Apple's M1

@ Steffen Huber

I am going by “real world performance” as measured with a quite complex component written in plain C at work. An AIX POWER9 machine outperformed the competition (e.g. Intel Xeon Cascade Lake at nearly 4 GHz IIRC) by quite a margin.

I totally believe that, AMD Zen 3 killed completely the Intel Xeon series (all of them) too.

But in the price/performance ratio, x64/Linux still won by a large margin. IBM AIX pricing is really quite eccentric.

Yup that’s the thing isn’t it? We have got kicked by PCs back in the days when they started to be cheap and slightly more powerful. Now ARM is taking back its throne because ARM boards are cheaper and more powerful… let’s see what happens with POWER10, again it’s very exciting and it should also be open source whatever that means for IBM lol XD

I did my CPU design involvment when I was younger, it is an art for the young for sure.

What does age have to do with it?

Some of the best damn engineers I’ve ever known (electronic, mechanic, and in one case steam!) were all retired and just kept on doing what they did for the love of it (and understood things in a way that young’uns just don’t have the attention span ¹ to cope with).

Hmm…

<thinks>

Kinda why we’re here, isn’t it? :-p

¹ I never had an attention span… more a fleeting moment of transient noticing.

NOW MAY WE PLEASE, PLEASE, PRETTY PLEASE get back ON TOPIC?

You can buy an Apple M1.

And even with Apple pricing it would be cheaper than a make it yourself ARM2 on streoids.

…back on topic…for sure…as can be clearly seen from Chris impressive benchmarks…the M1 and 64Bit is the way ARM architecture will go in terms of value/performance…and they just started…

Meanwhile I found some information that the virtualisation software is making good progress on M1. As can be read here

I’m not really experienced with virtualisation…would that mean it could emulate also a 32Bit Linux, where one could run the Linux-compiled Risc OS ? I remember on x86 I had once a virtualized WinXP running on Windows10…so a chance for Risc OS on the most modern ARM hardware…or it’s stupid and we let that 32Bt Linux run virtualised already on the latest PC hardware !?

But in the price/performance ratio, x64/Linux still won by a large margin. IBM AIX pricing is really quite eccentric.

Absolutely and the perf from the new M1X have just leaked, it appears that the tested CPU has now 6 performance cores and 16 GPU cores.

Now doubling the GPU cores makes total sense because they scale up, while not doubling the performance cores also makes sense since they do not scale up in that regards.

Anyway predictably the M1X will be able to compete with high end GPUs (not the ones using HBM2, but the slightly less pushed high end GPUs) and will obviously perform even better against AMD zen3 which is going to dominate PCs market on 2021.

Next on the list to appear probably during October/November 2021 may be the ARM based macbook Pro 14" and 16", in the end M1 already beaten in performance the actual Intel based Macbook Pro 16" so I believe with the introduction of the M1X there will be no need for Apple to keep producing the Intel based Macbook Pro 16".

I’ll wait for the M1X, but yes as I like to mention 2021 is the beginning of ARMageddon. Microsoft is also working on their new ARM based laptop and Qualcomm appears to have become serious about big ARM Chips for Server and Workstation, so if it keeps going in this direction AMD will keep the legacy PCs, Intel will become irrelevant until they figure out how to produce 7 and 5nm Chips… meanwhile Apple is already looking at the new TMC 3nm process.

I am interested in seeing what ARM Holdings will do with the new ARM X1 Architecture there is a lot of interest around it and it’s definitely the future for Desktop Application on ARM (if ARM Holdings does a good job with keeping it competitive).

And even with Apple pricing it would be cheaper than a make it yourself ARM2 on streoids.

absolutely true. The ARMv2a project if will complete and be improved can only probably match or beat an overclocked ARM250, an that is as far as we can go with the most optimistic view, BUT for the retro market it will ensure “eternal life” for RISC OS 2.x/3.x, which for me it’s fun and more than I could hope for for the retro stuff.

I mean even collecting documents to suggest a way (either hypothesis or theory) would mean no deliverable product: again licenses, prices, technical difficulties, team of experts that needs to be put together and willing to work for free and then fight legal battles…

Guys, IMHO, it’s easier to port RISC OS to C, move to 64bit, build a binary translator for old applications… just my 0.5c. Also having RISC OS running on the mainstream ARMs mean to keep it relevant for the market, while running on emulators or FPGA means just retrocomputing market/nostalgia which is something we should think of as well, but again just my opinion.

@ DavidS

Definitely interested to see the result of a ARMv2a implementation. Kool.

Not just ARMv2a, it’s a full Archimedes on a chip, and already runs Arthur, RISC OS 2.x and RISC OS 3.x fine, I’ll make a video of it on youtube soon. There are already plenty of people around the world using it to run old RIS COS games on the MiSTer board and recently it has got HD Image support, I am working on creating a pre-partitioned HD for people to Enjoy RISC OS on FPGA.

Performance wise is slightly slower than a A3000 right now, but it’ll get better and better. 26bit RISC OS lovers have future fully ensured now and I am also trying to convince someone to help me to port RISC OS 5 to the actual modern ARM32 board that uses that FPGA, so that the ARMv2a can be executed as we now run ArchiEmu in a window (or full screen), but running software in real hardware and so no performance glitches coming from emulation on RISC OS 5. So again a very very very exciting project and everyone in retrocomputing know MiSTer very well, if this last thing get a team we could also run Amiga and ATari cores in a RISC OS 5 window (or full screen) so you could have your dreamed platforms all-in-one ;)

My mistake.

Indeed so.

The clue was in the thread title. Of course the thread has started to drift again (away from the M1)

Meanwhile I found some information that the virtualisation software is making good progress on M1. As can be read here

I’ve updated the original post. The third column shows the results for Ubuntu 20.10 aarch64 guest running on Parallels on the M1.

I’m not really experienced with virtualisation…would that mean it could emulate also a 32Bit Linux, where one could run the Linux-compiled Risc OS ?

Parallels is using the Hypervisor Framework running directly on the host CPU at near native speed. So 64bit only still.

@Chris: Thanks for the info and adding those numbers…but can it be true that the Ubuntu code sometimes is even faster while hosted by a virtual system ? Or is that the different compiler optimisation effect like clang vs. gcc ? I thought the XCode package would contain also the gcc so you could compare the same gcc generated code ?

Anyway really impressive what the virtualisation can do…

@ Kuemmel

I thought the XCode package would contain also the gcc so you could compare the same gcc generated code ?

AFAIR the last XCode release that came with GCC included was release 4 or release 4.1, after that it had only LLVM, CCTools, Clang, Objective-C and then at some point added Swift.

You can however install GCC via homebrew and then configure XCode to use GCC backend to compile your C, C99 etc…

@DavidS: I doubt Apple will share any tech details…on the other hand, what would you wanna do with the information in the real world, create your own cpu ?

On a superficial level it’s very clear what they did, same story like on recent x86, you “just” add execution units or ports (like Intel is calling it). So a massive parallelism. Some German dude wrote an article, on picture 3 (“Bild 3”) you can see that massive amount of execution units. Link ist here
…if it just could be that easy…;-)

EDIT: Here’s also a Link from Anandtech regarding the execution units. To feed all those units the decode is 8-wide ! Find it here

as it is a long time before any follow on of RISC OS will run on AARCH64

I spy an incurable optimist here. Well, perhaps on an emulation of a Pi, but what’s the chance of that being any faster than a Pi, or 100% compatible?

The new Mac is fun technology – the Pi is useful, affordable and fast.
An extremely fast Pi (for RISC OS) is ten times cheaper than a Mac.
The Pi compute module seems more interesting than the Mac; carrier boards for compute modules could be used to create powerful high-end RISC OS systems.

Performance wise is slightly slower than a A3000 right now, but it’ll get better and better. 26bit RISC OS lovers have future fully ensured now and I am also trying to convince someone to help me to port RISC OS 5 to the actual modern ARM32 board that uses that FPGA, so that the ARMv2a can be executed as we now run ArchiEmu in a window (or full screen), but running software in real hardware and so no performance glitches coming from emulation on RISC OS 5.

I like the idea. While I’m using a Pi4 every day for work, I’m not against a modern-retro platform. Even if pricey. I’m pretty sure it’s easier to sell a 500£ Archi remake than a 100£ ROS/Pi solution.

On the other hand, the ArchiEmu emulation level is very good, and the emulated speed can go to more than 100 MHz. I hope that it will support RISC OS 3.5 one day :)

On the other hand, the ArchiEmu emulation level is very good, and the emulated speed can go to more than 100 MHz. I hope that it will support RISC OS 3.5 one day :)

Totally yup, I love ArchiEmu. Beside what you have mentioned it also has a good debugger for people who really want to learn how RO 2/3 worked, it supports the VideoEnancher and more. An impressive piece of software IMHO :)

The FPGA Core for RISC OS it’s just fun and again learning process. The cool thing is, if the above idea works (aka I manage to either find the time or people interested in working on it), the creation of a RiscPC-Like platform for RO 5 that goes beyond running the ARMv2a as a guest CPU.

For example MiSTer board can already run an Amiga in hardware like we did for the x86 PCcard on the RiscPC, an Atari, a BBCMicro, C64, ZX Spectrum, VIC20, many consoles and other 8bit and 16bit computers (there is also an FPGA based core for the 486 that runs Windows 95), original Arcades. So the idea is something like a RiscPC2 on steroids running RISC OS 5 as host OS on the modern ARM dual core on the same board ;)

re compute module: Is this what the RPi 400 is?

Yes but No; the compute module is socketed; so the board designer has options to decide what to do with the modules PCI slot; and can decide what else they want to add to the board; how they want to deliver power and so on. So it is possible to have a standard PC format board; add SATA or NVME or a graphics card; and even possible to design a board to support a cluster on a board using multiple modules.
The RPi 400 is great though.

e.g. A small board with NVME ssd support.
https://www.cnx-software.com/2021/02/26/raspberry-pi-compute-module-4-industrial-carrier-board-supports-m-2-nvme-ssd-4g-lte-modem/

sorry for vagueness: It is the same chip 2711ZPKFSB06C0T in both, different to the regular Pi4.
It is on the board, not on a module on RPi 400.

RPi 400 ONE example of the Compute module applied to a product?

RPi 400 is not Compute module based, it’s a board on its own

I’m not really experienced with virtualisation…would that mean it could emulate also a 32Bit Linux, where one could run the Linux-compiled Risc OS ?

qemu-arm runs OK on Ubuntu guest on Parallels so RISC OS Linux builds and runs OK too. The CPU %age always seems to be 7percent under qemu.

MOS Utilities   5.29 (13 Feb 2021)
FileSwitch      2.88 (30 Jan 2021)
C Library       6.09 (06 Feb 2021)
FileCore        3.77 (14 Dec 2020)
HAL-based system (Linux HAL PSR Manipulation) tested at Tue,02 Mar 2021.19:07:38
Filing system: IXFS$.HardDisc4.Software
Graphics Resoloution: 1024x768, 16M colours

Test                                                   Benchmark
Processor - Looped instructions (cache)                    13797       7%
Memory - Multiple register transfer                        74510   45993%
Rectangle Copy - Graphics acceleration test                12487    5159%
Icon Plotting - 16 colour sprite with mask                 55015    2750%
Draw Path - Stroke narrow line                              9501     609%
Draw Fill - Plot filled shape                              13631     934%
Draw Render - Render draw file                              9790    1080%
HD Read - Block load 8MB file                            5028752  168636%
HD Write - Block save 8MB file                           1881726   61878%
FS Read - Byte stream file in                                354     171%
FS Write - Byte stream file out                              328     170%

vfp and neon work too but slowly.

15.26s FixFrac_64BitMUL
6.74s FracNEON
29.1s FracVFP_Double
29.18s FracVFP_Single

@Chris: Thanks for testing…I’ll start porting my FixFrac_64BitMUL code to a non graphical AARCH64 Linux version soon (GNU Assembler) to get some native results…once that is done I continue with the other three Frac’s.

Just use CoPilot to write (most of) your (Risc OS) code – whether in ARM64 assembler, C, C++, C++20, BBC Basic, Pascal, BCPL, Forth, Logo, PROLOG, Javascript, JSON, Perl, Python…:
https://copilot.github.com/

Just type RISC OS 6 or 7 or 10? It may autocomplete the code to your next Risc OS – it is flawless AI :-] But DO NOT type skynet, that will jeopardise our whole civilisation :-(

Maybe copilot will some day be standard in your (Risc OS) IDE?

Rumor:

Jun. 18th 2021, Leaker claims new redesigned MacBook Pro will be launched in Q4 along with M1X Mac mini:
https://9to5mac.com/2021/06/18/leaker-claims-new-redesigned-macbook-pro-will-be-launched-in-q4-along-with-m1x-mac-mini/
Quote: “…
MacBook Pro having a…expanded connections — which includes an SD card reader, MagSafe, and HDMI port
…”

A month ago:

Linux ported to Apple Silicon (M1):

Linux on Apple Silicon:
https://asahilinux.org/

https://github.com/AsahiLinux/

08 October 2021, pcgamer.com: Developers finally get Linux running on an Apple M1-powered Mac
backup
Quote: “…
This massive feat was achieved by having a bunch of drivers merged (or in review) for Linux 5.16, which includes drivers for the PCIe, USB-C PD, ASC mailbox, etc. If you’re wondering if any of this is legal, don’t worry. So long as no code is taken from macOS to build Linux support, it’s legal to distribute.
…”

asahilinux.org: Progress Report: September 2021
Quote: “…
However, Apple is unique in putting emphasis in keeping hardware interfaces compatible across SoC generations – the UART hardware in the M1 dates back to the original iPhone! This means we are in a unique position to be able to try writing drivers that will not only work for the M1, but may work –unchanged– on future chips as well. This is a very exciting opportunity in the ARM64 world.
…
With these drivers, M1 Macs are actually usable as desktop Linux machines! While there is no GPU acceleration yet, the M1’s CPUs are so powerful that a software-rendered desktop is actually faster on them than on e.g. Rockchip ARM64 machines with hardware acceleration.
…”

-

Later? support for:

October 18, 2021, anandtech.com: Apple Announces M1 Pro & M1 Max: Giant New Arm SoCs with All-Out

Thanks to Acorn Computers for making ARM architecture and making it possible to innovate ARM architecture further.

-

Another port:

January 2021, corellium.com: How We Ported Linux to the M1
Quote: “…
So when Apple decided to allow installing custom kernels on the Macs with M1 processor, we were very happy to try building another Linux port to further our understanding of the hardware platform. As we were creating a model of the processor for our security research product, we were working on the Linux port in parallel.
…”

https://www.corellium.com/platform
Quote: “…
Corellium’s ARM2ARM feature offers transparent 32-bit translation for Android, so you can run apps with 32-bit libraries on hardware without 32-bit silicon without making any modifications to your code.
…”

-

(post copy at: https://www.riscosopen.org/forum/forums/9/topics/15400?page=3#posts-128137 )

16 Dec 2021, theregister.com: What does 2022 have in store for Asahi Linux on Apple chips? Drivers aplenty, but that GPU still needs tackling.
Handily, the team spotted a bug in Linux’s ARM SMMU support too
Quote: "…
“Amusingly,” Martin wrote, "while implementing support for this in Linux, we ran into a bug in Linux’s ARM SMMU support that had been there ever since 52-bit address support was introduced.
…
“This was breaking systems with more than 256 TiB of RAM – I wonder why nobody noticed? Either way, Linux now correctly supports standard ARM systems with up to 4 PiB of RAM.”
…
There is still plenty of work to be done, not least in the audio department. The next area of focus, according to Martin, is Wi-Fi. Besides a number of other bits of coding (including RTC support), there is also that pesky GPU kernel driver to be tackled (although Martin reported that the userspace is in “good shape”).
At the current rate of development, it seems entirely possible that Asahi Linux could be booting on a Mac near you in 2022.
…"