Nehalem is fast

[liquid stereo]

I use a MacPro with 8 cores. I also use 8-core AMD (Shanghai) CPUs.
I write my own parallel code (parallelized via MPI) and these codes scale linearly up to around 1,400 CPUs/cores.

The new CPUs are ridiculously fast. By that I mean my codes are 2 to 4 times faster than the 8-core AMD.

 

[Tallest Skil]

Yes, the Mac Pro is fast, but you don't have Nehalem.

 

[liquid stereo]

Correct. But I've seen the performance of my codes on Nehalem. Its ridiculously fast for codes/applications that are memory-intensive.

Yes, the Mac Pro is fast, but you don't have Nehalem.

 

[nanofrog]

I've also linked some geekbench scores in another thread, where a Core i7 920 was OC'ed to 4.2GHz, and kicked the crap out of the current Mac Pro.

Tom's Hardware also has a performance chart comparing 3.2GHz Penryn and Nehalem (Core i7 965) to one another. Very good results, and by extrapolating, will be worth the additional $$ IMO.

 

[liquid stereo]

Performance really depends on code

The performance I've seen really depends on code/algorithm.
Codes which are memory limited will show the greatest performance increase.

Codes which aren't are going to scale with clock speed. Meaning for some calculations the 2.8GHz Nehalem will be slower than the current 3.2GHz Xeon.

Fortunately my codes are memory intensive and show significant speed-up.

Cheers!

I've also linked some geekbench scores in another thread, where a Core i7 920 was OC'ed to 4.2GHz, and kicked the crap out of the current Mac Pro.

Tom's Hardware also has a performance chart comparing 3.2GHz Penryn and Nehalem (Core i7 965) to one another. Very good results, and by extrapolating, will be worth the additional $$ IMO.

 

[nanofrog]

The performance I've seen really depends on code/algorithm.
Codes which are memory limited will show the greatest performance increase.

Codes which aren't are going to scale with clock speed. Meaning for some calculations the 2.8GHz Nehalem will be slower than the current 3.2GHz Xeon.

Fortunately my codes are memory intensive and show significant speed-up.

Cheers!

I know.

The chart posted by Tom's made a decent attempt at using enough of a variety of commonly used apps to give a more well rounded outlook. Broad enough, it can be of use to a fair number of users.

Email and web browsing won't be significantly faster, but the EDA package I use is another story.

The code will hopefully progress enough where it can utilize such a system, but another two generations of hardware may be out by then.