Question: GPU based parallel processing in genomics tools
2
gravatar for hafiz.talhamalik
29 days ago by
Pakistan
hafiz.talhamalik210 wrote:

I am looking for new system for my lab. and I want to know that do typical genomics alignment assembly and annotation tools (specifically tools like BWA, velevet, spades, unicycler, Qiime2) support GPU based parallel processing ?? one of my friend told me to go for good GPU rather than buying new workstation. Any suggestion is highly appreciated.

softwares ngs assembly • 259 views
ADD COMMENTlink modified 29 days ago by JC9.1k • written 29 days ago by hafiz.talhamalik210
1

Related: Using Graphics Processing Units (GPUs) in bioinformatics
(and many more threads)

ADD REPLYlink written 29 days ago by WouterDeCoster42k
1

One more thing If someone could elaborate it for me. what is preferred option Workstation of rack based server ?

ADD REPLYlink written 29 days ago by hafiz.talhamalik210

Functionally makes no real difference, so it depends on your own requirements for space (if you havent got access to racks, you cant have a rack mount!)

Our lab has both. The workstation just sits in the corner of an office. The advantage of the rackmount is that IT services do contribute to keeping it alive.

ADD REPLYlink written 29 days ago by Joe15k

Depends on your situation. Workstation for us works better because we are completely in charge of it, not IT department. But rack mounted might get you better specs, and the noise/heat isn't in the office.

ADD REPLYlink written 29 days ago by SaltedPork100

TL;DR - almost none of them

ADD REPLYlink written 29 days ago by Joe15k
5
gravatar for Carambakaracho
29 days ago by
Carambakaracho1.9k
Switzerland/Basel
Carambakaracho1.9k wrote:

As far as I know GPUs don't play a big role in bioinformatics. Most software is designed to run on Linux servers, reachable via commandline only - at least in my company the focus is by far more on CPU than GPU. Especially the tools you mention, BWA, SPAdes, Qiime2 are mostly CPU and/or memory intensive. A combination of good CPUs and substantial memory will be more useful than spending money on expensive GPUs.

ADD COMMENTlink written 29 days ago by Carambakaracho1.9k
1

In addition to this, much older software that is able to utilize CPU multithreading reaches peak throughput at 8-12 CPU threads, with diminishing returns after that. So if your workload is small, you might not even see much increase in performance with more CPUs. Unless you start moving into running multiple samples in parallel with multithreads, in which case your RAM considerations will need to be much higher as well. Eventually you get to the point where it may be a better option to just move to cloud computing such as AWS or GCP.

ADD REPLYlink modified 29 days ago • written 29 days ago by steve2.4k
3
gravatar for colindaven
29 days ago by
colindaven1.8k
Hannover Medical School
colindaven1.8k wrote:

There are a few GPU projects, eg here https://developer.nvidia.com/Clara-Genomics (interestingly minimap2 is getting the gpu treatment)

and there have been a few in the past, eg. soap3-dp, bwagpu, some accelerated GPU blast-like programs, etc, but in general there hasn't been great uptake in the community.

CPU aligners tend to be fast enough and GPU is still tricky to program for and more tied to rapidly changing hardware. I've had more failures than successes trying to get GPU programs to work.

Oxford nanopore's technology does use a lot of GPUs for basecalling, polishing and now alignment.

ADD COMMENTlink written 29 days ago by colindaven1.8k

ok got that. I have also seen a [GPU Based Aligner]: https://nvlabs.github.io/nvbio/nvbowtie_page.html what about their acceptance in the scientific community ??

ADD REPLYlink modified 29 days ago • written 29 days ago by hafiz.talhamalik210

Acceptance still pretty low in my estimation. We have a CPU cluster and no GPUs, but we are working towards getting a GPU.

For image analysis and neural nets GPUs are great of course. If I were you I'd definitely get the workstation.

ADD REPLYlink written 29 days ago by colindaven1.8k

Thank. Currently I m not into image analysis and neural networks. so I will go for workstation

ADD REPLYlink written 29 days ago by hafiz.talhamalik210
2
gravatar for chrchang523
29 days ago by
chrchang5235.8k
United States
chrchang5235.8k wrote:

GPUs have a place in bioinformatics, but for now it's a small one since there's still so much underexploited CPU power on the table. I'd expect the workstation to be a better investment unless you already know you're bottlenecked on some computation where today's best GPU-exploiting software beats the best non-GPU software by more than an order of magnitude.

ADD COMMENTlink written 29 days ago by chrchang5235.8k
2
gravatar for JC
29 days ago by
JC9.1k
Mexico
JC9.1k wrote:

I agree on what is said before, GPUs are nice but is not commonly used in Genomics because most software doesn't need it or the cost/benefit is not great. Protein and numerical simulations fit better for GPUs. I have a Tesla in one server, I have never used it for any analysis in the past 3 years.

ADD COMMENTlink written 29 days ago by JC9.1k

Thankyou. What about the OS preference, which one is preffered ?? Ubunutu, centos or someone else ?

ADD REPLYlink written 28 days ago by hafiz.talhamalik210
1

Depends how much Linux experience you have. If you want an 'out of the box' experience, then go for Ubuntu.

ADD REPLYlink written 28 days ago by Joe15k
1

yes, I second Ubuntu, also fedora is good

ADD REPLYlink written 28 days ago by JC9.1k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1398 users visited in the last hour