Apple Releases M1-Powered Apple Silicon Macs, macOS Big Sur Releasing This Week

[muncrief]

Like really?


Zen 3 running at 5GHz at 15W reaches 1700 points and an unconstrained A14X should be "abysmally slow". The one on the picture is limited to less than 3W btw.

Ha!

Apple just makes stuff up, and generates huge reality distortion fields. That's why everything has to be tied to "performance per watt", instead of straight up performance.

I'd love to see real benchmarks between real CPUs - single core, multi-core, and all unleashed to full balls to the wall.

The M1 would come out rated as full on pathetic.

[birdie]

Ha!

Apple just makes stuff up, and generates huge reality distortion fields. That's why everything has to be tied to "performance per watt", instead of straight up performance.

You're either drunk or insane. GB5 shows "straight up performance" FFS. GB5 results have been corroborated by AnandTech as well who I trust a lot more than a random no one on Phoronix.

[Jumbotron]

Like really?


Zen 3 running at 5GHz at 15W reaches 1700 points and an unconstrained A14X should be "abysmally slow". The one on the picture is limited to less than 3W btw.

Yeah....the Big Metal x86 guys haven't grokked it yet. They will but it will take a few years. Moore's Law has killed any real progress for x86 simply because they are STILL trying to do too much via the CPU when compute loads now HAVE to be disaggregated and sent to the proper chip for the compute type needed. This is PRECISELY why the totally integrated HSA based APUs for AMD failed and now you see the rise of SiPs (System in a Package) and the rise of "Chiplets". We are getting to the limits of Newtonian Physics as transistors shrink. You now have to spread the transistors BACK OUT of the totally integrated SoC like the x86 APUs into chiplets and SiPs.

Hence the need now for either on die or in package GPU's, DSP's NPU's and FPGA's. The age of the heavy metal CPU is dead. Particularly x86. The only place where it will be relevant going forward is HPC and Supercomputers. But even there you also see the rise of DSPs, NPUs and FPGAs.

[ezst036]

There's a small chance Apple will continue using AMD video chips in their workstations but even that is not certain.

I was wondering about this after seeing your comment, if Apple had a small line about their new PCIe 16x graphics cards to compete with Radeons or Geforce. But Apple seems to be going the other direction with M1.

Graphics cards are not allowed - the support isn't built in:

https://techcrunch.com/2020/11/10/macs-with-the-m1-chip-do-not-support-egpus/

I suspect this means the M1 will only be in the low and midrange AIO macs, minis, and laptops. I find it a little hard to believe Apple is going to stop selling its ten thousand dollar Mac Pro with twin Radeon Vegas in it.

Hopefully someone else can add in more information than I am aware of, or correct an error if I made any.

[drakonas777]

These synthetic benchmarks mean shit. For example, Intel 9 years old Sandy Bridge garbage is still among fastests in SuperPI - so what, it gets destroyed by AMD ZEN in any real world workload.

Start a linux compile in the VM on that M1. It will lag like a effing turtle compared to ZEN3.

[Jumbotron]

What's a SBC? Single-board computer?

Yes, single board computer. Think Raspbery Pi.

Or in the case of Apple....the Apple Mini. Really most computers are now single board computers. We just think of SBC now as something that will fit in the palm of your hand.

[birdie]

I was wondering about this after seeing your comment, if Apple had a small line about their new PCIe 16x graphics cards to compete with Radeons or Geforce. But Apple seems to be going the other direction with M1.

Graphics cards are not allowed - the support isn't built in:

https://techcrunch.com/2020/11/10/macs-with-the-m1-chip-do-not-support-egpus/

I suspect this means the M1 will only be in the low and midrange AIO macs, minis, and laptops. I find it a little hard to believe Apple is going to stop selling its ten thousand dollar Mac Pro with twin Radeon Vegas in it.

Hopefully someone else can add in more information than I am aware of, or correct an error if I made any.

Probably new Mac Pros will be released with Zen 3 CPUs and RDNA 2.0 GPUs since the M1 chip is seemingly built to work only with on-die RAM (HBM or whatever it is) and people who need to have work done may require tons of RAM and GPU power.

[birdie]

Yes, single board computer. Think Raspbery Pi.

Or in the case of Apple....the Apple Mini. Really most computers are now single board computers. We just think of SBC now as something that will fit in the palm of your hand.

An SBC based on the M1 chip would be a beast - not gonna happen though. "It's Mac Mini or fuck off" - Apple :-D I guess close to 95% of PCs/laptops in the world are slower than what this little chip offers.

[Jumbotron]

While I applaud Apple for the M1 chip I'm really sad they won't offer it to other OEMs, so that we could have non-apple laptops and NUCs based on it. I for one would love to have a laptop with the M1 SoC with a comparable battery life and performance, only with standard DDR4 SO-DIMM RAM slots, standard M.2 SSD slots and a replaceable battery.

Uhhh....you do know there is Qualcomm, Mediatek and Marvel right? They are going to look at this M1 and use it as their benchmark in engineering and performance going forward.

Make no mistake. Qualcomm right now is looking to poach an Apple Silicon engineer as Intel did with AMD in order to make GPUs.

Apple has set the benchmark and the bar VERY HIGH in terms of ARM engineering and performance. There are LOTS of jealous engineers right now looking to do better.

Some of those will make WAY more performant ARM CPU's for the likes of Google and Microsoft along with Lenovo, HP and Dell very soon now.

[Jumbotron]

It's NOT DDR4

It's LPDDR4X. The SLOWEST LPDDR4X is as fast as regular DDR4. At its fastest it is around 40% faster than DDR4.

Also look how physically close the RAM chips are to each other and the CPU on the package. That belies the interconnect tech that Apple uses to tie the RAM into the Die and across the package. Once again, this SoC is zero copy HSA, the likes that AMD championed at one time back in the mid-2000's and abandoned.

Another table that shows the list of various generations of LPDDR memories is provided here. The generations of LPDDR memory have followed a similar pattern (like DDR memories) when it comes to improvements in the data transfer rates and voltage levels.

	Year of release	Data transfer rate (MT/s)	Voltage level (V)
LPDDDR	2008	333-400	1.2V
LPDDR2	2010	800-1066	1.2V
LPDDR3	2012	1600-1866	1.2V
LPDDR4	2014	3200	1.1V
LPDDR4X	2017	3200-4267	0.6V
LPDDR5	2020	6400	0.5V
LPDDR5X	2023 (expected)	–	–