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

Sure it does

I wonder if all islands in Canada are mobile chips.

 

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

 

Actually the quote from DigiTimes is:

Doesn't that mean that they have STARTED tape-outs not that tape-outs are complete?

 

It means they don't know the industry lingo. Which is totally not surprising.

 

Tape-out means sending the design to a manufacturer, nothing more. It isn't a long-term proces, so it doesn't make sense to distinguish between a started and completed tape-out.

 

Is it normal that a whole series of GPUs tapes out at once?

 

What if the first tape-out comes back with problems?

Isn't the tape-out procedure an iterative process?

http://en.wikipedia.org/wiki/Tape-out

If AMD already has all the Sea Islands taped out (and working) why are they waiting 6 months before producing them?

 

Taped out isn't the last step before production.

 

Then you do metal spins. If problems are too big, then you have to do a silicon spin.

No.

 

A tape-out is the act of FTP'ing a file to the fab. That's it. A tape-out doesn't come back. Silicon comes back.

Hopefully not. There are exist checks done before sending the tape. And the fab does input checks. Both normally pass.

That said: for the last 15 years or so, all tape-outs I've seen where a 2 step affair: first you tape-out the base layers. Then, a week or 2 later, you tape-out the metal layers. This is a convenient way to cut 2 weeks of the total tape-out to first silicon schedule, which is getting increasing longer for a state-of-the-art process. It used to be 4 weeks from base to silicon (that's when I was still so very young), now it's getting close to 3 months or so (it depends on how much you're willing to pay to hot, super hot or super super hot lot it, and on your standing with the fab.)

For 28nm? I don't know exact numbers, but let's say: 14 weeks from tape-out to silicon. Then a shit-load of functional verification. Then tons of system stress testing. In parallel a whole bunch of long term reliability testing (accelerated testing at high temperature/voltage etc.) It's hard enough to complete this in 6 months once silicon is back (let alone tape-out), though AMD and Nvidia are supposedly pretty good at it and do in less. How long? Some reliability tests run for 2 months, and they can't be started immediately after silicon comes back, so that's very optimistic lower bound.

 

They look perfectly on track for me, i remember South Island have tap out around the same date. Same for Cypress. In general this is when they start ramp up the production, some problem can appears and delay the launch. But this time it should have been fixed on TSMC side. ( Ramp up the production is the last step, there's ton of other steps between the tape-out and the ramp up)

 

Guys, do you want to play that "guess the specs" game?

"Tenerife" (let's call it in this way) might be a 5100 MTrans at ~410 mm^2 beast with 2560 SPs , 160 TMUs, 32 or 48 ROPs, 384bit GDDR5 and 30-40% faster than Tahiti.

There are also other specs for the smaller chips.

http://www.3dcenter.org/news/eine-erste-prognose-zu-sea-islands-aka-amds-radeon-hd-8000-serie

 

Well, that would mean something like GK110 ≥ Venus > GK104 > Oland > GK106 > Mars > GK107…. I'm wondering what the TDPs for the (high-end GPUs based on the) chips would be. Venus's especially.

 

I'd be calling it "Venus XTX," since we already know the GPU codename. Something to note -- AMD hasn't had an "XTX" part since the 1900XTX, which was the last time they ruled the roost (if you're not entirely conviced that AMD is the leader now with the 7970GE). AMD might think they've got a winner in their hands.

 

You know what?
This thingie is a mobile chip codename. And most likely the second biggest chip.

Venus XT, Venus Pro, Venus LE can't be HD8900 series.

http://forum.beyond3d.com/showpost.php?p=1649939&postcount=300

 

http://translate.google.com/transla...-zu-sea-islands-aka-amds-radeon-hd-8000-serie

 

The sad thing is that with ignoring the obvious, lack of thinking logically for the sake of posting and believing mostly meaningless article won't get you too far. I agree that the article is long and there is much to read but you have to distinguish what's worth it.
The other sad thing is that these fiddlesticks repeat every coming generation. The same was the nonsense with Tahiti, Pitcairn, etc codenames.

 

Geez, like your basis for calling it Tenerife is any better. What hypocrisy.

 

So, what are the chances of TSV/stacked DRAM/HBM (High Bandwidth Memory) debuting with HD 8000? We already know that it's on its way: http://semiaccurate.com/2011/10/27/amd-far-future-prototype-gpu-pictured/

GloFo is tooling its 28nm process for TSV, Hynix has been toying with it for ages, Haswell's rumored to have it... it looks like 2013 might be the dawn of the TSV age.

 

I would think the HD8000 series comes too early for that.