Hi Alessio.

- do I need at least 2 graphics cards (one for graphic purposes and the other for computation)?

if you do not run graphic applications NO.

Bedpostx_gpu needs a GPU completely free of any graphic task. When you run bedpostx, you should be in command mode (no graphical sessions).

On a centOS machine it is easy to do it.

- my need is to process single patient dataset, 50 slices, 64 gradients with bedpostX with nfibres=2.. which gpu card could allow me to stay in a computation time of 1 hour? Could a commercial

GeForce GTX 1050 or 1060 hit the score?

You can definitely use a commercial GPU and achieve accelerations of ~150X compared to a single CPU core. So, I think you can compute each dataset in less than 40 minutes. However, I would choose a GPU with better performance for double precision operations. These ones are really good in single precision, and most operations in Bedpostx_gpu are in single precision, but ~10% operations in Bedpostx_gpu are performed in double precision. If you have to choose between these 2, go for the 1060 6GB (twice GFLOPS in double precision compare with 1050).

- My aim is to set a machine like this and use it remotely from my Mac.. it could be a good idea?

I do not see any problem doing this. You can use it as a server.

Moises.

On 22 December 2017 at 03:18, am <[log in to unmask]> wrote:

Hi everybody,

usually I use macOS, but it has been arrived the time to reduce bedpostX computational time.

Where I worked 2 years ago I exploited sge with some multicore apple machines. Since I cannot use these machines anymore, I need to find an alternative way.

After some reading, I realize that the best option could be a centOS machine with dedicated nVidia card. So, I was thinking about the hardware needed.. can you help me with some advice?

I thought that could be enough to provide a desktop computer with a decent CPU, the maximum possible RAM, and at least one decent gpu nVidia card..

So my doubts are:
- do I need at least 2 graphics cards (one for graphic purposes and the other for computation)?
- my need is to process single patient dataset, 50 slices, 64 gradients with bedpostX with nfibres=2.. which gpu card could allow me to stay in a computation time of 1 hour? Could a commercial
GeForce GTX 1050 or 1060 hit the score?
- My aim is to set a machine like this and use it remotely from my Mac.. it could be a good idea?

Thanks in advance for every help

Alessio