All kinds of strange things crop up in this thread.
1. RAM requirements are pretty much independent of instruction set. Data is what requires RAM, the RAM footprint of the instruction code is typically much, much smaller.
2. Caches don't help at all with RAM size requirements. Caches help with average latency as seen by the cores, as well as average bandwidth. But NOT with footprint. If you need more than 8GB of RAM, caches won't help at all.
Exactly! I find it very strange that some people claim that ARM Macs will somehow magically double the effective amount of RAM. Memory is memory, Apple Silicon won't change that simple fact.
3. Fast SSD:s does help reduce the impact of paging for data as you run out of RAM. This is important, as that is the scenario where regular people most clearly perceive that they are running out of RAM. The clack-clack-clack of mechanical disks is gone, and speed from PCIe SSDs is more than an order of magnitude faster than the mechanical disks. It is also still more than an order of magnitude slower than RAM in both latency and bandwidth however, so if you run out of RAM, your performance will still drop massively. But for short bursts (swapping in the data of a background app shifted to the foreground for instance) it's nowhere near as painful as it used to be. CPU instruction set doesn't affect this at all. Hardware resources for encryption and compression can affect it to some modest degree.
Just a few things I'd like to add here. Most of the time running out of RAM means that the total memory usage of all the open apps exceeds the amount of available RAM. Fast SSDs and memory compression can work wonders in this case. Reloading browser tab memory from SSD is almost always faster than the usual tab switch animation, which results in a seamless user experience. So 8GB can indeed be more than plenty even for a heavy multitasker as long as the currently active tasks (the one the user directly interacts with) need less than 8GB.
Where compression/SSD won't help though if you actually need more RAM at a given time. If the system has to swap memory in and out for an active application, you will have lag. Same applies if multiple applications have to be active simultaneously (as is the case for some development workflows where you have multiple VMs open with active components).
Overall, I think that the RAM is becoming less critical to ensure smooth multi-tasking operation, but one needs to make sure that one has enough RAM for the relevant workflows.
4. A unified memory subsystem means that the graphics data consume application data space. This is business as usual (actually it might be a bit better) if you were running integrated graphics before, but it is worse in terms of RAM required if you are used to having a GPU with its own local memory.
This is not necessarily the case. Anything in the GPU memory has to first pass though the CPU memory, so even on a system with separate GPU RAM you will have significant system RAM usage dedicated to GPU operation. Furthermore, GPU resources are frequently mirrored in the system RAM by the driver so that they can be reloaded/synchronized quickly. In practice, having more VRAM actually often means that more of your system RAM being stolen by the driver.
Apple's implementation of unified memory simplifies this entire driver spiel massively. There are no copies to worry about, no synchronization, no DMA transfers. Everything is just virtual memory. Pages that are use less will be compressed/offloaded, and if the GPU needs them, they will be transparently loaded again. The overall memory allocation efficiency improves tremendously. Frankly, unified memory is probably the aspect of Apple Silicon that excites me the most — this will make GPUs so much more useful.
And of course, let's not forget about texture compression — Apple GPUs are leaders in compression standards support, which can dramatically increase the size of effective memory. Using latest ASTC compression, you can fit dozens (or more) of megabytes of texture data into the 8GB RAM space without any perceivable quality loss, while still having enough space for your application code.
That said, the average MR commenter seems to be running notorious resource pig Google Chrome with 15-20 tabs. If these are CS graduates, the world is doomed.
15-20 tabs? Why so little? I have 30 tabs open right now, and that's a fresh Big Sur install
Granted I didn't go to a big name school, but I'd say a good 1/3 of the students in my CS classes were using Macs. The ratio was higher in the Unix and (of course) the iOS classes.
In the classes I teach (mostly humanities) I'd too estimate that around 1/3 of students use a Mac.
