I just obtained an M2 Max Mac Studio 38 core GPU 2TB SSD and 96 GB RAM for scientific computing. I've never really known for sure how the various benchmarks (e.g. Geekbench) you see on youtube etc translate into real world computational speed in data processing, modeling and simulations. Up to now I relied mostly on 6-core trash can Mac Pros for my work. They've been a reliable workhorse over the years but now need something faster. I knew the Mac studio would be faster but was unsure how faster it would be for my purposes. As a quick initial test, I just ran a simple Python for loop printing out numbers from 0 to 10,000,000. It took the Mac studio 6.3 seconds to complete this. The trash can Mac Pro took 33 minutes! I'm hoping this speed ratio is borne out when running more sophisticated code but this simple test has blown my mind.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Studio M2 max vs Trash can Mac Pro
- Thread starter stardog555
- Start date
- Sort by reaction score
Interesting comparison, looks a bit “biased” but I’m sure it isn’t. What about bc on terminal, say a huge integer raised to a huge power?
Hopefully not biased but was a really quick test because I was curious about the differences. I should have added the for loop was run within the Spyder environment, but that is what I usually work with. When I run the loop in the terminal instead I get some curiously different results: it takes about 14 seconds to run on the Mac Studio (so longer than on Spyder), but the trash can takes 40 seconds... so using terminal trash can much faster than the same job on Spyder, and Mac studio is slower. So speed differential on terminal not as astounding but still about 3 times faster. Can you suggest an integer value and power to try?
Another simple test: initiate a 3-D numpy array of ones of size (1000 x 1000 x1000). Not much difference here between running on terminal and on Spyder. Mac studio : about 0.7 seconds: Trash can about 3 to 3.9 seconds. Both computers crash if we initiate a 10000 x 10000 x 10000 array due to exceeding RAM capacity.
I may open MP 6,1 later and check, but on a Mini M1 base model, using the bc on the terminal to raise 123456789 to 1234567 took 44.5 seconds. Six-digit exponents are not noticeable task, time-wise.
The later test you've tried is more indicative, I guess.
The later test you've tried is more indicative, I guess.
Just an update on this... I've now implemented a code that carries around a couple of thousand MCMC fits to data sets in sequence. On the trash can this would take me in the order of a week (5-7 days) to complete, often with RAM crashes and having to restart etc. and splitting the task between two trash cans. The Mac studio does this in around 1.5-2 days (depending on the number of fits) using a single spyder console. It is possible if I split the task over multiple consoles it would complete faster still. What's more the machine is stable against crashes... memory is not under pressure. I think getting the maximum 96GB unified memory helps here. So in summary for this task, I find a real world improvement of around 1/3rd the computing time of the trash can(s) but also with added stability against crashes.
Good question... I don't know exactly but will try to check this down the line. I can say the Mac Studio is extremely quiet to the point where I had to check to see it was on and computing away which it was.