Amd Or Intel Based Workstation For Bioinformatics?

Question

[Leszek]

I'm planning to get some new workstation for bioinfo work. It's going to be mostly NGS data processing.
I'm hesitating between AMD (Opteron or FX) and Intel (Xeon or i7). The former is much cheaper for 2x more cores, but much more power-hungry. For some bioinfo tools (ie. bowtie2) I noticed that 2 Opteron cores = 1 Xeon core at similar clock speed. So investing more money in Intel CPU makes sense, no?
What's your opinion?

[Alastair Kerr]

I think that memory (and storage) is more important just now that cpu-cores. If you get a dual processor machine this will allow more memory and you can get programs like STAR running in minutes where tophat would take hours. As STAR needs about 35GB of continuous memory to store one genome your min spec would be about 64GB. Ideally you want much more. 1TB is not a luxury for denonvo transcriptome assembly

[lh3]

1TB for transcriptome assembly is scary given that SGA can assemble a human genome in 64GB. Even the more memory hungry allpaths can do that in 512GB. Transcriptome assembly is of a much smaller scale. Back to the question, Intel is indeed faster than AMD in general, but I have not seen 2X difference even between a recent 3.5GHz Xeon X5690 and a much older 2.3GHz Opteron 8376.

[Alastair Kerr]

Human is only diploid though, I have seen at least one researcher using most of 1TB for a single run. Agreed on CPU performance differences. When memory is not an issue, most of our bottlenecks are IO, including internal bus and disk write.

[lh3]

I know transcriptome assembly poses new challenges, but the amont of data is nothing compared to whole genome assembly. Someone using 1TB RAM does not mean we should use 1TB. When buying hardware we should ask what is the requirement of a typical mainstream tool, rather than the requirement of a badly engineered in-house pipeline.

[Alastair Kerr]

I do agree in the most part as the 1TB instances (actually is was ~0.8TB) are rare and extreme and pre-processing the data is much more sensible (if there is sufficient skill and knowledge of the system to do so). The point I am trying to make though is that moving the data within the computer is the slow point with our big data, rather than the speed of a particular core. i.e. Keeping data in memory is usually the fastest speed up.

[Raygozak]

These kind of questions always seems to me whether you are deciding to use one brand of calculator against another to sum two numbers together.

[Leszek]

I wish it's that simplistic...

[Steve Moss]

Unfortunately, Intel have pushed ahead of AMD in terms of performance. It isn't a matter of brand, but which system can provide you with the best computational performance for your money.

Answer 1

[Josh Herr]

I have no personal opinion right now on AMD and Intel. We have 4 and 8 core machines of AMD Opteron and an i5 and i7 Intel machine and I can't tell much difference between the two when considering comparative processor speed. We don't have a Xeon machine. I would seriously just address price.

I think a lot of reasons to buy a machine come from other specs and things such as which operating system you'll put on it and ease of dealing with software. Here's some other somewhat related posts to discuss hardware in the last few months that may help your decision, but may or may not be pertinent to workstations in general:

Storage For Miseq In-House

Medium Sized Data Backup Strategies

Which Operating System Do You Prefer For Bioinformatics?

Linux Distros Best Suited For Bioinformatics?

Big Ass Servers & Storage

Storage For Miseq In-House

[Leszek]

thanks Josh! I will go over your links.

Answer 2

[Adam Cornwell]

Per-core performance is higher on the Intel side of things. Per-core price for AMD systems will almost always be lower, so if you're going to run something that parallelizes well, you could still be better off with more cores for the money on an AMD system. Memory performance is probably pretty similar for both now. As already mentioned, the more memory the better, so you can always go AMD for CPU savings and plow that cash into more memory. If you will be running anything that needs be fast but only runs on one or a couple of cores, then Intel is the way to go. When looking at benchmarks for AMD CPUs from the last few years- look up the CPU model if possible- the original "bulldozer" models that introduced their current architecture were not so great performance-wise. The newer "piledriver" models are markedly improved.

[Leszek]

thanks Adam. I went through plenty of benchmarks. Unfortunately, reviews are focused mostly on performance in gaming (sick!) and I was curious of anyone testing Intel/AMD systems in bioinfo work. I have noticed "piledriver" improvement both in performance and energy consumption over "bulldozer". But still it's far behind i7 in both performance and energy consumption I think. And difference in price between them can melt quickly if you consider electricity bills. Don't get me wrong, I'm AMD fan for many year:)

Answer 3

[Rahul Sharma]

Hi,

In our group, for bioinformatics jobs we have clusters with Xeon processors. Since most of the NGS data processing tasks demand for the high computation and long walltime jobs. According to my opinion Xeon processors work well in these situations.

Cheers

[Steve Moss]

Is there much difference between Xeon and Core i7 processors these days? Would be interesting to see some benchmarks!

[5heikki]

Replying to a very old post.. dual socket setups are Xeon only. Anyway, we're living interesting times and it sure seems like AMD's threadripper CPU's are the obvious choice right now..

[Eli Korvigo]

It largely depends on the things you do. If your software is heavy on MKL/BLAS/LAPACK, then AMD is a waste of money.

[Eli Korvigo]

The main difference is ECC memory. It might not be very important for regular consumers, but it is important for anyone with long-running computations in mind.