...
To Paul
Deadmeat's argument fell through the roof when he found out Cell would a 65 nm chip instead of a 90 nm one. Go figure.
No it doesn't. 4 CELL cores are still too many to fit into a chip fabricated on 0.065 micron. Each VU eats up as many gates as your low end general purpose CPU. To have 32 of those plus 4 PPCs and 4 MB of cache is a daunting task. Not even IBM's Power5+ can fit more than 2 cores per die, and this is a $2000 device we are talking about.
Think about it, all of those half a billion transistor chips are largely SRAM intensive(Itanium Medison, R500, nV50, Power5), logic gates take up 4~5 times the space of SRAM gates on the average and their density per mm2 is much lower. 36 processors represent lots of logic gates and I don't think even 0.065 micron processes can fit them all.
Let's separate fantasies from reality here.
Oh and I agree, his argument held no water.
He never said anything about disagreeing with me.
To Megadrive
with regard to transistor budget, sure we dont have an unlimited budget, but we have at least a 500M budget for the CPU, at the very least.
That is if CELL has 20% logic gates and 80% SRAM gates. Even at 20% of total transistor count, logic gates will eat up half the silicon real estate due to their low density relative to SRAM gates. How are you supposed to fit 4 PPCs and 32 VUs in 25~30 million logic gates???
the GPU certainly has a possibility of going over 1 billion transistors.
Actually, there is a hint of VU transistor count in GS3. Since both EE3 and GS3 should be of similar die size, GS3 trades in 16 VUs in exchange of 32 MB of eDRAM. How much transistors are 16 VUs then??? 400 million eDRAM transistors * 1/5th density = 80 million transistors, which is in line with my original assessment of 5~6 million transistors per VU.
If PS3's CPU is just 500M transistors, can we expect 1.5 billion+ for the GPU? impossible?
I don't know if it is possible to fit 200 million logic gate transistors into a single die with 0.065 micron process. Unlikely.