Let the beating begin!

But what do we know about actual margins? We know that ATI, excluding the chipset business (which is awful margins for them, we all know) has stated that the rest of the business is in the 34-38% range. Let's be conservative and give ATI the middle of that range at 36% for the GPU business, which is still something like 80% of their business.
Click to expand...

What is their chipset business margins if their GPUs are at 36% and the chipsets account for 20% of revenue? Are they giving these things away to get down to 28.2%?
 
Last estimate I heard was "high single-digits". When asked to promise 25% was in sight, Orton just wouldn't go there.
 
Dave Baumann said:
However, even a quick look at two figures would tell you that there is something seriously wrong with the calcuations. Your costings for a high end wafer was at $20K (while I don't have anything definate, thats I figure that I have had floating around in my head for a while now as well, so I'll bite that) and then there is 164 cores per wafer for R580. This puts the cost per core at around $121 and applying a 35% margin (IIRC this is what they said for desktop without integrated, and there is likely to be a premium for parts like R580 over that, but we don't know how quantifyable that is), which puts the selling price per core at around $164 - thats for all chips on a wafer to come out at the profit margins they are stating. That's not unreasonable, roadmaps from last year suggested that R580's die cost was ~$200, so allowing for some yeild drop off we are probably in sane areas.
Click to expand...
I'm sure they don't pay $20k per wafer, and I'm sure they're not even close to 80% yield for R580.


Josh, would you please share the formula you used to calculate the good net die per wafer? I consistently get 8-10 less dies without defect. That is, going with a fixed defect rate of 0,5 per cm², for a more correct number we would need the standard deviation of that value as well.
 
Xmas said:
I'm sure they don't pay $20k per wafer, and I'm sure they're not even close to 80% yield for R580.
Click to expand...

? I didn't say they were yeilding at that rate, but my point being is that, if Josh's die numbers are correct then there is probably a majority of chips lying around that will work with fewer quads, but this doesn't mesh with the fact that they've been shipping R580's for 3 months and been in production longer nor that there are rumours of a separe chip coming up designed to fill this SKU.
 
Xmas said:
I'm sure they don't pay $20k per wafer, and I'm sure they're not even close to 80% yield for R580.


Josh, would you please share the formula you used to calculate the good net die per wafer? I consistently get 8-10 less dies without defect. That is, going with a fixed defect rate of 0,5 per cm², for a more correct number we would need the standard deviation of that value as well.
Click to expand...


Were you adding in the 2mm exclusion?

Edit: Basically I used this tool from ICKnowledge. I was using Rev. 3, but they have Rev. 5 up now that can utilize different defect methods: http://www.icknowledge.com/misc_technology/miscelaneous_technology.html

It is the free die and yield calculator about halfway down.
 
Last edited by a moderator:
Dave Baumann said:
? I didn't say they were yeilding at that rate, but my point being is that, if Josh's die numbers are correct then there is probably a majority of chips lying around that will work with fewer quads, but this doesn't mesh with the fact that they've been shipping R580's for 3 months and been in production longer nor that there are rumours of a separe chip coming up designed to fill this SKU.
Click to expand...
And my point was that if they're paying $20k per wafer, they're not at $200 per die.


JoshMST said:
Were you adding in the 2mm exclusion?
Click to expand...
Err, is that supposed to be a 1mm border around every die? :?:

Anyway, I calculated with die area and your die per wafer count. But I used the poisson model.
 
Last edited by a moderator:
Dave Baumann said:
? I didn't say they were yeilding at that rate, but my point being is that, if Josh's die numbers are correct then there is probably a majority of chips lying around that will work with fewer quads, but this doesn't mesh with the fact that they've been shipping R580's for 3 months and been in production longer nor that there are rumours of a separe chip coming up designed to fill this SKU.
Click to expand...

I think Dave Orton has a pathway through his garden paved with all of the defective R580 die. It apparently looks very shiny and is a mile long...
 
Okay, you geeky bastiges this is Industry. If you're going to throw Murphy Model and Poisson Model around, links damn it.

:p
 
  1. The heavy memory dependency of R5xx for the register file (compared with prior generations) combined with the fact that memory is extremely easy to architect with ultra-high yield (through fine-grained redundancy) creates an opportunity to use enhanced levels of in-die redundancy
  2. "3:1" is still a funny ratio to me, and I honestly expect that each of the four shader units in R580 is constructed with 4x fragment quads, but with one of them sacrificed for redundancy (much like Cell in PS3 sacrifies one SPE for redundancy). This leaves each shader unit with 3x active fragment quads.
These two concepts, in my opinion, mark a dramatic shift in yield-modelling for R5xx, as compared with previous GPUs.

Jawed
 
kemosabe said:
Geo unless I'm mistaken those ATI chipset margins are now somewhere in the teens and climbing (very slowly).
Click to expand...

Hmm. I'll have to go look at my last couple sets of notes. That might be a blended#, as crossfire chipsets start to get something more volume-y. I know I heard that "high single-digits" number somewhere from them recently, but it might have been IGP-only.
 
Pretty sure the single digits was from a previous CC. They're "in the low teens" at the moment, according to later CC's/presentations.
 
geo said:
Hmm. I'll have to go look at my last couple sets of notes. That might be a blended#, as crossfire chipsets start to get something more volume-y. I know I heard that "high single-digits" number somewhere from them recently, but it might have been IGP-only.
Click to expand...

I remember reading a statement a short while about (about a week now), on this forum section I believe, that showed the current chipset margins being in the very low 20's.

IIRC though, it did seem a little optimistic so, could be wrong there. :eek:
 
After reading that "article", it makes me wonder why someone would post this knowning they don't have a full understanding of what they are writting about. With such gross misrepresentation of yeilds, it makes me wonder what the modivation really was.

These are the type of articles that make a website lose credibility, in my opinion.
 
You must log in or register to reply here.
Back
Top