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

OK, time to dig for the die size of the thing.

With some quasi-precise measurements, the die area comes up rather large at 432,6 sq.mm -- a notch higher than R600:

cgn.jpg


For reference, I used the distance between the mounting holes, presumably the same as for every high-end SKU from ATi/AMD ever since HD2900.
 
I used size of memory module as reference and my result was about 415 mm². It seems to be definitively closer to Cayman or R600 than Cypress (unless the picture is PSed).
 
OK, time to dig for the die size of the thing.

For reference, I used the distance between the mounting holes, presumably the same as for every high-end SKU from ATi/AMD ever since HD2900.

At 3DCenter.de Davidzo used gimp to rectify the pic:

http://www.abload.de/image.php?img=7970diesize48lps.jpg

=> the distance between the mounting holes seems to be higher at 53mm instead of 50mm(*) for the older 6900 series, measured from the middle of the left hole to the middle of the right hole.

*= measured with irfanview using a highres pic of a 6900-series board
 
Hm, is he using the board dimensions for base reference? This is prone to large error accumulation together with rectify, considering the poor state of the source photo. For instance, I opted to "bend" the measurement plane so it could match the perspective in the photo, leaving it intact.

For a base reference I used the length of the PCI-E slot (71mm sans the power contacts). The distance for the mounting holes is 53,7mm in my book (center-to-center), so I guess we are both right at least for this one.
 
no i used the mounting hole distance as a Reference.
Your first pic suggests 58.6mm for the holes which is close to the distance of the gtx580.
ATI mounting holes are 53mm since X1900 (same as G92)

i just inserted the overall dimensions larer to get a whole idea of the card, it has nothing to do with the scale of the picture.


http://www.abload.de/img/hd7970_entzerrtiwpag.jpg
[URL]http://www.abload.de/thumb/hd7970_mithilfslinienyaj7e.jpg[/url]


please dont take the hundredth of milimeters as extend of tolerance, its just what my cad programm offers as a standard.

the extend of tolerance is somehow in the range of +-0.5mm in both directions, so it is between 18*20 (360mm²) and 19*21(399mm²).
 
To be more precise, measure all combinations of mounting hole distances, i.e. both verticals, both horizontals and both diagonals, to correct most accurately for perspective/straightening.

And also don't forget that the packaging of a die adds 0.5mm, roughly, to the dimensions of the die inside.
 
Well, sadly the die itself on the photo has been tampered on (probably to hide various markings) and there's too few clear pixels where the actual edge is visible. For now, it's all about the good guess.
To be more precise, measure all combinations of mounting hole distances, i.e. both verticals, both horizontals and both diagonals, to correct most accurately for perspective/straightening.
I used the Vanishing Point tool in PS, that takes both vertical and horizontal dimensions to project the correct perspective, upon which the relative measurements were made.
 
So, it's within the 380mm - 420mm range, 384-bit, and GCN. Did S|A or BSN have an idea about the die size from an earlier article?
 
I am not closely following all the rumors but wasn't GCN mean to first appear in 2013 products?
 
I am not closely following all the rumors but wasn't GCN mean to first appear in 2013 products?

Not ever since it became public, it was indeed GraphicsGenerationNext

I, however, believe that AMD skipped one planned generation due delays from 32nm cancellation - as in, 32nm was supposed to be full VLIW4 (except perhaps lowend rebrands), and 28nm probably too, but since 32nm cancellation delayed HD6-generation, they ended up with hybrid generation of VLIW5 and VLIW4 highend adapted from the situation at hand (backup plan Barts @ 40nm + 32nm Cayman adapted for 40nm, lowend small tweaks to existing chips), and this delay allowed them to bring GCN 1 generation earlier (or rather, half gen if the delays had never happened) than the original intentions were.
 
I am not closely following all the rumors but wasn't GCN mean to first appear in 2013 products?
At the Fusion Developer Summit last summer Eric Demers discussed features that would be rolled out over the next couple years so GCN was not a specific generation though most have come to describe the Southern Islands generation as GCN.
 
At the Fusion Developer Summit last summer Eric Demers discussed features that would be rolled out over the next couple years so GCN was not a specific generation though most have come to describe the Southern Islands generation as GCN.
Demers was mainly speaking about the Fusion System Architecture. Not all mentioned features (there was a slide with several tableaus as some kind of evolution roadmap) will be in SI. The talk of Mike+Mike (Mantor and Houston) two days earlier detailed the GCN Compute Unit architecture that will be used in SI (higher end models, not the entry level and mainstream) and probably a few more future generations.
 
I am not closely following all the rumors but wasn't GCN mean to first appear in 2013 products?

No, late 2011/early 2012.

At the Fusion Developer Summit last summer Eric Demers discussed features that would be rolled out over the next couple years so GCN was not a specific generation though most have come to describe the Southern Islands generation as GCN.

At Fusion Developer Summit, Eric said that the GCN is used in the vga at the end of 2011.
 
Please consider that those values are in no way declared to come from inside knowledge but are explicitly just an estimate from the die size and thinking about some reasonable values.
 
At Fusion Developer Summit, Eric said that the GCN is used in the vga at the end of 2011.
Back then they didn't know they'd have to delay the launch by a few weeks due to capacity/yield issues at TSMC. Most rumors and alleged insiders point to a release in January now.
 
Back
Top