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

Companies don't tend to give employees any product they want though many including AMD have employee discounts or rebates.
 
As I ponder on the possibilities of the 8000 series I have to also wonder if it will mark the 1st card that will be capable of the same functionality as the next gen consoles. Which is rumored to be released in 2013. I'm not 100% confident that the 7000/600 series will be on the same tier once it's known what the next gen consoles can do.
 
Has CES ever been used to announce a new GPU? I can't imagine why a company would inflict such treatment on itself. That fact alone makes the rest suspicious. ;)

From that link:

According to data SweClockers aware of AMD will shortly announce new GPUs in the Radeon HD 8000M series, especially designed for laptops and other compact platforms.

It is not clear whether it is the existing architecture Graphics Core Next (GCN) or new GCN2.

Standalone cards from the Radeon HD 8000-series "Sea Islands" is not expected to appear until well into the new year, probably ahead in the spring or summer.

I guess the clue is in the 2nd quote.
 
Has CES ever been used to announce a new GPU? I can't imagine why a company would inflict such treatment on itself. That fact alone makes the rest suspicious. ;)
NV used CES 2010 to formally briefing the gaming press on GF100. Otherwise, no, new architectures have not been announced at CES in the last half-decade or so (as far as I can recall).

As for any potential AMD announcement, it's worth noting that AMD had a significant HD7000M (VLIW5 rebrand) presence at CES 2012.
 
NV used CES 2010 to formally briefing the gaming press on GF100. Otherwise, no, new architectures have not been announced at CES in the last half-decade or so (as far as I can recall).
Was that a press only private announcement? I guess it does make sense to do a briefing when everybody is there already.

As for any potential AMD announcement, it's worth noting that AMD had a significant HD7000M (VLIW5 rebrand) presence at CES 2012.
Ah, well. I'll have to be more specific and say 'major' announcement. ;)
 
I notice that there has been absolutely no word on GPU tape outs for either side.... did both companies get that much better at keeping that quiet? We should have heard about it before now. Maybe nobody's looking?

Hmm.

http://www.digitimes.com/news/a20120615PD210.html?mod=2

AMD confirms Sea Islands have been taped out, will enter production in Q4 and will be announced in Q1/13

http://forum.beyond3d.com/showpost.php?p=1651048&postcount=302

And after that many posts dedicated to the topic. ;)

I would bet it is GCN 2 as it will be the same architecture in Xbox Next and PS4.

I'd prefer to think it's just the Enhanced GCN, so not 2.0 ;)
 
Was that a press only private announcement? I guess it does make sense to do a briefing when everybody is there already.
Correct. They held it the day after CES since everyone was already there. Otherwise there's a strong disincentive to making major public announcements at CES since it's likely to get lost in the noise of the show.
 
"Solar System" HD8000M is out
12_17_amd82.jpg

12_17_amd81.jpg

20121217amd1.jpg
 
If the chip pictures are anywhere near accurate, then it suggests that there is a new GCN-based 28 nm chip with (≥) 384 SPs which succeeds the 40 nm Turks 7xxxM chips, since the chip for the 8500M-8700M series seems to be smaller than the chip for the 8800M series (which I'm assuming is Cape Verde or something similar in size).

As for the two "Coming soon" slots at the top right of the roadmap, maybe they'll be filled by the Cape Verde successor (which may be larger, hotter, and more powerful than Cape Verde) and the Pitcairn successor (Oland).
 
I'm not sure what the coming soon slot just above the 8600m could be.

Also, it seems likely that these 384SP chips are to mate up with current VLIW4 APU's, is it possible that we will see a re-think of GCN based APU's to first release with only 384SP's as well? Perhaps the lower positioned coming soon slot is reserved for a 512SP part, although it seems a little low for that, unless it's purposely crippled by only being available with DDR3 to match the APU it's paired with. Considering the delay it seems more likely that we won't see a part to match to GCN based APU's this release anyhow.
 
Man that's some sick naming right there. Three complete series (8500m, 8600m, 8700m) which only differ by clocks, and some of them have variable clocks within the series too. Oh and all are possible with ddr3 or gddr5 meaning a 8700m can be (quite a lot) slower than a 8500m...
That said I guess the chip should be decent, those solutions could be quite comparable to the HD7700M series (if comparing same memory type), assuming they've got a 128bit memory interface... 6 vs 8 CUs but generally higher clocks (though I'd expect them to not really be much of an improvement in perf/w but the chip should obviously be cheaper to produce and it should be a definitive improvement over the former non-gcn low-end mobile chips).
And yes the 8800m series looks like a rebadged Cape Verde to me too. Which leaves you wondering about the faster "coming soon" parts...

Using my l00t pixel counting skillz the chip seems to be quite small, just below 2/3 of CV, so about 80mm². Using that estimate though it seems it is either more densely packed or something more was chopped off than just CUs (half the ROPs would be my prime suspect in this case) as otherwise it wouldn't fit into that size (6 instead of 10 CUs is a bit less than 2/3 but the CUs only make up about half the chip). I don't think only 8 ROPs would actually hurt all that much neither (the CU/ROP ratio would just be slightly higher than on full Cape Verde or full Pitcairn and still be way below that of full Tahiti) - and when equipped with ddr3 8 or 16 ROPs doesn't matter anyway so since I'd expect tons of ddr3 designs (but look with 4GB!!!) skimping there would make a lot of sense (and actually it might be in a performance region where even ddr3 itself makes sense at least on the low-clocked versions in contrast to abominations like desktop hd7750 ddr3 which have ~70% more compute power than a 650Mhz 6 CU chip).
 
hardware.fr states 77mm² maybe that's more accurate even :). I'm not sure if I believe the 64bit memory interface they state for the 8500m and 8600m though. Would suck big time with ddr3 reducing this to a "just about IVB IGP performance gpu" (and with gddr5 would require clamshell configuration for 1GB which I consider unlikely). But then the 3 different series would make a bit more sense (even though it would discard the lower two series from being actually useful...).
 
hardware.fr states 77mm² maybe that's more accurate even :). I'm not sure if I believe the 64bit memory interface they state for the 8500m and 8600m though. Would suck big time with ddr3 reducing this to a "just about IVB IGP performance gpu" (and with gddr5 would require clamshell configuration for 1GB which I consider unlikely). But then the 3 different series would make a bit more sense (even though it would discard the lower two series from being actually useful...).

Well, im not sure about their information.. Anandtech will have write them directly if thoses one was available.
 
(and with gddr5 would require clamshell configuration for 1GB which I consider unlikely).
GDDR5 does not have a "clamshell" mode, at least not in the same way DDR3 does. You can increase the addressable size with more devices by splitting up the adress lines, i.e. a 64Mx32 device can operate in a 128Mx16 mode. You can lay that out how you want, they do not need to be either side of each other on the PCB.
 
GDDR5 does not have a "clamshell" mode, at least not in the same way DDR3 does. You can increase the addressable size with more devices by splitting up the adress lines, i.e. a 64Mx32 device can operate in a 128Mx16 mode. You can lay that out how you want, they do not need to be either side of each other on the PCB.
Well yes and that is called clamshell mode :). I've never heard that term being used for ddr3, but it appears in gddr5 papers for instance this one from elpida: http://www.elpida.com/pdfs/E1600E10.pdf
Are you saying we'll see such configurations in the mobile space? I know they aren't that uncommon for some desktop cards (using same pcb with either one or both sides populated).
 
Back
Top