Apple Neural Engine Internal: From ML Algorithm to HW Registers

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by RealityVoid on Thursday December 09, 2021 @12:14PM (#29500090)

The M1 is an impressive chip... for what it is. I bought an Air just to mess with it, while in the past I would not have touched Apple with a 10 foot pole. But, the whole framing around M1 being better than _everything_ in the x86 world was just weird and not true. Yes, perf/watt it was unbeatable, but the rest...if you just care about raw processing power, you'd still buy a desktop x86. I think Apple framed it this way and most media ran with it, but I don't see this a sue apple kind of deal.

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by tambourine_man on Thursday December 09, 2021 @06:10PM (#29504364)

M1 tops SPEC2017 while giving you 15-20h of battery life. I'm not mentioning hardware specific accelerated tasks like Final Cut, if you throw that in it beats Xeons at a fraction of the price and power. It's just the best portable chip for regular, boring, standard Floats and Ints computation. There's really no competition yet.

Oh, and it's got crazy fast IO. And it's last year's architecture and process node. M2 with 4nm and A15 derived architecture should be another 20-30% faster at the same wattage.

The real question is the Mac Pro. What can and will Apple do when performance is the main focus? Will they just stick a bunch of M1s together (which wouldn't be bad) or are they able to crank single threaded performance a lot higher with more fans and watts to spare?

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by Jcowell on Thursday December 09, 2021 @02:25PM (#29501761)

If raw performance is the only thing you care about then you would get a desktop. If you care about mobility as well or changes a lot.

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by marderfarker2 on Thursday December 09, 2021 @03:05PM (#29502232)

It’s funny that you mention mobility. I usually see people stuck an extra slab of USB C to HDMI/USB hub beside their thinnest and lightest Apple laptops so that they can use peripherals.

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by Toutouxc on Thursday December 09, 2021 @04:53PM (#29503456)

Usually, except for the overwhelming majority of people who don't use a hub.

(e.g. my Air has two cables running into it, power and display)

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by caconym_ on Thursday December 09, 2021 @08:04PM (#29505251)

Not sure I understand what point you're trying to make, but when I "go mobile" I am typically leaving behind my monitor, my wired keyboard and mouse/trackpad, my backup drive, and most of the other shit I might have daisychained off a USB hub. So I actually see it as an incredible convenience that I can run all of that, plus charging, plus wired networking, off a single USB-C port.

It feels far more "mobile" than a setup where I need to plug/unplug all that stuff into the body of the laptop whenever I want to use my laptop away from my desk.

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by photojosh on Thursday December 09, 2021 @09:44PM (#29505873)

And those USB-C hubs are like $20, so I can just get one for each spot I move between. Remember the old days of $200 hubs?

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by vmchale on Thursday December 09, 2021 @12:19PM (#29500158)

I've actually found it slightly better in single-threaded performance. Only slower with rosetta.

[ Parent | on HN ]