Linus Torvalds says ARM just doesn't look like beating Intel

[System Error Message]

In the past before intel went into the manycore CPU design the RISC CPU world and also many businesses used the sun sparc CPU. It was capable of handling up to 8 threads per core (intel only does 2) and had 16 core variants with support for many sockets when intel only had dual core CPUs. The platform was certainly interesting and it was used in many mission critical applications as it was much better at number crunching than a multi socket intel xeon coupled with the fact that sun did design high end GPUs specifically for the rendering bit of the application. It was a good platform.

What limited it was mainly that only sun solaris will work on it and in some lucky cases debian linux. At one point microsoft did have a version of windows for sparc but only 1 version. What killed it was pretty much a bunch of bad decisions and lack common platforms for it in a world where people's choices were either microsoft or linux (not unix). At that time OSX hasnt even come out yet, it was stlll the macintosh. The hardware themselves also werent cheap, as x86 became a more cost effective platform.

ARM is still expanding and its actually starting to replace the x86 platform. the cost of an ARM vs intel based chrome laptop for instance while providing all the same functionality. I wouldnt count out the performance of ARM just yet, unlike the x86 ARM doesnt include many hardware units that x86 does to accelerate some tasks (like AES) but more and more ARM platforms are coming up with these units so the gap is closing.

ARM will never replace x86 in the desktop world as ARM isnt a big CPU like intel, it lacks many PCIe lanes, bus bandwidths and so on. What i've seen is that less and less people are having desktops, using laptops, more on portability and lower price as it can perform the same tasks. But ARM is starting to replace x86 as the main platform in the house.

The other thing about the RISC platform is that unlike with x86 it shows a clear distinction between good and bad coders show. Being good at coding i can make my apps run well on ARM CPUs that are so slow that they can only operate android on the basic level so i dont need them to clock up but also take advantage of faster hardware when it is available.

[rollatorwieltje]

- What is the problem anyway? Neither ARM or x86 are going to dissapear in the next 20 years because both are bound by the same limitations which come with making chips.

I'm not entirely sure, but I suspect he isn't pleased with how ARM platforms are designed. If you look at an x86 machine, all the bootstrapping and hardware configuration stuff is done by the platform itself. On ARM platforms this isn't really the case, you often have to rely on closed binary blobs in your own software to do this. On x86 it's a closed binary blob as well, but it's on the platform itself (i.e. BIOS chip), so it's less of a concern. The entry point for your own software is always the same, doesn't matter who made the machine.
Just look at how annoying it is to run a random OS on an ARM platform. You always need a kernel specifically configured for that device. It's starting to improve with DeviceTree, but it's still far from perfect. On x86 that's not a concern, grab a Windows or Linux disk and it will install and run on every random machine.

[dannyf]

"On x86 it's a closed binary blob as well, but it's on the platform itself (i.e. BIOS chip), so it's less of a concern."

So you think he is blaming the cpu for the lack of an uniform bios for the CPU?

Then he couldn't have been more irrational.

[rollatorwieltje]

"On x86 it's a closed binary blob as well, but it's on the platform itself (i.e. BIOS chip), so it's less of a concern."

So you think he is blaming the cpu for the lack of an uniform bios for the CPU?

Then he couldn't have been more irrational.

He doesn't, look at his quotes. He even mentions the problem is not the CPU core or the instruction set.

“The instruction set and the core of the CPU is not very important,”
"What matters is the infrastructure around the instruction set. x86 has it and has it at a lot of levels.”
“as a hardware platform it is still not very pleasant to deal with.”
“It does not have the same unified models around the instruction set as we do in the PC space, but it is getting better.”

Also he is doing the interview with someone from Linaro, an organisation that attempts to standardize ARM platforms. DeviceTree is part of the solution. DeviceTree.org was started by Linaro.

[Karel]

- For starters I'm not convinced Linus is right about anything because he says so and sometimes I get the feeling he should leave because he is holding progress back.

Linux made a lot of progress and is now the most used OS in the world thanks to people like Torvalds.
If somebody or a company thinks they can do better, they are free to fork Linux and add their improvements.

[nctnico]

- For starters I'm not convinced Linus is right about anything because he says so and sometimes I get the feeling he should leave because he is holding progress back.

Linux made a lot of progress and is now the most used OS in the world thanks to people like Torvalds.

In the beginning yes but now it seems to have stalled mostly due to never ending backward incompatibilities which (for starters) prevents hardware manufacturers to create drivers. In companies it is rare to see the founder growing a the business into a large company because running a small company takes totally different skills than running a multi-million dollar firm. Same with Linux. Linus did great with starting it but the best thing he could do for Linux right now is leave it alone so others can take it to the next level.

[R005T3r]

The real stupid thing about IoT is that they should have put Security as their top priority.They didn't. As a result anyone with a bit of skill can hack the damn thing and when it's in it, it's compromised. No matter what.
Just think about the whole idea to have a pacemaker with security flaws: anyone can kill you.

[System Error Message]

This is a different Linus as the Linus in LTT used to be a sales person for PCs not programming/development for linux.

To sum it up, ARM will never beat x86 (not intel as theres AMD and VIA too), not in the direct sense of the tasks that x86 will continue to perform but ARM will beat x86 in common usage. Infact even russia is throwing out x86 and going with ARM because they can design their own, make their own and use their own software (the open source thing for security).

IoT is not relevant as plenty also use intel's low power CPU. You can find raspberry pi compatible boards that use quad core atoms. To me IoT as a concept is flawed because of the dependence of connectivity with the outside world rather than with the network created by user.

[David Hess]

but ARM will never take over the desktops.

ARM has demo'd server chips and there are ARM-based server boards out there. Having lived with Xeon for the last 10 or so years, I am more and more convinced that the future of (personal) computing is in those multi-core CPUs where individual cores are reasonably but not terribly fast - so they provide adequate performance on older applications. Such CPUs can take on dynamic load situations - sudden onslaught of multiple tasks running at the same time. This approach provides robustness of performance, unlike single core CPUs or CPUs of low core counts.

So if you project 10 - 15 or 20 years down the road, it wouldn't surprise me if ARM or ARM-like architectures become a mainstay on the desktop / server market.

And you have just touched on the problem Linus identified.  It is not x86 versus ARM.  It is one x86 standard versus many ARM standards which are taking x86 on individually.

When x86 took over the RISC workstation and server markets from below, x86 was larger than any individual RISC system.

[Karel]

- For starters I'm not convinced Linus is right about anything because he says so and sometimes I get the feeling he should leave because he is holding progress back.

Linux made a lot of progress and is now the most used OS in the world thanks to people like Torvalds.

In the beginning yes but now it seems to have stalled mostly due to never ending backward incompatibilities which (for starters) prevents hardware manufacturers to create drivers.

Stalled? You must be kidding, right? Linux development, and specially driver development is growing and better than ever.

In companies it is rare to see the founder growing a the business into a large company because running a small company takes totally different skills than running a multi-million dollar firm. Same with Linux. Linus did great with starting it but the best thing he could do for Linux right now is leave it alone so others can take it to the next level.

Those are your arguments? Really?

Like I said, everybody is free to fork Linux and take it to the next level but that's not happening
because "the watchers always know better than the doers".

[Karel]

And you have just touched on the problem Linus identified.  It is not x86 versus ARM.  It is one x86 standard versus many ARM standards which are taking x86 on individually.

When x86 took over the RISC workstation and server markets from below, x86 was larger than any individual RISC system.

Exactly this.

[Karel]

The real stupid thing about IoT is that they should have put Security as their top priority.They didn't. As a result anyone with a bit of skill can hack the damn thing and when it's in it, it's compromised. No matter what.

This plus the fact that all those IoT gadgets will not receive any updates after a two years (if any!) and you have the recipe for failure/disaster.

[timb]

In the past before intel went into the manycore CPU design the RISC CPU world and also many businesses used the sun sparc CPU. It was capable of handling up to 8 threads per core (intel only does 2) and had 16 core variants with support for many sockets when intel only had dual core CPUs. The platform was certainly interesting and it was used in many mission critical applications as it was much better at number crunching than a multi socket intel xeon coupled with the fact that sun did design high end GPUs specifically for the rendering bit of the application. It was a good platform.

What limited it was mainly that only sun solaris will work on it and in some lucky cases debian linux. At one point microsoft did have a version of windows for sparc but only 1 version. What killed it was pretty much a bunch of bad decisions and lack common platforms for it in a world where people's choices were either microsoft or linux (not unix). At that time OSX hasnt even come out yet, it was stlll the macintosh. The hardware themselves also werent cheap, as x86 became a more cost effective platform.

ARM is still expanding and its actually starting to replace the x86 platform. the cost of an ARM vs intel based chrome laptop for instance while providing all the same functionality. I wouldnt count out the performance of ARM just yet, unlike the x86 ARM doesnt include many hardware units that x86 does to accelerate some tasks (like AES) but more and more ARM platforms are coming up with these units so the gap is closing.

ARM will never replace x86 in the desktop world as ARM isnt a big CPU like intel, it lacks many PCIe lanes, bus bandwidths and so on. What i've seen is that less and less people are having desktops, using laptops, more on portability and lower price as it can perform the same tasks. But ARM is starting to replace x86 as the main platform in the house.

The other thing about the RISC platform is that unlike with x86 it shows a clear distinction between good and bad coders show. Being good at coding i can make my apps run well on ARM CPUs that are so slow that they can only operate android on the basic level so i dont need them to clock up but also take advantage of faster hardware when it is available.

I wouldn't be surprised if we saw an ARM based laptop running OS X from Apple in the next few years. Sell it for ~$600. They've got the software infrastructure in place to do it; basically they could flip a switch in Xcode and have it produce x64/ARM64 fat binaries. I bet most existing applications on the Mac App Store would require little to no code modifications for it to work.

For 90% of the people that use a computer, it would fit their needs perfectly.

[System Error Message]

There may be many implementations for ARM but they all share the same instruction set. The problem is that you have to optimise for specific processor.

With ARM good and bad coders are revealed.

What made people move from sparc to x86 was simply to do with cost as sparc was way ahead of x86 sold only in the server segment for critical applications after microsoft stopped supporting it as many people were using it on their desktops. The sparc CPU is a big chip despite being RISC architecture but they gave more than what intel could at that time even in workstations from the 64 bit architecture, 64 bit PCI, lots of registered memory slots, FCAL (this was before sata).

[R005T3r]

The real stupid thing about IoT is that they should have put Security as their top priority.They didn't. As a result anyone with a bit of skill can hack the damn thing and when it's in it, it's compromised. No matter what.

This plus the fact that all those IoT gadgets will not receive any updates after a two years (if any!) and you have the recipe for failure/disaster.

The recipe to disaster? I'm shocked that they permit to make medical equipment with it! Imagine a pacemaker with IoT: someone can kill you by sitting on a desk and writing a few commands...

[madires]

The real stupid thing about IoT is that they should have put Security as their top priority.They didn't. As a result anyone with a bit of skill can hack the damn thing and when it's in it, it's compromised. No matter what.

This plus the fact that all those IoT gadgets will not receive any updates after a two years (if any!) and you have the recipe for failure/disaster.

Just take all the security issues with SOHO routers (D-Link, Netgear and so on) in the last few years and multiply them by 10 or 20. And IoT has just started. In three years the factor will be maybe 1000. The big DDoS attacks we've seen recently are mostly driven by IoT and SOHO routers. The largest one was about 1.1 TBit/s (directly, no amplification), and speculations are that the IoT bot net could create 2 TBit/s direct traffic. This is scary!

[3db]

ARM will never ever replace intel x86 arch... it simply won't happen. what is already happening is that certain activities like reading mails and browsing the web are moved from desktops to mobile devices where ARM is prevalent (and since Intel ditched their mobile platforms ARM is the one and only)... but ARM will never take over the desktops. it simply won't happen because of the software. if there is no windows and office for ARM, then it means there will be no ARM for mainstream desktop. for server market it's the very same... some services might/will move to specialized ARM platform (e.g. WEB hosting could be moved to a massive multi-core ARM platform). but in general... as long as there is no windows servers, ms sql, oracle , SAP and many others  for ARM... there will be no significant market share for ARM.

They HAVE market share of the tablet/phone market.
It's the desktop market that's in decline.

3DB 

[System Error Message]

The real stupid thing about IoT is that they should have put Security as their top priority.They didn't. As a result anyone with a bit of skill can hack the damn thing and when it's in it, it's compromised. No matter what.

This plus the fact that all those IoT gadgets will not receive any updates after a two years (if any!) and you have the recipe for failure/disaster.

The recipe to disaster? I'm shocked that they permit to make medical equipment with it! Imagine a pacemaker with IoT: someone can kill you by sitting on a desk and writing a few commands...

In that case full steam ahead with IoT, than i can kill however i want easily 

The problem with IoT is the way people design and monetise it. There are many existing things that are constantly being developed and secure that allow for proper checking and for security. Perhaps i should make my own IoT just to prove it.

[linux-works]

I contracted at cisco for a while and did some work for an IoT group.

I asked the same kinds of questions: what is your security 'story'?

(crickets chirping is heard)

they didn't have an answer for me.  they didn't even care.  and the more I asked, the more I was ignored.

if big old cisco doesn't give a damn, no one else will, very likely.

IoT is doomed.  in a way, its a good thing its doomed.  its mostly about marketing stealing your info and phone-home attempts.  WAN access for IoT seems like a failure anyway; and its mostly companies that want this, not us users.  for me, I'd prefer ALL my IoT things sit on my lan and NEVER reach out to a public network for any reason.  home automation does not need remote access; not really.
'

[BobC]

I have two $99 10" tablets with 720p displays and 4GB RAM.  One is quad-core ARM+Linux (well, rooted Android), the other is dual-core x86+Win10.  Both are 32-bit only, and both run Chrome.

The ARM tablet is noticeably faster when web browsing.  The only thing the x86 does better is video playback, but only by a tiny bit.  I don't play games, so I haven't tested that.

I believe the only reason the x86 tablet even exists is to run Win10: It was purely a software decision, a marketing decision, not an engineering decision. If full Win10 ran on ARM, the x86 tablet wouldn't exist (but there would be two ARM tablets, one running Linux/Android, and one running Win10).

The goal in personal systems is to get the most computing done per watt.  Fewer watts means I can reduce or eliminate heatsinks, simplify power supplies, reduce size, reduce cost, and the list goes on.  It ripples through to more bang for the buck for the entire system.  ARM not only delivers more computing per watt, it also delivers more computing per dollar.

In the near future I believe Intel will only rule when installed in large boxes (high-end desktops, server racks) where only the large scale will let them prevail on the basis of compute power per watt.  Everything with a battery will soon be all ARM.

Why?  Early next year MS will have an ARM Win10 Mobile release with x86 emulation.  Assuming that goes well, I would not be at all surprised to see a full desktop Win10 release for Aarch64.  With much of .NET officially supported on x86 Linux, and with .NET Core already on ARM, I believe a full ARM build of .NET is simply waiting for release.  The path exists and is being traveled.

Personal use of x86 will die as MS support for ARM grows.  When x86 is no longer the exclusive Win10 desktop target, Intel will have no hardware leg to stand on in the commodity personal space. All that will be left for Intel will be the high-end desktop space and the cloud.

To hell with you, Torvalds.  What you should worry about is losing Linux installations when full Win10 comes to ARM!

[coppice]

I believe the only reason the x86 tablet even exists is to run Win10: It was purely a software decision, a marketing decision, not an engineering decision.

Most x86 win10 tablets exists because Intel supplied the SoC for less than zero dollars, and Microsoft supplied licenses for free. Once the crazy subsidies which brought this about stopped, the manufacture of most x86 tablets also stopped. Most of the x86 tablets actually dual boot - Win 8.1 or 10 and Android. I think most people use them as Android tablets.

[madires]

I contracted at cisco for a while and did some work for an IoT group.

I asked the same kinds of questions: what is your security 'story'?

(crickets chirping is heard)

they didn't have an answer for me.  they didn't even care.  and the more I asked, the more I was ignored.

That's disturbing, but I'm not really surprised. On some of their SOHO products (like the SPA VoIP ATA series) you can't enable https because it wouldn't work with any modern browser (firmware 1.3.5 from mid of last year). It's cheap stuff, and IosT (s for shitty) has to be cheap too. In case of Cisco it poorly reflects on their brand.

IoT is doomed.  in a way, its a good thing its doomed.  its mostly about marketing stealing your info and phone-home attempts.  WAN access for IoT seems like a failure anyway; and its mostly companies that want this, not us users.  for me, I'd prefer ALL my IoT things sit on my lan and NEVER reach out to a public network for any reason.  home automation does not need remote access; not really.

I'm not sure about average Joe. Take PC gaming for example. Users are buying games which don't run without an online gaming platform. When the publisher shuts down that platform for whatever reason the game is worthless. It's the same for any cloud based IoT device, also creating tons of e-junk. Unfortunately most people don't see this.

Edit: The DHS gets involved: https://www.techdirt.com/articles/20161122/08025036105/homeland-security-issues-strategic-principles-securing-internet-broken-things.shtml

[technix]

In the past before intel went into the manycore CPU design the RISC CPU world and also many businesses used the sun sparc CPU. It was capable of handling up to 8 threads per core (intel only does 2) and had 16 core variants with support for many sockets when intel only had dual core CPUs. The platform was certainly interesting and it was used in many mission critical applications as it was much better at number crunching than a multi socket intel xeon coupled with the fact that sun did design high end GPUs specifically for the rendering bit of the application. It was a good platform.

What limited it was mainly that only sun solaris will work on it and in some lucky cases debian linux. At one point microsoft did have a version of windows for sparc but only 1 version. What killed it was pretty much a bunch of bad decisions and lack common platforms for it in a world where people's choices were either microsoft or linux (not unix). At that time OSX hasnt even come out yet, it was stlll the macintosh. The hardware themselves also werent cheap, as x86 became a more cost effective platform.

ARM is still expanding and its actually starting to replace the x86 platform. the cost of an ARM vs intel based chrome laptop for instance while providing all the same functionality. I wouldnt count out the performance of ARM just yet, unlike the x86 ARM doesnt include many hardware units that x86 does to accelerate some tasks (like AES) but more and more ARM platforms are coming up with these units so the gap is closing.

ARM will never replace x86 in the desktop world as ARM isnt a big CPU like intel, it lacks many PCIe lanes, bus bandwidths and so on. What i've seen is that less and less people are having desktops, using laptops, more on portability and lower price as it can perform the same tasks. But ARM is starting to replace x86 as the main platform in the house.

The other thing about the RISC platform is that unlike with x86 it shows a clear distinction between good and bad coders show. Being good at coding i can make my apps run well on ARM CPUs that are so slow that they can only operate android on the basic level so i dont need them to clock up but also take advantage of faster hardware when it is available.

I wouldn't be surprised if we saw an ARM based laptop running OS X from Apple in the next few years. Sell it for ~$600. They've got the software infrastructure in place to do it; basically they could flip a switch in Xcode and have it produce x64/ARM64 fat binaries. I bet most existing applications on the Mac App Store would require little to no code modifications for it to work.

For 90% of the people that use a computer, it would fit their needs perfectly.

Apple have long unified the code base of macOS and iOS - they just need to ship an ipsw file for iPad Pro with a full installation of OS X in it, and ship a new release of Xcode enabling the existing aarch64 toolchain on macOS SDK, and you get an ARM-based laptop running OS X in the existing iPad Pro hardware, and app developers will get the apps ready in about an hours of time.

And you got it right about the code - it is one recompile away from universal compatibility. Apple have done quite a good job at keeping the software layers coherent across different hardware platforms so far, so it is pretty much that as long as you keep clear of lower layers of IOKit, the code is identical across all platforms.

[nctnico]

Edit: The DHS gets involved: https://www.techdirt.com/articles/20161122/08025036105/homeland-security-issues-strategic-principles-securing-internet-broken-things.shtml

They are not the only ones. Last week I've read an article that some Dutch politicians are saying IoT devices should pass a security test before being allowed on the market.

[ataradov]

IoT devices should pass a security test before being allowed on the market.

I agree. Do they know what kind of test? Because the same can be applied to OSes and other stuff too. How did we not think about that ourselves? Clearly Dutch politicians are smart dudes.