Building a Desktop Computer for Math/Physics

[DrZoidberg]

So you decided on an expensive high performance build afterall. In that case why not put an M2 SSD in there. Those are 4 to 5 times faster than SATA SSDs. So maybe a 512mb M2 SSD for the system drive plus a 1TB SATA SSD for data storage.
Of course it's total overkill but it's fun to have.

 

[RJLiberator]

'Twas just getting to be too much. Needed to stop somewhere. Room for improvements down the line :D.

 

[Aufbauwerk 2045]

Hi all,

Over the summer I have been working hard to net some extra cash to acquire a desktop computer for education based purposes. Physics research, modelling, and coding is what to be expected. I am hoping to have this computer be my main setup for the end of undergraduate (2 years) and graduate school, so I am looking at around a 6-8+ time frame for relevant usage.

You may want to consider parallel processing using the GPU. Check out https://developer.nvidia.com/cuda-gpus and also this story about scientists using the GPU instead of a supercomputer. http://spectrum.ieee.org/tech-talk/computing/hardware/use-a-gpu-to-turn-a-pc-into-a-supercomputer.

 

[Vanadium 50]

Let me repeat what I said:

It is true that you can get great calculational performance on GPUs, However, you need to be using code that is specifically written for GPUs. If you aren't writing the code yourself, or using a program specifically written for running on GPUs, it's not likely you will get better performance. You're better off saving your money or spending it elsewhere.

 

[vr0nvr0n]

Hello!

Here is a site where I put together computer parts and the site calculates the best deals and any compatibility issues: http://pcpartpicker.com/list/mjRdxY

This is actually my first build, and it is based on the "Moderate Gaming Build" tutorial found on the same site referenced. However, I feel comfortable with this since I am not doing any hard core gaming; I am doing more software testing/development.

If wanted, I can follow up with my experience for this what I find to be reasonably-priced build!

 

[Dr Transport]

Let me repeat what I said:

It is true that you can get great calculational performance on GPUs, However, you need to be using code that is specifically written for GPUs. If you aren't writing the code yourself, or using a program specifically written for running on GPUs, it's not likely you will get better performance. You're better off saving your money or spending it elsewhere.

I looked at using GPU's when I was in industry a few years ago, they were fast but not nearly as accurate numerically as we had wished, that is why we went towards pc clusters...

 

[Borek]

I looked at using GPU's when I was in industry a few years ago, they were fast but not nearly as accurate numerically as we had wished

They mostly use 32 bytes floats, but some of the newer ones can work in double precision (64-bits).