workstation hardware

[Berlinq]

Dear friends.
We are going to use Roche 454 FLX and SOLID3 platforms i our lab. I need suggestions about hardware configuration for the workstation to work with sequence data. One or two CPU's? How much RAM? What about GPU? I have read that there are some progs using GPGPU with CUDA. What about atistream, as ATI cards are much more powerful? Software: Win7 or Linux?
Thanks and greetings from Berlin

[maubp]

Most open source tools will work on Unix (e.g. Linux, often Mac OS X), but are less likely to run on Windows.

If you want to work with Roche 454, you will certainly need to consider running their closed source "Newbler" software - and that only runs on Linux. See:

http://seqanswers.com/forums/showthread.php?t=114

Whatever you are working with, you will need lots of memory. A 32bit machine limits you to 4GB of RAM, and is unsuitable but for the smallest projects (e.g. viral genomes). You will need a 64bit machine with lots of memory. How much depends on what work you'll be doing.

[Berlinq]

Thanks for the tip. Shall it be one or two (xeon) CPUs? The main application for us will be rna-seq.

[maubp]

I've not worked with RNA-seq so can't offer you any useful advice there (beyond try searching the seqanswers forum).

[westerman]

64-bit Linux. As much memory as you can afford. 1 or 2-CPUs doesn't matter since you'll be looking at slow over-night (or multi-day) runs in any case. Unless you have specific software that will use the GPU I wouldn't bother and would instead invest the money in memory.

[shurjo]

16G RAM at least

My project consists entirely of RNA-Seq data analysis and in my experience and on our Linux x64 cluster, 16G RAM is about a minimum requirement for most jobs with even one lane of Illumina data, if you want to keep runtimes on the scale of hours instead of days.

[The_Roads]

I'd agree, as much ram as you can afford and i'd say multicore as well. if you are going to be using two NGS platforms it sounds like you anticipate lots of throughput. if you have four or five users waiting to get on a single workstation that is taking tens of hours to do one persons work you will end up very frustrated. i'd look at how many users you realistically expect and then decide on a hardware option. in my personal experience when NGS projects start there is way more computing done than once a project is established and folks are familiar with how to see what they are looking for in their data, thus each new user/ or project creates more work than an established one. if you have good IT support you may want to try some sort of cluster. if you are going with workstations you may find that in terms of users you have a more content lab with two smaller workstations than one huge one. we use single quad xeon workstations with 32Gb ram running CLC and a shared linux cluster for in house bioinformatics. speaking of which software is another thing to consider in your budget if you want to go commercial

[Berlinq]

Thanks a lot to all for your tips. Really appreciated!