Need advice on buying a PC

[Homer_simpson]

My laptop has been my PC for 3 years now but it has its limitations so I'm looking into buying a PC and I need some advice on what to get for the specific purpose that I need it for. I dont need this PC to be able to run highend graphics or games. My main reason for getting it is to be able to run simulations,computations and data processing which are taking too much time to run on my laptop.
I would say I'm a noob when it comes to figuring out the parts best suited for the task but I'm guessing highend processor and the motherboard are what I need. Most of the preassembled PCs that I've looked up seem to be designed for gaming/media and are more expensive since they have components that I dont need.
So my question is would I be better off trying to build a PC part by part or is buying a preassembled PC?
Also what kind of processor is better suited for my task?
any other helpful tips and comments are also appreciated.

[ProgrammingFun]

What is your price range?
Depending on your hardware knowledge level, a custom PC will be cheaper but may end up with hardware conflicts/a lot of time to look up and buy the required parts.

On the other hand, a pre-built PC will be more expensive but you will spend less time and thinking power

HP did release a new laptops (mostly centered towards entertainment though) which were briefly on sale for $850: http://www.engadget.com/2011/05/18/hps-pavilion-dv6t-and-dv7t-available-with-1080p-screens-people/ (click the two LogicBuy links at the bottom of the article for the savings)
I currently use an older generation of the dv6 (last year) and it does have great capabilities.

As for desktops you could try this (I know I am being HP-centered...) in which you can opt to not choose the entertainment parts.

As for processor, you could get an Intel Sandy-Bridge if you want the latest though the older generation is also pretty good. I don't really like/use AMD so I guess an i5 or i7 would be good enough.

[Sur_real]

um, I'm biased too...but towards the hatred of HP, lol sorry PF

Quote:

My main reason for getting it is to be able to run simulations,computations and data processing which are taking too much time to run on my laptop.

Since you need basically power and maybe speed, I suggest going for a custom build with a SB i7 and a SSD (if you want)

[Tony]

Since this is for a particular (set of?) application, I'd recommend actually profiling the system resources to see where the bottlenecks are.

Are you _sure_ that you are CPU bound? If so, how well do the tasks parallelize? That is, are you better off with more cores, or a single faster chip?

How much memory is actually used up? There is no performance gain from having idle memory.

Does HDD access speed matter?

[Homer_simpson]

I'm running into speed problems mostly when using mathematica to process large sets of data or to run simulations and on occasion simple C++ codes with large loops. I dont know too much about the way mathematica uses my processor but when i check resource monitor Both my CPUs are running at 100% with mathematica having 38 threads(I dont know what threads are). Also I'm not sure if this is of any significance but mathematica seems to run a separate file called MathKernel.exe simultaniously as the mathematica.exe file.
there seems to be no HDD access required and I have 4 GB of ram which is not maxed out. At this point I'm wondering if it is a short coming of my computer that these simulations are taking too long or this is simply the limit of the best time I could achieve, In other words will there be significant gains if I get a more up to date PC. Currently certain simulations take well over 30 minutes which is too long for my application.

I guess I should mention that I'm currently using a dual core intel u7300 laptop @1.3 GHz with 4 GBs of ram.
Also I tried doing a bit of online research on CPUs but theres too much scattered information that are contradictory and not so credible everywhere. for my application should I be looking at i7 or AMD alternatives?

Thanks for the responses.

[chrisbrown]

A high-end CPU will definitely perform, but if you are serious and can justify the time to learn a new skill set, it might be worthwhile to look into GPU acceleration - Obligatory video.

I don't know much about it, but as I understand things, GPUs are optimized for "x=y"-type statements, but struggle with conditionals. So if your code is light on if-statements, the GPU will greatly outperform any CPU (by a factor of 40 in some cases).

* Forgot to mention, Mathematica 8 (and maybe others?) supports this through different SDKs.

If that's too over-the-top, a high-end chip from either vendor will still give you noticeable results. Midrange laptops just don't have the hardware to process big things efficiently.

If you give us a budget and what you already have (case? monitor?), I for one would be happy to make some recommendations.

[mirhagk]

Apparently batman arkym assylum built a physics engine (to animate a wall of bricks exploding) completely on the GPU. I have a book that discusses creating physics engines and such on a GPU. THe biggest benefit is the extreme multi-threading, so long as everything can be run independently, it'll be so much faster

[ProgrammingFun]

mirhagk @ Tue May 24, 2011 11:17 pm wrote:

Apparently batman arkym assylum built a physics engine (to animate a wall of bricks exploding) completely on the GPU. I have a book that discusses creating physics engines and such on a GPU. THe biggest benefit is the extreme multi-threading, so long as everything can be run independently, it'll be so much faster

That comment was incredibly well-timed: http://www.engadget.com/2011/05/25/cray-xk6-supercomputer-smashes-petaflop-record-humbly-calls-its/

Engadget @ http://www.engadget.com/2011/05/25/cray-xk6-supercomputer-smashes-petaflop-record-humbly-calls-its/ wrote:

Sure, IBM's ten petaflop supercomputer may sound impressive, but Cray can do you five better -- the outfit just announced the Cray XK6, an upgradable, hybrid supercomputing system capable of more than 50 petaflops of computational muscle. Powered by Cray's Gemini interconnect, AMD Opteron 6200 processors, and NVIDIA Tesla 20-Series GPUs, the XK6 system blends x86 and GPU environments with the firm's own flavor of Linux. The folks at Cray won't resort to bragging, however -- they're humbly declaring the machine to be the first "general-purpose supercomputer based on GPU technology," and not, as they put it, a stunt to place high on any Top 500 lists. Suggestive, aren't they?