Statistical Computing - i7 or Xeon processor?

trenkol

Distinguished
Dec 10, 2008
2
0
18,510
I am upgrading from an older desktop and a lot of my work is doing statistical computing (rather large datasets, matrix manipulation, simulations, etc) with R and Fortran programming. I have limited super computer access so would like to do as much as possible on my own computer and am tired of waiting days for simulations to complete. I'm a grad student so my budget's only around $2000, which from my limited research would put me at a mid to high range desktop or a lower-end workstation-grade machine. I realize this isn't optimal for what i'm going to be using it for, but it would certainly be better than what I have going now. :)

For the most bang for my buck given my applications, would you recommend going with the i7 (desktop), Xeons (workstation), or even back a processor generation (i.e. throwing the savings towards RAM instead)? I'm currently eyeing the Dell Studio XPS with the i7 and 12 gigs of RAM and the Precision T5400 with 2 Xeon e5405s and 4gigs of RAM, but am open to other suggestions, if that helps....

Thanks!
 

ffakr

Distinguished
Jan 31, 2009
13
0
18,510
The big difference, aside from new SIMD that your compilers won't use.. will be the memory bandwidth and latency. That's where the i7 will have a big advantage over current Xeons (though Core2 isn't exactly awful).

Some code is very dependent on memory (isn't it all though?). If you're manipulating really large matrices that'd be a good example.

Without knowing exactly what you're doing, I'd suggest a little benchmarking.

Bench your simulations, if possible, on an Opteron/Phenom and relatively comparable Intel system. If you're really bandwidth or Iop bound you might see a notable advantage to AMD.

On your current system, monkey with the bios to change the memory bus clock (if that's an option). How does this affect your simulation performance? Does performance and memory clock scale linearly? If performance doesn't decrease as fast as memory clock you're probably CPU bound.

If you look at Core2 v. i7 benchmarks, you'll see the i7 excels in SIMD (SSE4) aware code (like video encoding) and in memory benchmarks. It's faster overall but these are the strong points. I suppose the real question is, will your Fortran and your particular workload benefit from the architectural changes.

Here's another option.. Consider a small cluster. If you've got a Frys by you, pick up the weekend special a few times. I got a QuadCore Phenom w/ motherboard for $100. It's a fantastically impressive machine for twice the price (as they say). The big downside was only 2DIMMs on the board.. However, $2000 would easily build you a 4 node cluster (16 cpu cores).
 

TRENDING THREADS