My dumb Xenos die size calculation

[superguy]

Knowing the transistor count, and ASSUMING ATI counts fairly consistently across it's own products..

Xenos shader core 232m transistors
R580 352m transistors

First I tested it on R520. Not knowing R520's die size, I used the 321m transistor figure.

321/384=.8359

So .8359*352mm^2=R520 predicted die size=294mm^2 estimate for R520. R520 actual figure? 288mm^2. So we're pretty close. And confidence grows in the method.

So Xenos =232m, =.604 as big as R580, =213mm^2 by my estimate.

G71=196mm^2 for reference.

All measurements by Dave.

 

[Guden Oden]

Why are you bothering to calculate the result when it's just a matter of looking at a picture of the die and measure it that way

 

[Mintmaster]

Do we have a picture with some sort of scale on it? Or at least something we know the size of nearby?

 

[Edge]

From this page, R580 is 384M transistors, and not 352M that you list: "R580 352m transistors".

edit: Oh I see, you used the correct transistor count in your calculations versus R520.

G71 is packing a lot more transistors in a smaller space, 278M at 196mm2, than your Xenos estimate, 232M at 213mm2. Memory is more tightly aligned, over logic, so maybe more memory cells on G71?

The fact that a different architecture like G71 compared to R580 is packing a lot more transistors in a smaller space could also mean that your Xenos estimate could be way off. Or Xenos is more similar to R580 than G71.

 

[superguy]

From this page, R580 is 384M transistors, and not 352M that you list: "R580 352m transistors".

edit: Oh I see, you used the correct transistor count in your calculations versus R520.

G71 is packing a lot more transistors in a smaller space, 278M at 196mm2, than your Xenos estimate, 232M at 213mm2. Memory is more tightly aligned, over logic, so maybe more memory cells on G71?

The fact that a different architecture like G71 compared to R580 is packing a lot more transistors in a smaller space could also mean that your Xenos estimate could be way off. Or Xenos is more similar to R580 than G71.

I had heard something to the effect ATI doesn't count cache memory in it's transistor counts, just logic, where Nvidia counts all.

This is just a rumor and one possible explanation..the general gist being that supposedly ATI counts transistors more conservatively. But do I know that for fact? Not really.

 

[Mintmaster]

Okay, so here's my calculation:

From this page, Anandtech compares Xenos with Yonah, which he/Intel claims is 90 mm2. Measuring pixels, I find the parent die to be 1.94x Yonah and the daughter to be 0.74x Yonah.

That gives us 174 mm2 for the parent chip and 67 mm2 for the smart eDRAM.

However, that photo makes it look like the full Xenos package is 35mm x 35mm, as it is the same size as a Pentium M. By looking at the B3D die shot, I find Xenos to be 186 mm2, and the daughter to be 70 mm2. Not far from the above measurement.

It looks like there's isn't much density difference between Xenos and G71, maybe 10% or so. Somewhere I also heard 250M for the parent die, so even this could be wrong.

 

[superguy]

Yep, I think it's pretty clear from eyeballing it to eyeballing a G71 photo they look around the same size, very roughly.

Xenos die looks more sqaure shaped and G71 a bit more rectangular, but they look fairly close one way or another, as all these estimates in this thread agree.

Oh and 232 I think is the official figure of the shader core..perhaps you're thinking of the 257 with daughter die logic included that some people use?

The daughter die is 105 and the EDRAM is figured to be around 80, so that's where the 25 more logic idea comes from.

 

[SiBoy]

However, that photo makes it look like the full Xenos package is 35mm x 35mm, as it is the same size as a Pentium M. By looking at the B3D die shot, I find Xenos to be 186 mm2, and the daughter to be 70 mm2. Not far from the above measurement.

Correct