CPU choice for statistics

Hi,

I'd need some advice concerning CPU choice for my particular application. I will partly use my laptop to run statistical programming softwares such as R and WinBUGS, possibly SAS. I don't know whether these softwares can benefit from multi-cores or increased cache but I guess they do. I'm planning to get a ThinkPad X220 and hesitate between a core i5 2520M or a core i7 2620M...or even another machine and get a i7 2720QM quad-core if that would really be beneficial (though I'd highly prefer an ultraportable machine...).

First on the X220, would the increased cache size of the i7 2620M help speed things up in R compared to the i5 2520M or is the difference barely noticeable? I guess +200Mhz does not count much here...

With that kind of use, would you consider another laptop to get a quad core or would the X220 still be fine? I want the X220 because it's small and light and also because I want a good screen (IPS). If that is not possible, would you recommend another laptop below 2kg (<4lbs), max 13" good screen with quad core?

I don't need blazing fast computations and I won't run R all the time but since i plan to keep the laptop for a long time, I'd prefer an educated choice specific for my application.

Thanks for any input

 

I would just get a dual core. Im not sure about your case, but in my experience of scientific computing its either been very fast (<2sec so it doesnt matter), or very slow (use cluster for 20 days, not on laptop unless you are really desperate). Ive used an Atom before to run matlab (just normal m file), it runs surprisingly well.

Not sure if stats programming is the same.

If you write a program then run it, then you might as well get the laptop which is in the form factor you like, i.e. the x220, since you will probably test on a small case then go bigtime on the data server or whatever, so processing speed is not too important.

However, if its more kinda interactive, then the difference between say a 4 sec run time and a 2 sec run time if you are doing it 100s of times a day will be more important. A bit like the engineering cad type stuff where this sort of usage model occurs.

 

I used R on an older core2duo and I never experienced any delay. Everything was pretty much instant. Then again I didn't process huge amounts of data so maybe that's why

 

If you're doing heavy arithmetic computation and matrix manipulation (like in MATLAB) then clock speeds are absolutely the crucial factor (over parallel cores).

More cores allow you to run more instances of R or your statistic package (e.g. on different sets of data, or different operations on the same set of data) in parallel. But if you tend only to use one instance of your statistic package, then go for clock speed over number of cores.

 

After digging a little more, it seems that R is a single-threaded application so a quad-core won't help here I guess. Probably that the i7 2620M with its highest clock is then the best option (and it brings USB 3 on the X220 by the way). But is the larger cache size of any benefice here (4Mb on i7 vs 3Mb on i5)?

 

I must say that there are benefits to be had for sure, but the price-performance increase doesn't appeal to everyone. Going from the i5 to the i7 (difference in terms of clock speed and cache), if i remember correctly, is a $100++ premium. You may gain some small performance increase (in terms of how fast your computations can be completed), but to many people, its not worth that $100++ price increase.

I personally chose to go with the i7, since I do academic work (simulations, scripts, matlab, data processing, LaTeX, illustrator, photoshop) and every little bit of speed increase helps.

 

Yes it's about 100$ more but I plan to keep the laptop for a good while and will probably use USB3 devices when they'll become widely available. If you consider that a USB 3 express card adapter costs about 45$, the CPU upgrade become "only" 55$ compared to the i5 2520M.....not that bad!

Thanks guys for all your help!