A Prescott Northwood..........

[iwodo]

I read most of the article on the net and they all seems to point out one thing. Give credit to intel because they manage to make a chip the same only only just slighly slower with a 55% increase of pipeline. While i personally think this is quite a achivement as well but it makes me wonder something.

What happen if strain silicon, all the good optimaztion on ALU, and trace cache etc were put on to Northwood.......

or put it in other word. A Prescott with 20 Stage Pipeline and 512K cache ( but much lower latancy ).

This will make Prescott much faster than today's Northwood. ( i mean if it could perform the same with all optimazation and 55% increase of pipleline. it is hard to imagine what heppen if the 11 stage of pipeline penatly is removed. )

This means......It would be cooler then current northwood because it is using 0.09 with strain silicon. It is faster.... And scale a bit more. ( may be not 5Ghz............. )
And using less power as well.

Because i am getting fed up with intel CPU now more like a heater in my room!!!!!!!

Just give me a god damn 533Mhz FSB 2 GHz Dothan CPU on my desktop!!!!!!!

 

[jpetermann]

 

[Corsairpro]

Originally posted by: iwodo


Just give me a god damn 533Mhz FSB 2 GHz Dothan CPU on my desktop!!!!!!!

They have that, its called an Athlon 64.

 

[iwodo]

Well........... I would think apart from the 64 bit Athlon 64 isn't as advance as Dothan........ At least Dothan 2Ghz Maximam power is only 42W............................

 

[Sahakiel]

Originally posted by: iwodo
Well........... I would think apart from the 64 bit Athlon 64 isn't as advance as Dothan........ At least Dothan 2Ghz Maximam power is only 42W............................

Depends on your definition of "advanced." Obviously, Athlon64 has the newer ISA support. Dothan also borrows heavily from the P6 core.

At 90nm, there is also the issue with leakage current on the transistors. This basically means that even though the transistor is "off,", it's still using power. I remember hearing how it's becoming a larger problem with successive die shrinks.

If you remove the 11 stage penalty from Prescott, it's hard to tell how much of an increase you'd get especially if you don't have the tools to find out. The current 20 stage pipeline already suffers from severe branch penalties and the Prescott supposedly addresses part of the problem with a better branch predictor. Supposedly, it borrows the indirect branch predictor of the Banias core. If I remember correctly, compulsive branches and conditional branches have about 98% or better prediction rate. Indirect branches using the same hardware have about 50-60% prediction accuracy. From an architectural standpoint, it would seem that clock for clock, the Prescott core would beat out the older core.
However, you also have to consider the extra logic required to implement all the new optimizations. More logic = slower clock, that is the general rule. Given just the extra branch prediction hardware, it would be a good assumption to say that the branch logic is now slower. That alone would slow down the overall clock without some serious tweaking to the pipeline. Even if the pipeline were re-balanced to produce simliar clock rates to the older core, it would imply that the branch penalty would be higher to compensate.
All in all, it's hard to say.

 

[Corporate Thug]

Originally posted by: iwodo
Well........... I would think apart from the 64 bit Athlon 64 isn't as advance as Dothan........ At least Dothan 2Ghz Maximam power is only 42W............................

Dothan is delayed