Apple M1 ARM Performance With A 2020 Mac Mini

[carewolf]

Just look at native results for SPEC on AnandTech. Beware, it's a massacre...

You mean it loses consistently to the latest from Intel and AMD? It is doing very well per watt, but we will have to see how the models progress. It will be interesting. But it is not over. Please stop drooling over things you imagine.

[carewolf]

I wonder how much IPC they gained by just increasing the page sizes and dropping a layer from the virtual tables, plus increasing iOPs to the memory by integrating it. It might tell us something about how limited they are in expanding the memory.

[CochainComplex]

The results are fantastic!

Not being able to run anything other than MacOS is craptastic.

Zero documentation on the innards to prevent other OSes from utilizing the hardware properly? 1000 times worse than NVIDIA.

/thread

Spot on! You are totally right.

[Jumbotron]

According to Statisica:

"In 2020, it is forecast that a total of 20 million Chromebooks will be shipped globally, an increase of 17 percent from the 17 million units that were shipped in 2019."

According to Gartner:

For Apple specifically, Gartner's data indicates shipments of 5.26 million units, which is down 3% year-over-year, and a marketshare of 7.5%.

LOL...yeah...because.....PANDEMIC !! Mommy and Daddy suddenly became a home school teacher and most schools are recommending a cheap and cheerful Chromebook since classes are now online, and Mommy and Daddy are NOT going to invest in a $ 1,000 Macbook for little Timmy or Suzie to dump a glass of juice on while doing an online course.

I know...I've had to buy 3 Chromebooks....one for the son, one for the daughter and another to replace the first one my daughter got.

Also...people don't buy Macs every year like iPhones. There is no need. There are still 10 year old Macs doing yeoman's work for people. Why upgrade if it is still performant for most people's needs.

Chromebooks....yeah...they're like the impulse aisles at Wal-Mart stocked full of gum and chips and beef jerkey. While you're waiting 10 deep for a check out, why not get a back of chips....or some gum. It's a cheap disposable item.

[Jumbotron]

For chrissake all embedded devices and APUs share the same RAM between their processors and accelerators. Smartphones from 5 years ago do that. AMD APUs do that, Intel CPUs with graphics do that.

HSA is an inter-operatibility standard and software middleware designed to be used to run computing tasks on hardware of different vendors, which is completely different from this that is a SINGLE system created by a SINGLE vendor that controls the full stack from the OS down to the hardware design.

Wow, it's using a feature DirectX 11.1 had in 2018 and is a core feature in both DirectX 12 and Vulkan, and calling it a new and revolutionary thing! Much excitement! Very Apple! Wow!

Yay, apple has made an embedded device that is using all embedded device technologies everyone else uses! Much innovation! Very Excitement!

Wow....you've been gone a while and you STILL came back an ignorant bastard !! LOL !!

Ok, junior, let's go back to school shall we ?

Yes, most embedded devices with SoCs, and AMD APU's and Intel CPUs with iGPUs share system RAM. But NOT zero copy and NOT with Cache Coherency. Only AMD's FUSION APU's based initially on the Bulldozer cores starting with Kaveri and ending with the Excavator based Carrizo and Bristol Ridge were full HSA compliant and shared the same 48 bit address space between the CPU cores and the GPU cores in the SoC for Zero Copy of data. Intel has NEVER had this and they do not even now. And AMD abandoned HSA and Zero Copy when they developed ZEN. Bridgman himself confirmed that ALL ZEN based APUs do NOT share address space between the CPU cores and the GPU cores in the SoC to ease development of code that would be used for both APU GPUs and discreet GPUs which means NO zero copy and NO cache coherency in ZEN APUs.

The only shot of an HSA type Zero Copy and Cache Coherency platform from AMD is the upcoming 3rd Gen Infinity Architecture that will debut with the rollout of ZEN 4 Ryzens and the EPYC part called Genoa along with the next iteration of RDNA and CDNA Radeon GPUs. Once again, confirmed by Bridgman this will ALSO depend on further work on OpenMP, HMM in kernel and ROCm. This will happen sometime towards the end of 2021 into 2022. And who the hell knows when we will see a ZEN 4 / Infinity Architecture based APU from AMD? 2023...maybe? And watcha bet it runs a tad bit hotter than 13.8 watts like the M1 even if the thing is processed at 5nm....which may be a stretch even for TSMC.

On top of that, only Apple and the M1 is using HBM memory tied to the cores via high speed cache coherent interconnects, probably CCIX. Yes, it's not stacked such as certain AMD GPUs, AMD has pretty much given up on HBM stacked on the die of Radeon GPUs. Intel won't have CXL interconnects until next year, and AMD is still at 2nd gen Infinity Fabric and the full zero copy, cache coherent 3rd Gen Infinity Architecture won't arrive until, as I said above, late 2021 into 2022 and not for APUs until at least 2023.

And what in the bloody hell are you mumbling about Direct X what ever and Vulkan? Apple's Metal runs rings around both Direct X and Vulkan. And on the subject of Vulkan....is that RADV, AMDVLK, AMD GPU-Pro, or maybe Intel's version of Vulkan ? Oh..that's right...we have to pick the right Vulkan driver for 1,400 different Intel and AMD SKUs of GPUs. Some of discreet GPUs, some of Integrated GPUs, and some have GPU's that share the same address space as the CPU's like HSA APUs.

Yeah...with surfeit of choices, that REALLY drives home performance doesn't it ? LOL !!

Long story short, ignorant Apple hater, the M1 is the most innovative consumer chip since the original Athlon/Opteron, with the possible exception of the last of AMD's fully HSA compliant APU's and the last of the Excavator line that being Bristol Ridge. That was a true marvel in silicon engineering. But it was too little, too late and saddled with a CPU core design that was just compromise after compromise and had to be iterated over 5 years into something that had a chance until AMD's focus radically changed with the onboarding of Lisa Su and her relentless push to get AMD back into making high margin Server chips and custom SoCs for the likes of the PS 5 and the X-Bone.

Intel has nothing like the M1. AMD has nothing like the M1. And neither will EVER have anything like it at that kind of thermal envelope. The best Intel and AMD can do is try to make an M1 like SoC for desktops in a couple of years that could compete on raw compute. But if they even attempt to meet the thermal envelope of an M1, which they can do, it will suck compared to the M1 in performance. It is impossible to make a x86 SoC with a built in Neural Processor, Advance Matrix Math Processor, Audio DSP, Video DSP, 8 CPU cores running up to 3 Ghz for the Hi-Perf cores and over 2 Ghz for the "Low" power cores AND have all cores running simultaneously, all tied together with an HSA like UMA with zero copy and cache coherency. And make where that x86 SoC only has a max 13.8 Watt TDP.

You really do need to educate yourself son.

[vein]

Yawn!

Yes the M1 is complete shite that no one but apple fanboys will buy.

So nothing to see here..just move on...

[PerformanceExpert]

In that case shouldn't we also use some dynamic translator, or even better output the correct binary in the first place? What's the reason for not doing that? (I understand needing to support older/more generic stuff, but is not possible to support both?).

The key difference here is the microarchitecture of M1 which is incredibly fast. The simpler Arm ISA has fixed-length RISC instructions which allows 8-wide decode. This graph shows the steady progression of Apple's microarchitectures, showing clearly x86 has run out of steam.

No, both static and dynamic translation always add overheads. Compilers keep improving, but they are already very good, and there isn't that much opportunity to do better. Also many improvements help all ISAs, so it wouldn't close the gap.

[PerformanceExpert]

You mean it loses consistently to the latest from Intel and AMD? It is doing very well per watt, but we will have to see how the models progress. It will be interesting. But it is not over. Please stop drooling over things you imagine.

Did you even check my link? Just asking, because which CPU wins against the fastest Zen 3 desktop chip here?





For the uninitiated, SPEC is the benchmark that CPU designers use to improve CPU performance.

[blackshard]

I'm absolutely not a fan of Apple, but I'm pretty impressed by the numbers of both the M1 hardware and Rosetta software.

The chip is very interesting because it can compete fairly well with x86 heavily optimized hardware in terms of IPC. Don't be fooled by the encoding/decoding tests, that's probably because those benchmarks are using manually optimized SIMD code for x86 but not for ARM. In general purpose number crunching, the M1 is doing wonders. Maybe it will lose against latest Ryzen parts, nonetheless it is in the same ballpark.
Also 7zip compression numbers are huge, so memory subsystem here is doing wonders - or maybe they just have a lot of fast cache.

Rosetta is losing some terrain due to emulation, but it is always in the range of one order of magnitude, which is a great great achievement: emulating an architecture on another one is quite a complex and slow task, and that piece of software is doing wonders there too! Probably it has some assistance support in the hardware, but indeed the mix is very effective!

Two things are missing here to get the real picture of this potentially revolutionary chip: power consumption and frequencies.

[PerformanceExpert]

Where do you funny guys come from? From the same place where you have such fairy tales benchmarks?

From the real world. Check the links I posted, it's from sites that do serious benchmarking (I hope Phoronix will do some native benchmarks too). Yes I know it's not easy to admit Apple won the performance crown and even beats Zen 3, but that's a fact.