While there are a lot of duscussions about this – I haven't yet found any comprehensive, real world test. Does anybody have knowledge of existing test that go beyon simple benchmarking?
Thanks
Thanks
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.
Comprehensive tests of M1 Max 32gb vs 64gb unified memory
- Thread starter Nightjar84
- Start date
- Sort by reaction score
Use your imagination 
Interesting results here:
Not very systematic testing but interesting watch: yes, using more ram if it's there but no noticable performance impact.
Interesting results here:
Not very systematic testing but interesting watch: yes, using more ram if it's there but no noticable performance impact.
I would also like to see this too - I saw MaxTech’s video on ram, but it was a flawed test IMO. At least for Logic, he was using a CPU stress test and not something that would stress RAM.
I wouldn’t expect the amount of RAM on the M1 Max to have any performance difference. Both the 32 GB and 64 GB have the same 400 GB/s bandwidth. It is different than the M1 Pro 32 GB which has “only” 200GB/s. So unless your application needs more than 32 GB, expect the performance to be equal on the M1 Max.
What some of the few tests around have shown: The applications use more if it's there. But the question is: will there be a significant performance difference and it yes, under which circumstances?
How is this a hard question to answer? If you have working sets >32 GB you will need >32 GB of RAM, either to even be able to load them at all, or to keep your machine from actively swapping while working with them.
If you only use >32 GB of RAM for various separate applications but where each working set is significantly smaller than your available RAM (think running a whole bunch of apps and never closing the files you're working on), you might not even notice an upgrade to 64 GB of RAM in regular use as the system doesn't have to swap data out to disk and back into RAM while you're actively using it (what we used to call thrashing in the days of spinning rust).
As a a few tests (also with the m1 air) have shown, this 'old' thinking might not apply how the m1s work. Tests are needed – and not just short render tests, but also long, intense renders.
Said it better than I could.
It's not about stress tests. Though if your total RAM needs are >32GB but each set is smaller, you might appreciate the less swapping of data to and from the SSD. Will this matter in the overall average lifespan of the SSD? Maybe? It is something to consider though.
Data is data, there's no "new" or "old" thinking to argue about. If the software you use needs more RAM than is available in your machine it has two ways to go about it: Fail to load it, or swap parts of your data between disk and memory as required. The latter will be dozens of times slower than operating on the data in memory (up to 7 GB/s vs up to 400 GB/s), which you definitely will notice. The only explanation for cases where people don't notice a slowdown on the M1 platform is that they work with sufficiently small datasets and never actually needed the additional RAM in the first place.
I‘m interested in the battery life penalty of having 64gb vs 32gb on the m1 Max. I suspect it to be greater then expected and this could be the biggest part of the reason that some reviewed m1 Max models had shorter battery life (I think this is a topic of concern). I found it surprising that reviewers would only relate this to the GPU while ignoring the memory. While most reviewed m1 Max MacBooks had 64gb on board.
Last edited:
Same, especially since as I type this the 32GB version would arrive before Xmas but the 64GB pushes out 2 weeks further
