hojnikbThings like emulation exists and most software can be ported over to ARM without too much difficulty.

Emulation is slow. Web servers and databases can't afford to be slow. Emulation is a last-resort that nobody wants because it is trash.

hojnikbOther than some very purposeful machines (think CNCs) everything else can easily switchover to arm

Apparently you didn't read the part of the post where I said "think".

I ask the super experts that clearly populate this comment section. Isn't it possible to just build a hybrid processor, with both arm and X86 cores, maybe with a unified cache, and adapt the os to make use of this hybrid structure like a big-little configuration?

hojnikbThings like emulation exists and most software can be ported over to ARM without too much difficulty.

Emulation is not realistic for real-world usage. Even today, emulating a tiny ARM CPU on a powerful x86 CPU runs very slow, just imagine doing the opposite. If you are intending to do any kind of hardware emulation that is "fast enough" to use real productive software, then you've actually implemented x86 and it's no longer emulation, and would defeat the purpose of going with ARM in the first place.

Compiling most software with ARM using the base feature set isn't hard, but the performance would be terrible. If the application will do anything requiring real performance, you'll have to rely on application specific acceleration, which means you may have to write specific code paths for each feature set you want to target, requiring an immense amount of low-level code to get decent performance. The main strength of x86 is that it's inherently faster in generic performance, and the ISA is very stable across microarchitectures (in contrast to custom ARM designs which are very different), which makes it easy to make performant code across multiple generations of CPUs from both Intel and AMD (and VIA :D).

AssimilatorEmulation is slow. Web servers and databases can't afford to be slow. Emulation is a last-resort that nobody wants because it is trash.

For old software, emulation very well might be fine if it otherwise means using archaic hardware to run the OS to drive it. A great example is playing Xbox 360 games on Xbox One. It runs fine and those games are compiled for IBM's POWER u-arch, but is emulated on an x86 CPU. It really depends on what you're doing.

chris.londonI am not saying you are completely wrong, but you are comparing a Geekbench 4 result to Geekbench 5.

Ok, let's move to GB 5 and "comparable" TDP then.

Geekbench 5

Snapdragon 865 (2.5 W - 3 W): 3464 Multi-core, 934 Single-core, 3171 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html
Apple A13 Bionic (2.5 W - 3 W): 3338 Multi-core, 1288 Single-core, 6273 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html

Intel Atom x5-Z8350 (2W): 523 Multi-Core Score 168 Single-Core Score browser.geekbench.com/processors/1969
Intel Core i7-1060G7 (9 W): 2151 Multi-core, 1268 Single-core browser.geekbench.com/v5/cpu/1372971

hojnikbYou need to wake up in 2020, sir. Things like emulation exists and most software can be ported over to ARM without too much difficulty. Other than some very purposeful machines (think CNCs) everything else can easily switchover to arm, if and when performance will be there. It's that much easier for apple, since it has tighter control on their platform.

Correct. Google Play Store has millions of apps, so the argument about software is invalid..

AssimilatorFFS. Always the same idiots with the same so-called arguments.

IT'S NOT JUST ABOUT PERFORMANCE YOU MUPPETS. It's about HAVING SOFTWARE THAT PEOPLE RELY ON running an Arm. It's about the fact that x86 has an enormous, decades-old library of applications, many of which no longer have source code available, many of which are relied upon by massive organisations for their day-to-day operations.

S**tty Apple laptops that going to do nothing more than run a web browser and MS Office and various other trash from the iStore do not have any of those concerns. Nobody buys or uses Apple latops to do real software development (sorry JS users, VS Code and NPM aren't real software, they can be run on a toaster). So the x86 ecosystem isn't something that Apple needs to care about, just their own, which is small and has been moving to Arm for the last decade so that Apple can offer One OS To Rule Them All across all their devices, not just phones. Thus it is trivial for Apple to replace x86 CPUs with Arm CPUs in their "laptops" that are actually iPads with permanently-attached trackpads and keyboards.

And stop posting that stupid AnandTech article. It's essentially free advertising for Amazon/Graviton2 because it contains ZERO real-world benchmarks. NONE. Amazon can provide all the Arm hardware in the world, the end-user is still responsible for the actual software running on those instances and guess what, you aren't going to find very much cloud-relevant software that compiles to Arm because nobody gives a s**t about Arm in the cloud. There are no big-name webservers with Arm binaries, there are no big-name databases with Arm binaries, and there is nobody in those projects rewriting their code to be performant on Arm because there is no incentive for them to do so.

x86 is not going away, ever. Arm is not going to displace it, ever. Stop dreaming, start thinking for a change.

I wouldn’t go quite so far as to say “not going to displace it, ever.” But for the foreseeable future, yes - you are correct.

ARM, IBM Power, or any other architecture can be faster in some scenario. We can look at a benchmark and proclaim, “Oh, so much faster than x86!” It doesn’t matter at all. The world runs on x86 OSes and software with a slim, few exceptions.

Anyone who says it’s easy to port or recompile software has never worked development or is grossly understating the effort required for the sake of their argument. You can easily port or rewrite some basic tool with 1000 lines and no dependencies but imagine business software with 10k or 100k lines of code (or more)? What about all the 20-30 years of technical debt baked into a lot of big enterprise software packages? Hell, most of the accounting software I’ve worked with is reliant on Windows and would need to be recreated from the ground up to even run on Linux.

ARM *could* become a dominant desktop or server architecture. But it’s going to take many years for that to be a possibility.

SamuelLI wouldn’t go quite so far as to say “not going to displace it, ever.” But for the foreseeable future, yes - you are correct.

ARM, IBM Power, or any other architecture can be faster in some scenario. We can look at a benchmark and proclaim, “Oh, so much faster than x86!” It doesn’t matter at all. The world runs on x86 OSes and software with a slim, few exceptions.

Anyone who says it’s easy to port or recompile software has never worked development or is grossly understating the effort required for the sake of their argument. You can easily port or rewrite some basic tool with 1000 lines and no dependencies but imagine business software with 10k or 100k lines of code (or more)? What about all the 20-30 years of technical debt baked into a lot of big enterprise software packages? Hell, most of the accounting software I’ve worked with is reliant on Windows and would need to be recreated from the ground up to even run on Linux.

ARM *could* become a dominant desktop or server architecture. But it’s going to take many years for that to be a possibility.

We don't know what's after AMD's Zen 5, and there is extremely high uncertainty about Intel's future manufacturing. What if their current "++" node is the last one and they never move to 5nm, 3nm, and the dreamt 1.4nm?

phanbueyI don't think this product is intended for people who know any of the words you just said.

This needs clarifying...

efikkanARM is a nightmare to support for computers in general, it's only suited for embedded devices or devices within a tight eco-system.

That's not entirely true.
ARM is well suited for some server tasks as well.
It's really a question of having software designed specifically for the architecture.

We live in a x86 world. That's why ARM seems like a nightmare to use. But if we went this route 30 years ago, we would have a complete and robust ARM ecosystem.
If we really focused on migrating to ARM, it may become a fully featured universe in maybe a decade.
But we could also focus on improving and optimizing x86 - getting to ARM's efficiency without any big sacrifices.

x86 and ARM are different, but both can do what CPUs are expected to do. It's not a question of what moving to ARM can give us. It's a question of: why bother at all?

R0H1TI'm not sure why I keep reading these long prose on x86 & its apparent/inherent superiority?

Because most of our software today is written for x86. Because moving to ARM would be - as @effikan noticed - a nightmare.
ARM is more efficient in some scenarios but x86 is better in other.

x86 is built as a universal architecture. ARM is often custom made for a particular system.

ARM has big.LITTLE which is responsible for a lot of it's efficiency, but Intel's hybrid core architecture will close the gap.

Fact 1 ~ there's nothing that competes with Apple's highest end chip in a similar power envelope.

But that chip is designed to work well with other hardware Apple uses and optimized with iOS.
It's a comfort most of the computing world doesn't have. x86 is universal and flexible - the cost is efficiency. No other away.

You've mentioned Graviton, which is a great example of that optimization. It's a chip designed to be very efficient in AWS infrastructure and - probably - with Amazon's own Linux.
Phoronix tested the first generation chips:
www.phoronix.com/scan.php?page=article&item=graviton-linux-os&num=1
Amazon Linux was the best allround. RHEL was the worst, but that's the most popular server distribution.
So if you have the comfort of using a distribution, Graviton could be an interesting choice. If you don't (your software runs only on Red Hat family), x86 will remain more efficient.
And of course we don't have a fully features ARM Windows yet.

Another good example would be IBM's POWER and Z architectures.

No, we live in an ARM world where every single of your smartphones and tablets runs on ARM-based SoCs.

And there is more healthy competition in the ARM market - Qualcomm, Apple, Samsung, MediaTek.
While in the x86, there are AMD, Intel and only left maybe VIA...

ARFCorrect. Google Play Store has millions of apps, so the argument about software is invalid..

ARFx86 is inferior. It is slower and yet consumes 10 times ! more power :laugh:

ARFNo, we live in an ARM world where every single of your smartphones and tablets runs on ARM-based SoCs.

You know... maybe I should ask moderators for permission to call you stupid. As a fact, not an offence - that way it probably wouldn't be against the netiquette.

ARFNo, we live in an ARM world where every single of your smartphones and tablets runs on ARM-based SoCs.

And there is more healthy competition in the ARM market - Qualcomm, Apple, Samsung, MediaTek.
While in the x86, there are AMD, Intel and only left maybe VIA...

Just because it's a comparatively new environment. But in the years the actors will shrink just like it happened for the x86 environment. helathy competition is not a stable state of things with our kind of market...

MirkoskjiJust because it's a comparatively new environment. But in the years the actors will shrink just like it happened for the x86 environment. helathy competition is not a stable state of things with our kind of market...

I think it's because of Intel's predatory instincts to try to kill every single of the other x86 suppliers.
Too arrogant for a simple supplier to IBM.

ARFI think it's because of Intel's predatory instincts to try to kill every single of the other x86 suppliers.
Too arrogant for a simple supplier to IBM.

I think it's every enterprise's innate predatory instinct to try to kill every single of the other potentially competing enterprises, or inglobate them. Once you start having an egemonic position in the market it's a spontaneous behaviour.

MirkoskjiI think it's every enterprise's innate predatory instinct to try to kill every single of the other potentially competing enterprises, or inglobate them. Once you start having an egemonic position in the market it's a spontaneous behaviour.

It's illegal in every context - law, ethics, morals, social contribution, free market economy, etc...

ARFIt's about time.

askscience/comments/opg0h
Apple always innovates. They were the first with Retina-class displays, now they will be the first ditching x86 altogether, they will be the first to introduce an unlimited detail camera in their iPhones. :eek:
wccftech.com/exclusive-future-iphone-to-feature-unlimited-detail-camera-for-pin-sharp-digital-zoom-at-any-setting/

" retina display" is just a marketing term for Pixel density, it's not a certification or a standard, it's not a new display technology... So how is this an "innovation"?

ARFIt's illegal in every context - law, ethics, morals, social contribution, free-market economy, etc...

Yes, that's why the hegemonic position is usually shared by two entities that can thus try to do cartels. Or enterprises are split into two or more different entities (that again will try to do cartel and so on and so forth). By the time the law is enacted the damage has already been done and diversity has been already almost completely engulfed by the few remaining positions. But I will self stop. off-topic.

notbThat's not entirely true.
ARM is well suited for some server tasks as well.

Sure, there are some that prefer core count or power per core etc.

notbIt's really a question of having software designed specifically for the architecture.

We live in a x86 world. That's why ARM seems like a nightmare to use. But if we went this route 30 years ago, we would have a complete and robust ARM ecosystem.
If we really focused on migrating to ARM, it may become a fully featured universe in maybe a decade.
But we could also focus on improving and optimizing x86 - getting to ARM's efficiency without any big sacrifices.

This is not accurate. The ARM eco-system have been around for a long time, compilers and tools are very mature, and there are probably more ARM devices than x86 in the world, by far. There are also many more companies and researchers working on the ARM design, so if anything the ARM eco-system is in a stronger position.

There are some fundamental facts that many fail to realize;
Firstly, RISC is inherently slower, it needs more operations to do the same work. In a world where we are scaling towards the "clock wall" and the "memory wall", such inherent disadvantages will only increase. In order for ARM to compete, it needs to become more "CISC", as x86 keeps increasing the work done per instruction.
Secondly, the way most low-power ARM or MIPS chips today achieve performance is through ASICs which do all the heavy lifting, not through their core instruction set. This is how your Blu-Ray player manages to play movies with a tiny slow CPU, or your phone manages to play YouTube videos. Such features varies a lot between ARM designs, and some, especially Apple, have a lot of their own. Applications then need to be coded specifically to utilize these, which is a nightmare to develop for, considering all the various ARM designs. This custom chip approach is a conscious decision, and is a strength for ARM in many ways, but it's not for the general computing market in PCs.

notbx86 and ARM are different, but both can do what CPUs are expected to do. It's not a question of what moving to ARM can give us. It's a question of: why bother at all?

Yes, it's a very important question.
The fact is that from an ISA standpoint, moving to ARM will not yield any substantial benefits, it will only yield a lot of performance loss. All modern x86 microarchitectures have already solved this "problem"; if you want higher performance or better energy efficiency, you design the microarchitecture accordingly, but the ISA remains the same. Having very high compatibility across platforms is one of the key strengths of the PC. Imagine for a moment if the PC market were dominated by a semi-custom ISA; Dell, HP, Lenovo, etc. all had their own CPUs that were slightly different software compatibility would be a pain, and the PC market would probably be much weaker as a whole. We sort of had this in the beginning before IBM "standardized" the PC.

The only thing to gain is easier licensing and the ability to customize, but the latter is only an advantage for embedded or specialized platforms.

notbARM has big.LITTLE which is responsible for a lot of it's efficiency, but Intel's hybrid core architecture will close the gap.

Sure, it's a part of it.
Other major parts are CPU front-end complexity, cache efficiency/bandwidth, ALU/FPU efficiency, etc.
As you see, most factors which make these designs more energy efficient have little to do with ISA; it's a design choice.

MirkoskjiIsn't it possible to just build a hybrid processor, with both arm and X86 cores, maybe with a unified cache, and adapt the os to make use of this hybrid structure like a big-little configuration?

Probably. But nobody is going to do that.

AquinusFor old software, emulation very well might be fine if it otherwise means using archaic hardware to run the OS to drive it. A great example is playing Xbox 360 games on Xbox One. It runs fine and those games are compiled for IBM's POWER u-arch, but is emulated on an x86 CPU. It really depends on what you're doing.

Yes, of course. But my statement is still true: nobody actually wants emulation, because it's slow and complex. And I was specifically talking about emulation in the context of Arm and x86.

ARFCorrect. Google Play Store has millions of apps, so the argument about software is invalid..

What does Google have to do with a topic about Apple?

notbYou know... maybe I should ask moderators for permission to call you stupid. As a fact, not an offence - that way it probably wouldn't be against the netiquette.

You don't need to ask for permission to state facts.

ARFOk, let's move to GB 5 and "comparable" TDP then.

Geekbench 5

Snapdragon 865 (2.5 W - 3 W): 3464 Multi-core, 934 Single-core, 3171 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html
Apple A13 Bionic (2.5 W - 3 W): 3338 Multi-core, 1288 Single-core, 6273 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html

Intel Atom x5-Z8350 (2W): 523 Multi-Core Score 168 Single-Core Score browser.geekbench.com/processors/1969
Intel Core i7-1060G7 (9 W): 2151 Multi-core, 1268 Single-core browser.geekbench.com/v5/cpu/1372971

These results are pretty interesting. Intel Atoms have compareable TDP but still their Multi-core performance is only 15% of the ARM-based counter-part, and Single-core performance is only 18% of an ARM-based counter-part.

That means Intel violates the EU standards for energy efficiency.

AssimilatorYes, of course. But my statement is still true: nobody actually wants emulation, because it's slow and complex. And I was specifically talking about emulation in the context of Arm and x86.

Sure, but if that's your only option, you take it. Remember when Apple used to have 68k and the transition to IBM's POWER? How about going from POWER to x86? Eventually older software was phased out. Businesses who used Apple weren't given an option. You had some level of cross compatibility with fat binaries for the 68k/PowerPC days and universal binaries for PowerPC/x86, but Apple eventually (for the x86 transition,) did have emulation for older software which was eventually phased out. It worked and it worked well enough.

The bottom line is that if you're dependent on software that you can't control or don't have a service agreement for, then it's a ticking time bomb. It will eventually explode if you have no support. This is something Apple has done before, several times, and none of those times resulted in their demise. This a nice way of telling businesses, "Hey, we're about to change everything and we'll support you to some extent for so long, but you're expected to get your shit together and keep up with the times."

Honestly, if Apple wants to move forward and not be constrained by the past, this is the best way to do it, but if we always stuck with the "well, we've always done it this way," mentality, there would never be truly substantial progress.

Everyone talking about server, when a huge chunk of apple clients are content creator, designers, artist, so I'm just wondering what this move will mean for application like photoshop, after effect, première etc... ( and nvm the few x86 games ported on macos)

FlankerI'm holding my breath until I see the exact same application (hopefully the exact same source code), compiled to respective architectures, then compare their performance and power usage

From what I know so far from existing devices, ARM CPU will wipe the floor with Intel's in perf/Watt but no way any ARM CPU can replace an 8core/16 thread CPU in the high-end laptops destined to creators, scientists and gamers.

AquinusSure, but if that's your only option, you take it. Remember when Apple used to have 68k and the transition to IBM's POWER? How about going from POWER to x86? Eventually older software was phased out.

Move to PowerPC happened when:
a) software wasn't as "globalized" as it is since the late 90s, i.e. you were fine with a text editor, mail client, calculator, compiler, snake-like game - mostly stuff provided with the OS (phones looked like that until mid 2000s),
b) PowerPC had potential to become a mainstream architecture.

But the IT world changed. We all started using pretty much the same OS, the same office suite, the same photo editor. And, obviously, the same games - bigger, made by 3rd party studios.
And it was all made for x86.

So when Apple decided to go x86 in ~2005, it wasn't just another choice of architecture. They were fixing a mistake made 15 years earlier. And it changed everything, because suddenly Macs were usable for so many use scenarios. They became popular with coders, scientists, analysts. In business as well (especially when MS Office arrived).

The way I see it: if they really cracked x86 64-bit emulation (including GPU support) with no or little performance penalty, they can use whatever architecture they want - no one will care.
But if they launch a MacBook that can't run MS Office, Adobe Photoshop, Matlab, Visual Studio and so on... they'll end up with just a posh linux laptop with subpar repository. And the number of linux laptops offered today tells us exactly how big that market is.

efikkanThis is not accurate. The ARM eco-system have been around for a long time, compilers and tools are very mature, and there are probably more ARM devices than x86 in the world, by far. There are also many more companies and researchers working on the ARM design, so if anything the ARM eco-system is in a stronger position.

But that's a perspective of a PC enthusiast or coder. Yeah, you can compile for ARM, you can make nice RaspberryPi projects. You can do many things.

As of today you can't migrate most of the normal, everyday tasks we do: both professionally and casually (e.g. gaming).
As long as we use locally installed software, the mainstream program availability dictates which platform sells and which doesn't. People won't suddenly replace MS Office with LibreOffice because an ARM laptop will have double the battery life.

Maybe Apple is betting on a cloud strategy, when these MacBooks will merely be terminals. Fine. But that would mean only the weaker models (MacBook Air) will get an ARM chip - essentially becoming an iPad Pro with a keyboard. MacBook Pro and other Macs will keep x86 to run the software locally or slowly vanish from the lineup.

As you see, most factors which make these designs more energy efficient have little to do with ISA; it's a design choice.

Which is what I've mentioned earlier: the reason why ARM chips shine in some use cases is because they were made specifically for that scenarios/clients. x86 focused on being as universal as possible.
But there is no reason why big x86 clients wouldn't work with Intel and AMD on customized chips. Especially when the world moves towards mobile and cloud, so less chip variants are needed.

ARFOk, let's move to GB 5 and "comparable" TDP then.

Geekbench 5

Snapdragon 865 (2.5 W - 3 W): 3464 Multi-core, 934 Single-core, 3171 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html
Apple A13 Bionic (2.5 W - 3 W): 3338 Multi-core, 1288 Single-core, 6273 Compute www.pcworld.com/article/3490156/qualcomm-snapdragon-865-benchmark-performance.html

Intel Atom x5-Z8350 (2W): 523 Multi-Core Score 168 Single-Core Score browser.geekbench.com/processors/1969
Intel Core i7-1060G7 (9 W): 2151 Multi-core, 1268 Single-core browser.geekbench.com/v5/cpu/1372971




Correct. Google Play Store has millions of apps, so the argument about software is invalid..

Those results better be well optimized for AMD64 and AVX2/AVX-512, else it's just ARM propaganda.

GoldenXThose results better be well optimized for AMD64 and AVX2/AVX-512, else it's just ARM propaganda.

Sure... :kookoo:

Ampere Altra is the first 80-core ARM-based server processor
venturebeat.com/2020/03/03/ampere-altra-is-the-first-80-core-arm-based-server-processor/

80 cores at 210-watt TDP and faster than AMD's 64-core and Intel 56-core server CPUs.

holyprofFrom what I know so far from existing devices, ARM CPU will wipe the floor with Intel's in perf/Watt but no way any ARM CPU can replace an 8core/16 thread CPU in the high-end laptops destined to creators, scientists and gamers.

They can do chiplets. Imagine 20 ARM chiplets in 60-watt.