AMD starts gpu production in their fabs in 06?

highly unlikely, without Fab36 going full blown for CPU's they won't be able to supply Dell, and they need a chipset volume to match ;) so thats going to take a good deal of fab space aswell.
 
Last edited by a moderator:
If it's true that the fab is only 15-20% capacity right now, a ramp up to 50% capacity for CPUs would still be a significant increase and allow AMD to supply Dell. The remaining 50% for GPUs should still be good enough.

I think it could very well be possible.
 
hard to say, if the rumored Dell deal is as big as they say, AMD will need double the capacity they have right now.
 
Actually, it sounds like the prospect of making GPU's on AMD's process is tougher than you think. How much will AMD's line will have to be reconfigured to handle non-SOI or DSL wafer production? What kind of design changes need to be made to these GPU's to work on AMD's process? If AMD can only do SOI production, what changes need to be made to ATI's designs to make them work?

I highly doubt AMD will make many, if any, GPU's for ATI this year. In the future, yes... right now? Probably not.
 
JoshMST said:
Actually, it sounds like the prospect of making GPU's on AMD's process is tougher than you think. How much will AMD's line will have to be reconfigured to handle non-SOI or DSL wafer production? What kind of design changes need to be made to these GPU's to work on AMD's process? If AMD can only do SOI production, what changes need to be made to ATI's designs to make them work?

I highly doubt AMD will make many, if any, GPU's for ATI this year. In the future, yes... right now? Probably not.
Click to expand...

Why would AMD need to handle non-SOI wafer production for GPUs?
 
^M^ said:
And the deal between AMD and ATI won't be closed before the end of the year.
Click to expand...

That doesn't necessarily have a bearing on whether or not AMD can grant ATI space in their fab.
 
As expected:
http://www.digitimes.com/systems/a20060726VL203.html
AMD EVP Henri Richard said:
In practice, we're not going to change anything. ATI is going to continue to rely on a fabless model. We want to nurture the great relationship that exists, here in Taiwan, with TSMC. And you're right, because they're on bulk wafers, and we're on SOI, even if there was a good reason for us to go and think about doing things differently, it's not practical in any short period of time. As I mentioned, one day it will make sense to put a GPU and CPU on the same die. In that case, of course, they would have to be manufactured on the same process.
Click to expand...
 
Maybe late 08 or 09 you could see a budget part made at an AMD fab if they had enough capacity which I doubt they would.
 
Gotta say though SOI could be great for xenon like chip seeing as you could use z-ram:p
 
Last edited by a moderator:
bloodbob said:
Gotta say though SOI could be great for xenon like chip seeing as you could use z-ram:p
Click to expand...

At what size point would a PC orientated GPU clealry benefit from an on-die Z-RAM buffer?
I gather a 32-bit 1920 x 1200 6xMSAA image consumes ~55MB disregarding Z/ Stencils/whatever-else buffers, would you need enough storage for double or triple buffering for performance advantages to outweigh the additonal costs?
Could you make do with holding a single image on die for manipultion then dump to main memory once completed?
Would it be worthwhile maintaining a complete Z-Buffer on die and eliminate the current cache hierachies used for Z tests?
 
Well you would probably only need a single buffer which is dumped to the main memory then. But I still doubt your going to fit the whole thing on the die even doing what the xenon does and put it on the same package is going to be diifuclt so if it happens you going to need tiling until we move to 3d chips/packages we are going to have to tile :(. ... Well then again we aren't giong to continue to increase resolution for ever so that might change.
 
bloodbob said:
Well you would probably only need a single buffer which is dumped to the main memory then. But I still doubt your going to fit the whole thing on the die even doing what the xenon does and put it on the same package is going to be diifuclt so if it happens you going to need tiling until we move to 3d chips/packages we are going to have to tile :(. ... Well then again we aren't giong to continue to increase resolution for ever so that might change.
Click to expand...

On a 65nm process 64MB of Z-RAM would consume ~120mm^2 give or take 10% according to Innovative Silicons promotional material so I would think theoretically you could shrink a R580 level GPU, include such a buffer and still have a die smaller than the current one.

Does anyone think that with Xenon's production experience in mulitchip modules ATI might head this way with desktop chips sooner rather than later? It's not like expanding beyond 256bit buses is realistic due to pinout routing problems. I know it's been publicly stated it's not on the cards for R600 (no pun intended) but would R700 be taking this into consideration?
 
The_Wolf_Who_Cried_Boy said:
At what size point would a PC orientated GPU clealry benefit from an on-die Z-RAM buffer?
I gather a 32-bit 1920 x 1200 6xMSAA image consumes ~55MB disregarding Z/ Stencils/whatever-else buffers, would you need enough storage for double or triple buffering for performance advantages to outweigh the additonal costs?
Click to expand...
For 55MB you're looking at roughly 450~500M transistors, with some redundancy built in.
Doesn't look feasible today.
TWWCB said:
Could you make do with holding a single image on die for manipultion then dump to main memory once completed?
Would it be worthwhile maintaining a complete Z-Buffer on die and eliminate the current cache hierachies used for Z tests?
Click to expand...
Z-buffer only is the most likely scenario IMO if there's not enough space to hold everything.
 
Umm, it was asked and answered in the confcon on Monday. No changes in Fabs for '06 and '07. Flat and unequivocal.

Q: Fabs? Why not bring it over to AMD? Barriers?
A: ATI has great experience with foundries and hard to duplicate. Years of work. We are on a different manufacturing. In the near term not a serious consideration. No synergies commented on are fab-related. "Won't consider it" in the near term. AMD constrained. 1-2 years internal capacity is entirely AMD. No plans in that timeframe. Outside of that we think of ATI products being part of our products to use our capacity.
Click to expand...

http://www.beyond3d.com/forum/showpost.php?p=797320&postcount=2

I'm crushed y'all don't read these things. :cry:
 
zeckensack said:
For 55MB you're looking at roughly 450~500M transistors, with some redundancy built in.
Click to expand...
I think that's overstating things by about 2x, or at least 50%.

If they went 1T ram, then that's almost a 5x overstatement.
 
geo said:
Umm, it was asked and answered in the confcon on Monday. No changes in Fabs for '06 and '07. Flat and unequivocal.
Click to expand...
Any news on whether AMD is bringing any new fabs up in the next 2 years? I recall that they just had one go online in the US this year.

epic
 
You must log in or register to reply here.

Similar threads

L
A rambling history of AMD and nVidia's rise to the top
2
Replies
20
Views
6K
Leovinus
L
Shortbread
Next-Generation NVMe SSD and I/O Technology [PC, PS5, XBSX|S]
105 106 107
Replies
2K
Views
217K
Karamazov
K
T
Takeover Talk Drives Action in NVidia Options
Replies
4
Views
3K
INKster
I
R
id Software releases Open-Source Wolf3D iPhone port
Replies
13
Views
6K
Ailuros
Ailuros
T
[Inquirer] APM tech improves Xenon yields, Chartered to start Cell production in Q3
2
Replies
30
Views
5K
The_Standard
T
Back
Top