Best Computer for Computing work

[hellebyuck]

First off, if this isn't the proper place for this, you have my humble apologies.

That said, I'm looking at buying a new computer, and pretty much the only thing I do which really requires any processing power is a lot of work in Matlab and Maple. Since it can be quite a pain waiting on slow computers to get done with the number crunching, I obviously want to get a system that can best handle this sort of thing. Unfortunately, all of the benchmark tests on different processors that I've seen are geared towards graphics, as opposed to pure computing, and I really don't know how well the two correlate. If anyone has any advice on what types of processors and computers run things like MATLAB simulations most efficently, I would greatly appreciate it.

 

[Dr Transport]

The one thing you want is gobs and gobs of memory, if you can afford 2 GBytes do it. When it comes to number crunching on a PC the more memory the better. A few years ago somone did a study, for example, a 1.5 GHz machine with a gig of ram ran faster than a 2.0 GHz processor with 512 M ram.

 

[PerennialII]

On a general level yeah, 1 Gb memory, 2 Gb if you got the wallet depth sounds like a good idea. CPU thing is 'easy', just get the latest & the fastest available, if you can afford getting all the way to dual core 64 bit support ones sounds tasty. One typically omitted aspect is the OS, if at all possible, get a 'real' OS (=anything other than windows). The performance gain of running Linux rather than windows (and typically is, in the line of stuff I do usually 20-100%) on an identical hardware is good to harvest.

 

[hellebyuck]

Thanks guys.

From research on other forums, I'm going with an AMD Athlon 64 4000+ and a gig of RAM. I can afford this now, and the system I'm buying can easily be upgraded later, especially for RAM. Actually, the first after market things I'll be doing is bringing it up to 2gigs of RAM, because I can buy and install it cheaper myself, and adding a liquid cooling system for the processor for better overclocking.

But yeah, turns out Matlab supports 64bit architecture, so the AMD Athlon 64 chips can provide the best performance for my needs.

 

[-Job-]

I don't understand what was said by 'real OS' = 'other than windows'. Linux has some nice advantages over windows and some nice disadvantages as well. If you're a programmer, Linux is nice because it gives you a lot of freedom, otherwise i would go with windows for their hardware/software compatibility, and the 20%-100% mentioned above is certainly not across all architectures or versions of Linux, and i would question whether the author of that statistic isn't somewhat biased. Linux uses some different memory management and process scheduling algorithms than windows which, again, have advantages and disadvantages, and scenarios where they perform fast and where they perform slow.
I have dual 64-bit 3.0 Ghz processors with 1Gb of memory on my PC running Windows Server OS and the latest SUSE Linux, and Windows is significantly faster than Linux, a problem which i try to play down to driver support.
The advantages of Linux over Windows are mostly due to its being free and open source, anything else will be a tough case to make.

 

[PerennialII]

I don't understand what was said by 'real OS' = 'other than windows'. Linux has some nice advantages over windows and some nice disadvantages as well. If you're a programmer, Linux is nice because it gives you a lot of freedom, otherwise i would go with windows for their hardware/software compatibility, and the 20%-100% mentioned above is certainly not across all architectures or versions of Linux, and i would question whether the author of that statistic isn't somewhat biased. Linux uses some different memory management and process scheduling algorithms than windows which, again, have advantages and disadvantages, and scenarios where they perform fast and where they perform slow.
I have dual 64-bit 3.0 Ghz processors with 1Gb of memory on my PC running Windows Server OS and the latest SUSE Linux, and Windows is significantly faster than Linux, a problem which i try to play down to driver support.
The advantages of Linux over Windows are mostly due to its being free and open source, anything else will be a tough case to make.

I can understand your point very well since been balancing between different Linux releases and Windows for quite some time. Granted the "20-100%" is certainly not an exact number, but as said, has applied pretty well in CPU intensive numerical PDE work (and is a basis of the figure overall) that have done and seen software houses benchmark (when talking about numerically solving PDEs). Comparing performance of windows contra Linux (primarily suse & rh) and other OSs (ones like for example HP-UX and Solaris) has produced such differences in our applications, not particularly favoring windows. If you're getting good performance out of windows that's great, however the memory management and parallel performance alone make in unfeasible in our 'line', in addition to the performance questions. As I see it quite a bit of the difference arises from compilers, numerical libraries and scaling, which make up solvers etc. which for us at least clearly exceed the performance of anything windows has to offer (Anyways, pre- and post-processing is done in win due to software support - the crunching in Linux). Sure could get a more accurate readings if "de-biasing" various reported benchmarks, but probably not that crucial of a point in this case.