If swap is often needed, then it is quite good to know about this issue and how much it affects performance. Same for people who copy a lot of files around.
This is where I think the benchmarks are getting extrapolated beyond their domain of validity.
When people say the M2 SSD is "slower" they point to the Blackmagic disk speed test. Blackmagic is a company that sells video capture and editing hardware and software so they’re concerned with the ability to push and pull streams of data from storage to make the most of their products. There are a few things worth noting about how that test works:
- It writes between 1 and 5 GB in one continuous, sequential operation.
- The data is written procedurally. Uniformly distributed random numbers are generated by the CPU and pushed out to the file. Essentially no work is done to create them, they don't live in memory, they just flow from CPU across the interface to disk in one continuous, sequential stream.
- The 1 to 5 GB of data is then read back in one continuous, sequential stream. Nothing is done with the data, they are read and discarded.
This is a good way to test the maximum throughput of the interface because it gets all the other sources of delay and potential bottlenecks out of the way.
This is a theoretical maximum. It's a useful number to know because it tells you at what point your drive becomes the bottleneck. The only thing Blackmagic the company cares about in these results is that the 256GB drive doesn’t become the bottleneck until you are trying to work with 12k 60fps ProRes video of which this drive will hold just under 3 minutes worth.
So let's think about the two scenarios you raise, starting with swap. Swap is a completely different beast than streaming video. Memory is organized into 16kB pages. When pressure increases, the system compresses memory and keeps it in RAM. When pressure increases further it starts writing those pages out to disk starting with the ones you're least likely to need soon, if ever.
So while a video file can easily be 16GB, a memory page is a million times smaller. Throughput isn’t important to swap performance, latency is. The reason SSDs make swap feel so much more fluid than HDDs isn’t because of the really high throughputs, it‘s because of the really short latencies— the time to the first bytes back.
I’ve seen no good data on the latencies to the two drives but there’s no reason to expect the 256GB is slower and some reason to expect it might be slightly faster. Generally, when people talk about setting up RAID systems, the advice is “don’t use it for your boot drive”— the reason is because of latency issues. If you have a drive designed to push massive throughput, it doesn’t mean it’s good for virtual memory paging or even for launching applications.
There’s a ton of quanta involved that makes reasoning about this a bit hard. There’s the virtual memory page size, and the compressed memory page size, and the RAID stripe size and width, the NAND page size, and garbage collection batches, that all come into play along with the Apple RAID controller strategy. Data online about SSD RAIDs isn’t all that useful because Apple has certainly used the fact that they own the hardware and OS to tune all those parameters knowing they’re using this as a boot drive.
So I haven’t seen any tests that are able to isolate swap performance, but the Blackmagic tests tell us nothing, and it’s not impossible that the smaller drive might actually swap more efficiently.
Now there’s another kind of “swapping” that has been getting attention lately that I don’t really think of as swapping so much as caching— browser tabs. I think of swapping being a way to work with large active datasets. Browser tabs are mostly caching the page so the machine doesn’t need to rerender the whole page. In another thread someone was complaining that MR can sometimes take 1GB per tab, presumably because the ad service has lost control. Using that as a reference point, that GB page probably gets compressed by a factor of 2 when memory compression happens, so 0.5GB gets paged out to disk. If we assume it all gets paged out contiguously, which is still a big if, the write happens invisibly in the background but the read of this obscenely big tab will take 150ms longer with the smaller drive— literally a blink of the eye.
Ok, what about your second scenario: file copies? It all depends on what you’re copying to or from…
If you’re copying from the drive to the drive, it’s instantaneous regardless of the file size or drive size. APFS does a copy on write, so the file copy just creates another pointer to the file. So you’re not going to notice.
If you’re copying to a NAS over Gbit ethernet, it maxes out at something like 100MB/sec compared to the 1500MB/sec of the 256GB M2 drive, so you’re not going to notice.
USB 3.1 at 5Gbps will max out just shy of 500MB/sec, so you’re not going to notice.
If you have a dedicated 10Gbit link over ethernet or USB 3.2, then it will max out at about 1000MB/sec, which is still slower than the 1500MB/sec of the smaller M2 drive, so you’re not going to notice.
Macs don’t support USB 3.2 2x2, so those drives will max out at 1000MB/sec and you’re not going to notice.
Thunderbolt 2 drives could conceivably get you to 2000MB/sec (still not the full speed of the 512GB M2 drive), and TB 3 or 4 could get you to 4000MB/sec theoretically.
So if you’ve got a TB3 drive and, let’s say you’ve left enough room to copy a 100GB file to your internal drive. Remember, this is a tiny drive, so you’re not working with terabyte files. It will take you an extra 30 seconds to copy that file to the 256GB drive (67 seconds rather than 37 seconds). Now, finally, we’ve found a case that the drive speed makes a difference.
But how much is that going to save you in a day? Presumably you’re going to do something with that file, not just copy it back and forth all day. And you must be taking it mobile someplace you can’t really copy it back easily, otherwise you’d just work locally on the external drive. Maybe the difference is a minute a day— copy local at the start of the day, process, copy back at the end?
So a minute a day is the difference it makes.
Sure, I’d guess someone can contrive a situation where the impact is greater than that, but for most people this is what it comes down to.
