That article seems confused and/or misinformed on many points, on many levels.
As for the GP cited use case, there's no plausible way in which the compilers involved in the V8 build process would start using less memory due to their memory management machinery getting magically transformed by the hardware from tracing GC to refcounting GC, even assuming LLVM had been using tracing GC in the first place.
(Also the tracing GC memory overhead claims originating in Apple's marketing copy are way hyperbolic, modern tracing GCs don't require double storage used in "mark and sweep" GCs)
Many memory hungry apps on Apple - Web browsers, compilers, photo/video editing, games, dev tools like vscode/electron/Jave IDEs/emacs/databases - are in this category.
I think the point is that memory bandwidth and SSD bandwidth and latency have improved so much that M1-based Macs are really fast at swapping. Combine that with compression (which macOS has done for a long time) and I could imagine iPadOS-like performance.
For normal users (such as people not compiling V8), swapping may be virtually unnoticeable.
Macs have had fast SSDs for a while and as you say compression has also been done for a long time. M1 made no difference whatsoever to memory usage. The posts similar to above seem to just be an attempt to fanfic-justify what's an actual regression for some.
For many people 16GB is absolutely enough, and Apple had to start somewhere. But if 16GB wasn't enough for you yesterday, it isn't enough for you now.
Do you have a link? MKBD just said it was "slightly faster" than his comparison system and the numbers for the M1 appear to be around 3GB/s reads and 2.7GB/s writes in disk speed test, which is pretty typical of a last generation NVME drive (current pcie gen4 drives are hitting 7GB/s reads & 5GB/s writes)
Then sounds like Apple put some really garbage SSDs in the 2020 Intel-based Air, not sure what you're looking for here?
But 2-3x won't change swap from grinding to a halt to perfectly smooth, either. It's still over 20x slower than RAM.
More significant for swap usage though is random reads and read latency, neither of which are going to be particularly impressive on the M1's SSDs. You need something like Optane to make that a compelling case.
I do wonder if it’s taking advantage of the cheap CPU power and memory bandwidth to be much more aggressive about memory compression. That could reduce swapping for some workloads, though probably not a v8 build.
What do you mean exactly? It is normal for swapping to happen even if you are using less than the physical memory, so as to trade rarely-used RAM for frequently-used disk cache which improves overall throughput
