Forum:M1 Pro/Max and future

Question

[fredhstein]

Apple's M1 Pro and Max appear to lead in performance per watt due to architecture and TSMC's process. Assuming another large step up in a year by both Apple and TSMC, how much of the current bioinformatics software will be ported to and optimized for these Mac systems?

Should we think about this is terms of market size for Apple, or in terms of advancing the field of bioinformatics?

If this is covered elsewhere, please route.

Answer 1

[GenoMax]

how much of the current bioinformatics software will be ported to and optimized for these Mac systems?

There is no critical need to optimize software since most of it works via conda already. Headroom on the new chips should provide a good amount of performance boost as is. Apple already makes it possible to create universal binaries so the onus is on developers to do the needed effort to make them for software that is to be made into apps. Bioinformatics is probably a small market for Apple so I would be surprised if they make any additional effort. Intel only recently worked with Heng Li to optimize bwa.

[dariober]

Hi- you say: most of [the software] works via conda already. I've become an enthusiast of the conda ecosystem despite its glitches, so glad to see that comment. Do you have any reference or perception you could share to back up the statement that most software is installed via conda (I guess that's what you mean)?

[Istvan Albert]

we are basically running the entire terminal in the so-called Rosetta emulation mode, hence all code compiled for Intel will run

notably, even in emulated form the code runs faster on M1 than on the original Intel CPU

the above does imply that native M1 code might not work in the same terminal as non M1 code

(I have not tried this yet, since there is no bioinformatics software compiled for M1 code yet that use)

[GenoMax]

Your comment led to me to hunt for some information. The native terminal on M1 mac is actually arm64 so it is not running in emulation mode. But one can set a terminal up to run in Rosetta mode as noted here. So any intel software can be installed using this special terminal.

Answer 2

[Alex Reynolds]

While most tools will probably do with a simple recompile to a new target, it might also be worth taking a look at benchmarking of genomics tools on the cloud. Cost or performance could also drive efforts to target (and optimize) existing tools for this new platform.

For example, AWS offers Graviton2 instances — which are ARM-based, like M1 — that run bwa and other tools more efficiently than on x86-64 architecture, in terms of overall cost and performance.

It may be useful to repeat those kinds of analyses once Amazon offers "Apple silicon"-based Mac machine instances, in order to measure efficiency of key bioinformatics tools on Graviton2 and M1*.

Answer 3

[Istvan Albert]

I think M1 chips are a huge step forward in Bioinformatics, but Apple would also need to offer more RAM to their stock systems

With 64-128GB of RAM, the majority of bioinformatics pipelines would be readily parallelizable and runnable an iMac.

At that point that opens a huge market for life scientists and their teams running their analyses locally rather than paying the cloud fees that may start small but can accumulate quite fast.

[GenoMax]

Problem is because of the design of Mac CPU's getting that much RAM onto a physical die is going to make these CPU expensive. Students (people paying for these from their own funds) are already balking at the price of a M1Max MBP.

That said rumor mill says that next M1 chip has 20- and 40-core variations are planned, comprising 16 or 32 high-performance cores and four or eight high-efficiency cores likely destined for Mac Pro. Such a machine is bound to be coming in at a price point where it may make sense to get a proper intel based server since that would allow future upgrades with standard components.