Building a linux machine for MiSeq

[nermze]

Given virtually unlimited budget, what type of machine would you recommend for analysis of MiSeq data, and genome assembly? We are mostly working on Tuberculosis.

We are looking at a HP workstation with 128Gb-512Gb of memory, and 20-24 core CPU.

Is this overkill for bacterial DNA? Will double Xeon CPUs work great under Linux? How much storage should we aim for?

Any input is appriciated, thx in advance.

[GenoMax]

If you have the budget go for the machine described with as much RAM as you can afford. MiSeq runs themselves are not that large (~50G per run, even if you count raw flowcell data). Always get CPU's that are a one or two steps below the top one available (significant cost savings that can be used towards other components like memory/disk). Dual CPU's will work great under linux.

[nermze]

Ty so much, much appriciated!

Could I just ask 1 more question? Is there a preffered ratio of CPU Cores vs amount of RAM?

[GenoMax]

Since you are going to be working with bacterial genomes you may be able to get away with ~3-5G per core. For human genome size data one generally needs ~30G RAM. Since different programs work in different ways it would be hard to come up with a fixed recommendation though.

For de novo assemblies you will likely use SPAdes. Check their manual for recommendations on RAM/hardware from examples they list.

[nermze]

I see, thx for the quick reply.

We are starting with bacterial genomes, but are moving on to more complex organisms in the future. We have a Core I9 with 64Gb of RAM atm, and its working fine, but we want to future proof these new machines as much as possible. We will also be performing alot of runs in the start, there will be several other research groups. Am I right to assume, that although not necesarry, lots of cores/ram will generally speed up the process?

[GenoMax]

I see, thx for the quick reply.

We are starting with bacterial genomes, but are moving on to more complex organisms in the future. We have a Core I9 with 64Gb of RAM atm, and its working fine, but we want to future proof these new machines as much as possible. We will also be performing alot of runs in the start, there will be several other research groups.

A standalone machine will always have some limitations on how much you could do with it. If there is a need to make the resources available to other research groups simultaneously then you may be better off looking at putting a small compute cluster together. That could require resources (e.g. dedicated server room space, sys admin expertise) that you may have to acquire.

Am I right to assume, that although not necesarry, lots of cores/ram will generally speed up the process?

While technically correct that is a bit of oversimplification. At some point you will hit limits elsewhere (e.g. I/O, software) that may make some of those cores idle while they wait for the data to show up.

[nermze]

Thank you all for great input, I really appriciate it.

[Carcharodon]

What do you guys think about AMD's Threadripper 1950X for these kinds of builds? $1,000 for the processor with 16 cores/32 threads, can be overclocked stable at 4.0 GHz? Good value?

Not sure how it performs with Linux.

[GenoMax]

Threadripper is very new. On paper it does offer a lot for the money and may very well become the go-to CPU for bioinformatics in a few months.

It may be best to wait for some time to build a critical system with it considering some of the microcode issues Ryzen appears to have suffered earlier this year.