"Will ATI Be Broken Up?"

I'm a bit surprised by this part :

Custom logic all the way ?

 

custom logic, longer pipelines, process shrinks. this is pretty much standard concepts in the industry, not really nostradamus.

 

Where is AMD going to get the necessary number of circuit designers to push the GPU pipes that fast?
If the GPU is on the same process and same die, there's going to have to be a significant engineering investment in implementing any GPU, but AMD still has to worry about its x86 lines.

It would take time for two pools of engineers to get up to speed, and I don't know if before the buyout that the low-level anal-retentive kind circuit design needed for x86 was that much of a priority at ATI.

At the same time, the discrete GPUs will probably not be on the same process (not until AMD gets more fab space), so they'll need their own pool of engineers.

Even if AMD stays in the high-end segment, there's a fair question about how well it can do there without making sacrifices in either department.

 

I think they are dreaming in the mobile handset space. x86 for mobile? Show me an x86 chip that has standby power consumption of .10 mW and average consumption of 200 mW loaded. What's happening in mobile is fixed-function everything. A moderate CPU coupled with fixed function blocks accelerating everything from java, image codecs, to cryptography. This will remain more power efficient than trying to stick a faster CPU in there to perform the same functions. With the exception of uber-PDA style phones, most handsets are not going to go the way of trying to make them into UMPCs, short of a revolution in fuel cell batteries and human interface. I simply don't want my Motorola RAZR to turn into a generalized computing device, I want to it to do a few things well using as little power as possible.

Intel didn't understand this either, which is why they failed as well, and it's why TI, STM, Hitachi, Toshiba, Renesas, Motorola, Broadcom, et al are kicking ass.