Apple M1 processor for bioinformactics

Question

[1215045934]

Hi everyone,

I am thinking about buying a new laptop for bioinformatics. The new Macbook Pro with M1 processor looks really powerful. But does anyone know if there is any compatible issues for bioinformatic softwares with the Apple M1? Is there any potential issues?

Thanks a lot

[Dunois]

Following up on GenoMax's comments, just to mention some non-Apple alternatives: there are some fairly decent 14" machines (e.g., Thinkpad L14, Elitebook 845 G7, Tuxedo Pulse 14 Gen 1, MSI Modern 14) with the Ryzen 4000 4/6/8 core processors that go up to 64GB of RAM that should be a good compromise between portability, performance, and price.. (Heck there are even some good 13" machines with the Ryzen 4000 series processors and upto 32GB of user-serviceable memory.) None of these are going to beat the M1 in terms of performance, of course.

[5heikki]

They most certainly beat the M1 (in current Apple laptops) in performance in tasks where e.g. a file larger than 16GB is read into memory for whatever tasks. That is assuming that they have been configured with more than 16GB of RAM..

[vytarasov]

There are reviews where it is shown that there is no performance drop in 8 GB MacBook versus 16 GB. The same can be apply for bigger files. I have not seen an explanation why. (Who knows how this files have been read, completely at once or part by part.)

[i.sudbery]

I suspect once you are over a certain amount, more memory is not about more performance, but what is and isn't possible.

You'll never map to the human genome with STAR on a 8GB for example.

Its not a case of how fast it runs - the index is just bigger than 8GB (24GB I think) and needs to be all held in memory at once.

[GenoMax]

Likely because of the phenomenal memory bandwidth available to M1 cores (From AnandTech):

A single Firestorm achieves memory reads up to around 58GB/s, with memory writes coming in at 33-36GB/s. Most importantly, memory copies land in at 60 to 62GB/s depending if you’re using scalar or vector instructions. The fact that a single Firestorm core can almost saturate the memory controllers is astounding and something we’ve never seen in a design before.

[dante]

Hi Dunois and all,

In addition to the models listed above, are there any laptops good for bioinformatics in 2023? My 2011 MBP is dying soon. I basically use laptop for lightweight analyses using python or R.

Thanks.

[Dunois]

This would depend on your budget. Perhaps an Apple M1 machine might be the best replacement given you are probably well-invested in the Apple ecosystem now?

Answer 1

[GenoMax]

Let me start with limitations since they can be significant depending on what you intend to use this machine for. You are currently limited to running macOS. Projects to port linux have been announced but they will take time and may be of limited utility in short term.

1. Memory
   Since the processor uses DMA architecture, memory is included on the same die as the CPU/GPU. While it provides fast access, it is limited to a maximum of 16GB. This will preclude using any software that needs more memory than 16. Upgrade to 16G is significant cost.
2. Form factor
   AFAIK M1 has been released in a single hardware version. M1 in the form factor of mac mini is likely to allow more sustained high frequency compute since macbook air does not have a fan and will throttle processor, if you are doing sustained compute on it.
3. Storage
   Unless you plan to use external thunderbolt storage enclosures, you are limited by internal storage. You are going to pay a premium to get a TB or two (max). You can only get that during the build since you can't add more storage later (internally).
4. Xcode
   There is a version of Xcode for M1 but some things may still be missing from open source software. These will eventually catch up. So if you need to compile something you may run into odd glitches.
5. External monitor
   You can connect only one. May not be a problem but may want to note.
6. External GPU
   Until AMD/NVIDIA make user mode graphics drivers available there is no possibility of using an external GPU.
7. Docker
   No support for docker at this time. Coming in future. Docker has released a version that now supports M1 processor (Apr 2021)

Plus points:

1. Performance
   M1 is running rings around Intel's offerings as you have already seen.
2. Battery life
   For casual use battery life is really great. Truly an all day machine if you are simply going to ssh into central/cloud compute for serious work.
3. Rosetta2
   Rosetta2 is making things seamless. So far conda runs fine (and thus much of the bioinformatics software should). R/Rstudio works as well. There does not seem to be much of a performance hit. Again I have not done any serious compute.
4. User experience
   Flawless. Indistinguishable from an Intel mac. macOS feels/works identically on both architectures. Universal binaries will work.

[Alex Reynolds]

Another limitation is no Boot Camp support for Windows or Linux. This may change in the future, but it might not.

[Alex Reynolds]

A port of Linux compatible with the M1 is being worked on. More information here: https://www.patreon.com/marcan and here: https://asahilinux.org/

[GenoMax]

Linux has been successfully ported to M1: https://corellium.com/blog/linux-m1

[alanh]

Efficient, fast graphics for Linux are likely going to lag quite far behind base Linux OS support. My understanding is that Apple does not provide the GPU documentation that would make this easy.

[5heikki]

I wouldn't do any serious bioinformatics on a laptop ever, although certainly some things can be done with very limited specs. So basically for me laptop only serves as a way to access a server/cluster and that's pretty much it. In regard to this, IMO the three most important factors are screen size/quality (minimum 15" and preferably 4k resolution), a good keyboard (e.g. good typing feeling, correct layout (MacBooks and e.g. ThinkPads fail here)), and portability (I don't want to carry around a 5 kg monster). My current work laptop is Dell Precision 5550 (basically 15" XPS but better specs). It's almost perfect. One annoyance is the touchpad which is just too large/sensitive (and also mushy if you're the type who actually wants to click a touchpad), but maybe it will be fixed with some future software update..

[GenoMax]

The point I was making was for all intents and purposes a M1 laptop behaves just like any other PC/Mac laptop. If whatever you are doing fits in the hardware/software limitations, it should work more or less seamlessly.

Compared to the G5 --> Intel transition, Apple did their homework and made sure Rosetta2 works well. Having hardware they designed/understand made this feasible.

Many commercial bioinformatics programs have all worked without issues (e.g. SnapGene).

We are already learning about future M chips that can have as much as 32 cores and should be able to use more RAM. What the cost would be is going to be the more important question.

Answer 2

[alanh]

This is going to vary depending on the tools you want to use, at least at present.

There are a few Bioinformatics tools that are accelerated by some of the Intel AVX instructions. For example, GATK's HaplotypeCaller and Mutect2 both use PairHMM which is accelerated by at least an order of magnitude when run on AVX-capable CPUs, vs non-AVX fall-back mode on Intel hardware (in my experience at least).

I don't have an Apple device with an M1 chip to run comparison benchmarks, but AVX/AVX2/AVX512 are specifically listed as not supported by Rosetta2

For some other tools, you might be limited by available RAM. For example, the STAR RNA aligner loads the indexes and genome into RAM, and needs 10x the genome size or around 31GB for human and mouse genomes. As of March 8, 2021, the maximum amount of RAM that you can get in an M1 Mac is 16 GB, and the RAM is not expandable (it's integrated on the same chip as the processor).

Personally, I'm excited for the future, at least on a compute-per-watt basis, but software will need to catch up, and future Apple Silicon chips will probably provide more RAM.

[Kermit]

It's almost 2023 and the M2s are out. Does anyone know if HaplotypeCaller supports Apple Silicon yet?

Answer 3

[vytarasov]

IMHO, such powerful processor can not be ignored by bioinformatics, especially because coming in 2021 the next generation of M processors with more cores. At such low energy consumption they are the best for such tasks as reads assemblies. It is simply a future for bioinformatic! Intel RIP. And, I am not even talking about how perspective Swift for bioinformatics :). Just listen the Chris Lattner about swift to realise its power. More scientific apps written in swift is a near future.

[GenoMax]

such powerful processor can not be ignored by bioinformatics

That is unlikely but at this time above limitations are significant especially for working with really large data. Apple is formally out of server market (discounting use of mac mini farms that companies offer and they will indeed offer M1 mac Mini at some point) at this time.

If Apple ever decides to market real servers again (multiple-socket M processors) then things can get interesting indeed. Intel needs to start worrying some but they are not likely to be written off anytime soon. An opening salvo has been fired across their bows.

[vytarasov]

Here is a link about ARM server processors coming from the same Apple origin. https://www.techspot.com/news/86874-nuvia-raises-240-million-make-arm-based-server.html

[GenoMax]

ARM based servers are already available. In fact Amazon offers them right now via AWS. Tight integration of M processors and OS makes them special.