recommendation on workstation

[liuwensui]

I'd like to buy a workstation to do statistical computing with 64-bit R/Octave/Python.
what hardware will you recommend? what OS?
Any suggestion would be highly appreciated.

 

[alain]

Any flavor of Linux 64 Bits. an Intel Nehalem processor (multiples are probably better). A lot of Ram (8 GB are inexpensive now) and a nice Tesla C2070 GPU computing for extra performance

 

[yzia]

Any flavor of Linux 64 Bits. an Intel Nehalem processor (multiples are probably better). A lot of Ram (8 GB are inexpensive now) and a nice Tesla C2070 GPU computing for extra performance

Agreed. The Tesla is expensive however, so only if you plan on coding for the GPU should you consider it. You can always add it in later.


If you dont want to build it yourself, I'd go down the Dell Precision series route. Solid desktops there. What's your budget?

 

[liuwensui]

Agreed. The Tesla is expensive however, so only if you plan on coding for the GPU should you consider it. You can always add it in later.


If you dont want to build it yourself, I'd go down the Dell Precision series route. Solid desktops there. What's your budget?

i am thinking of levnovo thinkstation.

 

[Andy Nguyen]

You can always run VMWare to host other OSes if there are critical applications not ported over to your main OS.
Any off the shelf computer these days can do number crunching. You can also rent CPU cycles from Amazon server farm if the need arises.

 

[joel_b]

Those GPU's are pretty crazy. If you got an expansion chassis you could run it through a macbook pro via the expresscard slot, no? I wonder how reliable that would be :p

 

[liuwensui]

Any flavor of Linux 64 Bits. an Intel Nehalem processor (multiples are probably better). A lot of Ram (8 GB are inexpensive now) and a nice Tesla C2070 GPU computing for extra performance

alain, why you recommend linux? what are advantages of linux over windows in statistical computing?

 

[tobias elbert]

Any flavor of Linux 64 Bits. an Intel Nehalem processor (multiples are probably better). A lot of Ram (8 GB are inexpensive now) and a nice Tesla C2070 GPU computing for extra performance

Alain, just wondering why you'd recommend a fairly powerful GPU for work that is not very graphics intensive? Excuse my naivity on this topic, but can you redirect eg threads to be run on the GPU instead of (or on top of) the main CPU(s)?

If so, how would that work in eg .Net? Does the .Net framework automatically recognise the GPU and allocate threads to it?

 

[Eugene Krel]

The google is your friend: http://gpgpu.org/

you can also use quantnet search feature.

 

[alain]

Alain, just wondering why you'd recommend a fairly powerful GPU for work that is not very graphics intensive? Excuse my naivity on this topic, but can you redirect eg threads to be run on the GPU instead of (or on top of) the main CPU(s)?

If so, how would that work in eg .Net? Does the .Net framework automatically recognise the GPU and allocate threads to it?

GPUs are being used heavily in general computing now, mostly in fields where matrix operations are involved. Look at the NVIDIA site.

About .Net, I don't know much. However, last time I checked, the .Net framework only allow 25 native threads with the implementation that comes with their library. Check the internet. That problem already bit a group in my company in the ***.

---------- Post added at 08:33 PM ---------- Previous post was at 08:25 PM ----------

alain, why you recommend linux? what are advantages of linux over windows in statistical computing?

for one, it is free. It is widely use in the scientific community. It has massive support so any problem you have, you will probably find a solution really fast.

 

[tobias elbert]

Wasn't aware of that, but that's quite interesting.

I know about the problem as to how the native .net threading library manages multithreading - the developers working next to me have been cursing it for quite a while. We switched to some opensource .net library (I think) that handles multithreading for us.

I am not a developer, but find this http://gpgpu.org/ interesting nevertheless. Cheers!

 

[Andy Nguyen]

Build Your Own
From a link posted by cgorac on the NVDIA CUDA thread

I'm a DYI guy so that's something I'm tempted to build in the future unless their certified station becomes cheap commodity like a Dell.