without even increasing unit count. Then, of course, they could add more units on top of the aformentioned changes, which would make for some very enticing products.

Hopefully AMD can catch up to Nvidia on power efficiency and perf/mm2 on the high end. I don't recall how the smaller chips fare on the cost front, but they do pretty well on the power side of things. Add the better theoretical compute performance, and AMD should have a real winner on their hands. That is, until Maxwell launches.
 
I just hope that for the next series AMD allows downsampling, it's a joke that they do not offer it and have(perhaps by accident) ruined the 3rd party program that people used to force downsampling on AMD hardware.
 
Been saying it for almost a year that AMD's next series would still be on 28nm. There's no way TSMC's 20nm is anywhere near ready by October.

Nvidia are likely to do a Kepler with Maxwell, ie release the "midrange" part first and hope to charge high end prices. With Titan being $1k it shouldn't be too hard to charge $500+ for a slightly slower GM104. It'll have a much smaller die, however that is only going to offset the much higher early node cost.

If we assume the 9970 (or whatever name it gets, I wonder if AMD is about to change the Radeon brand as well?) can match or just beat the 780 then they don't have much reason to hurry to 20nm in H1 2014. A large 28nm die will be just as economical as a medium-dized 20nm die for a very long time, and TSMC's 20nm is nothing spectacular in terms of speed or power improvement - http://www.tsmc.com/english/dedicatedFoundry/technology/20nm.htm (notice the or in the link - the improvement from 28nm to 20nm is going to be nothing like as impressive as 40nm to 28nm was) -
 
Last edited by a moderator:
Sea Islands is just marketing, it's not linked to a specific graphics IP level.

I disagree.
The leaked (and retired) ISA published on AMD site was reporting Sea Islands ISA inside as title.
"Sea Islands Series Instruction Set Architecture"
 
...so you make a full ISA using that name, leak the document on internet, then retire it, as a marketing technique.
Hard to believe, to me. Much easier, actually, that the ISA was made available too much before due.
 
Volcanic islands launching this year?. And if so, with improved architecture or the same as Tahiti just with more units ( CUs, ROPs, geometry engines... )?.
 
No idea what the source is ..

http://www.guru3d.com/news_story/amd_radeon_hd_9000_series_might_arrive_in_october.html
More and rumors are circulating that AMD may introduce its Radeon HD 9000 as early as October. The company is anticipated to skip the 8000-series as it already sells Radeon HD 8000-branded parts to OEMs. The first Radeon HD 9000 family parts are anticipated to be Curacao and Hainan, two 28nm chips based on the GCN 2.0 architecture. The new architecture reportedly has an improved front-end with 4 asynchronous computing engines [ACEs] and 3 geometry engines, as well as a higher number of stream processors.

Since at present AMD sells re-badged Radeon HD 7000 products as Radeon HD 8000-series products to OEMs, it is logical for the company to utilize Radeon HD 9000 sequence for its new line of graphics cards. ATI Technologies’ original Radeon 9000-series products based on code-named R300-family graphics processing units revolutionized the market of graphics cards a decade ago and helped ATI (which now belongs to AMD) to become the producer of market leading graphics solutions.

The Curacao XT graphics processor is expected to feature 2304 stream processors (36 compute units), 144 texture units, 48 render back ends and 384-bit memory controller. The Hainan is projected to have 1792 stream processors (28 compute units), 112 texture units, 32 render back ends and 256-bit memory controller. Both chips will share the same front-end (just-like current-gen Radeon HD 7900 and 7800 do) with 4 asynchronous computing engines [ACEs], 3 geometry engines, command processor, global data share and so on.
 
Back
Top