4 ms·

> The Mac Mini M1 maxes out at 16GiB of RAM ...and that tight coupling between RAM and CPU must be a big part of its performance. I wouldn't really be surpris

by usrusr 6y ago

> The Mac Mini M1 maxes out at 16GiB of RAM ...and that tight coupling between RAM and CPU must be a big part of its performance. I wouldn't really be surprised if they went for a somewhat lazy architecture for the stationary/pro segment that simply (well, "simply", it won't be easy, just maybe less hard than a properly scaled up M1, perhaps much less hard) put n of those units on a shared carrier board with lots of memory slots that would either serve as slower RAM, relegating those 16 GB to an awesome "L4 cache" or maybe even just as one big volatile SSD swapspace. A "lazy architecture" might appear a bit out of character for Apple, but that pro segment must be so far from their core focus that this approach could work out really well. A conventional strategy whati new tech gets introduced in the pro segment and then trickles down until it eventually appears on the iPhone wouldn't really cut it because the phone is their unquestioned flagship and a "pro board" could make trickling up so easy that they might theoretically even make those things end-user upgradeable.

floatboth 6y ago
It's not really tighter coupling than in any other ultra portable laptops. It's just located on the package instead of on the mainboard, because they can. Shorter traces are better, but they don't magically make an incredible difference in performance. The newest x86 laptops run >4000 MT/s just fine through the board. Overclocked desktops run more through a socket, a big ass board, connectors, and full size DIMMs.
- usrusr 6y ago
  Oh, I would have thought that they'd at least get something like massively improved latency, if not something comparable to what dedicated GPU cards do (using completely different memory standards, and likely sacrificing latency for bandwidth). If the M1 units only integrate the RAM for the form factor and to avoid all the hassle supporting a wide range of RAM modules (or the hassle of communicating that they are almost all unsupported) then it's probably not much of a deal just building a bigger module. A "cluster of MacBooks" style board might still be an interesting approach to offer something to the market segment that hungers for an Apple workstation without relegating the MacBook to a second class device. "For up to four threads it's just as fast as the mighty stationary monstrosity" would be a much better sales pitch than "if the mobile isn't fast enough for you you might be interested in or stationary offering".

by usrusr 6y ago

floatboth 6y ago

It's not really tighter coupling than in any other ultra portable laptops. It's just located on the package instead of on the mainboard, because they can. Shorter traces are better, but they don't magically make an incredible difference in performance. The newest x86 laptops run >4000 MT/s just fine through the board. Overclocked desktops run more through a socket, a big ass board, connectors, and full size DIMMs.

usrusr 6y ago
Oh, I would have thought that they'd at least get something like massively improved latency, if not something comparable to what dedicated GPU cards do (using completely different memory standards, and likely sacrificing latency for bandwidth). If the M1 units only integrate the RAM for the form factor and to avoid all the hassle supporting a wide range of RAM modules (or the hassle of communicating that they are almost all unsupported) then it's probably not much of a deal just building a bigger module. A "cluster of MacBooks" style board might still be an interesting approach to offer something to the market segment that hungers for an Apple workstation without relegating the MacBook to a second class device. "For up to four threads it's just as fast as the mighty stationary monstrosity" would be a much better sales pitch than "if the mobile isn't fast enough for you you might be interested in or stationary offering".