SoftBank Completes Acquisition of ARM

[btarunr]

SoftBank today announced that it has concluded the acquisition of British chipmaker ARM, in a USD $31 billion purchase. The acquisition was first announced in July 2016. Following this, ARM will be de-listed from the London Stock Exchange from the 6th of September 2016. Although ARM is a CPU architecture designer with a mere $1.5 billion in revenue last fiscal, and licenses the architecture to other SoC makers such as Qualcomm, Samsung, Apple, Huawei, etc., SoftBank is betting on ARM CPU architecture's emergence as the prime-mover of the IoT (Internet of things) revolution. This is SoftBank's largest tech acquisition following the $20 billion acquisition of American cellular network Sprint, and a $15 billion investment in Vodafone Japan.



View at TechPowerUp Main Site

 

[ZoneDymo]

would be nice to see ARM moving to pc cpu's as well but im sure someone more knowledgeable on this website will tell me how completely infeasible that is.

 

[silentbogo]

would be nice to see ARM moving to pc cpu's as well but im sure someone more knowledgeable on this website will tell me how completely infeasible that is.

ARM-based netbooks and nettops are already on the market, so why not?
The only problem is Microsoft's abandonment of Windows for ARM platform, but otherwise I wouldn't mind having a mini-PC powered by a modern 64-bit ARM CPU.
There are tons of high-performance parts available already: just look at Tegra X1 or Exynos 8890, which in most cases outperform both Intel and AMD SoC (even some U-versions of mobile Core i3/i5 CPUs!!!).

 

[Frick]

ARM-based netbooks and nettops are already on the market, so why not?
The only problem is Microsoft's abandonment of Windows for ARM platform, but otherwise I wouldn't mind having a mini-PC powered by a modern 64-bit ARM CPU.
There are tons of high-performance parts available already: just look at Tegra X1 or Exynos 8890, which in most cases outperform both Intel and AMD SoC (even some U-versions of mobile Core i3/i5 CPUs!!!).

It's pretty unfair and meaningless to compare them like that though, unless for very specific tasks. And Windows without all the avaliable Windows software is pretty pointless tbh. WinRT was not bad, just early. In a few years they might try again, but then x86 is so effecient so there's still little point to it. Android otoh is made for it and has a vast ecosystem. They just need better support for large monitors.

 

[Ferrum Master]

Linux?

Try linux atop a normal flagship android phone using linux deploy etc.

Or some odroid etc board...

You will see that it is already running very very powerful and does run well.

With Gentoo like approach it even more can squeeze performance and effiency.

And the party ends here... just ARM is not enough.

The problem again is a closed source block. Problems like the GPU and media engine and depending on manufacturer buggy binaries. So the dream ends here.

 

[silentbogo]

It's pretty unfair and meaningless to compare them like that though, unless for very specific tasks. And Windows without all the available Windows software is pretty pointless tbh. WinRT was not bad, just early. In a few years they might try again, but then x86 is so effecient so there's still little point to it. Android otoh is made for it and has a vast ecosystem. They just need better support for large monitors.

Multiplatform development is not something new or exotic - it does not even take much extra labor to get it done nowadays.
Most of the software runs in virtualized environment anyway, so the actual software developer for any modern platform should not even care what hardware he is coding for, as long as it supports the runtime environment he/she is using(e.g. Java, ART/dalvik, .NET whatever). All it takes is for MS to continue the development of Windows ecosystem for ARM platform(s) and, as evil as it sounds, keep promoting Windows Platform as the best way of developing software for Windows.

Take Linux/BSD, for example. I can easily run pretty much the same OS (Debian, Fedora, FreeBSD, Android etc.) on a wide variety of devices starting from Raspberry Pi or a $35 Android TV box, to Intel Atom - powered tablet/netbook/nettop, all the way to my hexacore Xeon workstation. There are some things that are going to be different, there is lots of software that is not available for either platform, but overall my basic tasks can be performed similarly on either one of them:
- All can run the same OS with the same desktop environment
- All are capable of running Chrome, Firefox, LibreOffice, Thunderbird, Dropbox, PDF viewer and other productivity software
- All of them include capabilities of HD video playback, and in most cases do it very well with hardware acceleration (except older Allwinner chips when not running Android)

So, the only argument that is left - is games, but even in this aspect of computing we will soon catch up with the second wave of game streaming services. And don't forget Angry Birds!

 

[alucasa]

ARM based servers exist I believe and some unix has ARM kernel.
The problem is software support or rather lack of support.

 

[Shamalamadingdong]

ARM-based netbooks and nettops are already on the market, so why not?
The only problem is Microsoft's abandonment of Windows for ARM platform, but otherwise I wouldn't mind having a mini-PC powered by a modern 64-bit ARM CPU.
There are tons of high-performance parts available already: just look at Tegra X1 or Exynos 8890, which in most cases outperform both Intel and AMD SoC (even some U-versions of mobile Core i3/i5 CPUs!!!).

There are no ARM SoCs that outperform any Intel U processors (the TDP alone will make sure of that). Which others are you referring to? Core M? Not those either. Atom? They're on the way to be discontinued. What AMD SoCs? Bristol Ridge? Probably not but let's say maybe considering the relatively low IPC. Also, while recently launched, it's pretty much a stopgap until Zen is launched which will also be faster.

 

[silentbogo]

There are no ARM SoCs that outperform any Intel U processors (the TDP alone will make sure of that). Which others are you referring to? Core M? Not those either. Atom? They're on the way to be discontinued. What AMD SoCs? Bristol Ridge? Probably not but let's say maybe considering the relatively low IPC. Also, while recently launched, it's pretty much a stopgap until Zen is launched which will also be faster.

http://www.phoronix.com/scan.php?page=article&item=nvidia-tegra-x1&num=1

Snapdragon 820 is on par with Tegra X1. If we take into account potential benefits from OpenCL-accelerated applications, then there is even more advantages.
At the current stage both CPUs are capable of outperforming:
- majority of Bay Trail CPUs, which are widely used in netbooks, nettops and low-end office workstations.
- pretty much all Cherry Trail CPUs
- most AMD E-series APUs
- Low-power AM1 CPUs
Next year we get Helio X30 and Snapdragon 830, which are rumored to be almost twice as fast, and support 4K/2K @60Hz.

I am talking about low-power x86_64 chips, which do have a very significant market share on PC market. So, no need to throw into pile Zen , Skylake, or Skull Canyon, since there are no 35W+ equivalent ARM CPUs on the consumer market yet.

 

[evernessince]

http://www.phoronix.com/scan.php?page=article&item=nvidia-tegra-x1&num=1

Snapdragon 820 is on par with Tegra X1. If we take into account potential benefits from OpenCL-accelerated applications, then there is even more advantages.
At the current stage both CPUs are capable of outperforming:
- majority of Bay Trail CPUs, which are widely used in netbooks, nettops and low-end office workstations.
- pretty much all Cherry Trail CPUs
- most AMD E-series APUs
- Low-power AM1 CPUs
Next year we get Helio X30 and Snapdragon 830, which are rumored to be almost twice as fast, and support 4K/2K @60Hz.

I am talking about low-power x86_64 chips, which do have a very significant market share on PC market. So, no need to throw into pile Zen , Skylake, or Skull Canyon, since there are no 35W+ equivalent ARM CPUs on the consumer market yet.

There's a major problem with your comparison, you are comparing x86 processors to ARM. If you consider the fact that x86 architecture is designed from the ground up to be as fast performing as possible while ARM is designed to save power it's going to be obvious which is going to do better. ARM only supports simple instructions so even if we assume that they could someday get IPC onto Intel's level (which would take allot of work given ARM is rather lazy when picking up threads) it would still have severe hardware limitations. Another thing to consider, the operating systems also plays a part. Windows is significantly more advanced than android and iOS. It can handle a far wider range of devices, from phones to desktop PCs, and has far more functionality. Just look at the high level APIs available on them and you'll see. Android and iOS don't even share the same Vulkan API and instead they have to deal with a stripped down version of the one running on PCs.

 

[R-T-B]

There's a major problem with your comparison, you are comparing x86 processors to ARM. If you consider the fact that x86 architecture is designed from the ground up to be as fast performing as possible while ARM is designed to save power it's going to be obvious which is going to do better.

I know you think it's obvious, but it's not obvious the way you think it is. Given this statement, ARM would do better for a given TDP.

ARM only supports simple instructions so even if we assume that they could someday get IPC onto Intel's level (which would take allot of work given ARM is rather lazy when picking up threads) it would still have severe hardware limitations.

Not really. There is an ARM-extension for pretty much everything there is an x86 extension for. They just haven't jumbled them into one die-soup yet.

Windows is significantly more advanced than android and iOS.

Not really.

It can handle a far wider range of devices

No. The linux kernel, upon which Android is based, is FAR MORE flexible than Windows.

Just look at the high level APIs available on them and you'll see. Android and iOS don't even share the same Vulkan API and instead they have to deal with a stripped down version of the one running on PCs.

No. A lot of the chips in phones can run full DX and Vulkan, they just choose not to as to not rape the handhelds battery. Pretty much the only thing missing from OpenGL ES vs OpenGL is GLUT, which is an unneeded battery hog anyway.

 

[Frick]

Multiplatform development is not something new or exotic - it does not even take much extra labor to get it done nowadays.
Most of the software runs in virtualized environment anyway, so the actual software developer for any modern platform should not even care what hardware he is coding for, as long as it supports the runtime environment he/she is using(e.g. Java, ART/dalvik, .NET whatever). All it takes is for MS to continue the development of Windows ecosystem for ARM platform(s) and, as evil as it sounds, keep promoting Windows Platform as the best way of developing software for Windows.

Take Linux/BSD, for example. I can easily run pretty much the same OS (Debian, Fedora, FreeBSD, Android etc.) on a wide variety of devices starting from Raspberry Pi or a $35 Android TV box, to Intel Atom - powered tablet/netbook/nettop, all the way to my hexacore Xeon workstation. There are some things that are going to be different, there is lots of software that is not available for either platform, but overall my basic tasks can be performed similarly on either one of them:
- All can run the same OS with the same desktop environment
- All are capable of running Chrome, Firefox, LibreOffice, Thunderbird, Dropbox, PDF viewer and other productivity software
- All of them include capabilities of HD video playback, and in most cases do it very well with hardware acceleration (except older Allwinner chips when not running Android)

So, the only argument that is left - is games, but even in this aspect of computing we will soon catch up with the second wave of game streaming services. And don't forget Angry Birds!

Aye true all of it, nowadays. UWP didn't exist when WinRT arrived, and Android was on 4.x iirc. Windows is less tied to decades of software now, so it's not far off.

 

[R-T-B]

Aye true all of it, nowadays. UWP didn't exist when WinRT arrived, and Android was on 4.x iirc. Windows is less tied to decades of software now, so it's not far off.

Heck, all we need is a good ol' x86 EMU envioronment and as sufficiently powerful arm core for the "old stuff" and I could see them pulling an "Apple Classic" mode on old PC software.

It's been done before. On PowerPC granted, but it has been done.

 

[Brusfantomet]

For IoT i would think a ATiny or something simple would be more fitting than a ARM core.

 

[Ferrum Master]

For IoT i would think a ATiny or something simple would be more fitting than a ARM core.

Well they are there, but attiny usage not being basically as the core device. ARM and linux kernel atop of it give more flexibility and access due to more easy software approach. Bare metal programming is bare metal... It will never take off for all, only for specific mission critical applications and needs. As a companion module, communicating via bus, yes always... but not the internet of things core. ARM just gives... as Jeremy Clarkson would like to say - more powerrr...

 

[Aquinus]

ARM only supports simple instructions so even if we assume that they could someday get IPC onto Intel's level (which would take allot of work given ARM is rather lazy when picking up threads) it would still have severe hardware limitations.

Jeez. NO! RISC does not mean that the instruction set is any more or less complicated than x86. The difference between most RISC and CISC CPUs is that RISCs tend to be LOAD-STORE architectures where every access to and from memory must be executed with one of those operations whereas in X86 (such as CISC,) you could have operations where it can run in a mode where one of its operands actually will automatically do a load to complete the operation or a store to put it back in system memory from the registers. You as the engineer (or compiler if we're talking about higher level languages than assembly,) the benefit of this is that ARM tends to have more registers so it's more likely that ARM can do a bulk load on data then perform operations on it unlike X86 whereas you don't even have to consider loading data, you can just use a memory address as an operand.

I personally think that ARM has the advantage here because it gives you control over how and when data is taken out of system memory, even more so if the data you need is stored sequentially in memory which would make for a very fast series of loads compared to purely random ones. This isn't to say that it's faster than most x86 CPUs but, it is to say that it's probably a more efficient CPU in terms of computational capability versus power consumption.

 

[silentbogo]

For IoT i would think a ATiny or something simple would be more fitting than a ARM core.

At the present day and time ATTiny is a thing of the past. Cortex M0/M0+ already costs as much as barely usable ATTiny2313. I did a few automation projects for my previous work, and I'm telling you from the personal experience that even a relatively simple task requires more memory than most ATTiny MCUs can provide. I had to write firmware in AVR Assembly, in order for it to fit into a tiny ROM space, and it was real pain in the ass. Especially when you do a highly-branched algorithm or a state machine, and hit the jump limitation, space limitation or hardware limitation few days into development.
Then we switched to Cortex M0 for the rest of our projects.
8-bit AVR architecture is pretty much dead. I still occasionally use ATTiny85 and ATTiny13 for very small stuff, but it is only because I have a cheap MCU supply from local manufacturer's overstock (and I can buy small bundles of 5-10 chips). Otherwise I would choose something like NXP LPC800-series or Freescale Kinetis L-series for these tasks (at almost the same price).

 

[Shamalamadingdong]

http://www.phoronix.com/scan.php?page=article&item=nvidia-tegra-x1&num=1

Snapdragon 820 is on par with Tegra X1. If we take into account potential benefits from OpenCL-accelerated applications, then there is even more advantages.
At the current stage both CPUs are capable of outperforming:
- majority of Bay Trail CPUs, which are widely used in netbooks, nettops and low-end office workstations.
- pretty much all Cherry Trail CPUs
- most AMD E-series APUs
- Low-power AM1 CPUs
Next year we get Helio X30 and Snapdragon 830, which are rumored to be almost twice as fast, and support 4K/2K @60Hz.

I am talking about low-power x86_64 chips, which do have a very significant market share on PC market. So, no need to throw into pile Zen , Skylake, or Skull Canyon, since there are no 35W+ equivalent ARM CPUs on the consumer market yet.

There are many problems here. You post a link to a benchmark on a Tegra X1 chip with a TDP of 15W that is getting curb-stomped by i3 processors with no turbo boost (aka severely crippled) in certain scenarios. It's often trading blows with an Atom CPU at a roughly 2W TDP. I mean c'mon: what kind of comparison is this?

The SD820 is a disappointing chip in general. I mean it's much better than SD810 but which 'modern' chip isn't? It will be interesting to see if Qualcomm will showcase any meaningful improvements with SD830 or if they'll be forced to throw in the towel and just go with A73 or its successor for the SD840.

Intel doomed their own Atom line by giving it the scraps of their resources. It was crippled from the start as to not compete with their Core offerings and by the time they really needed it to perform versus the rapidly growing ARM business it was too late and now it's slowly being phased out in favor of low TDP Core and Xeon products.

Skylake and Zen both scale down to 4W, so why is it not a fair comparison? You threw in a 15W ARM SoC vs a 2W x86 SoC and said "Look, it can beat it in some scenarios". It's a bottom of the barrel turd that Intel dug out versus the best of the best ARM-based SoCs have to offer.

If I recall correctly, ARM doesn't scale very well with high TDP although that's probably due to poor design. Will be interesting to see if AMD's K12 will bring ARM to servers in a meaningful way. The opposite could perhaps be said for x86 not scaling very well with low TDP but I do believe the newer Core M products do a decent job.

AMD's processors have generally been notoriously bad over the last six years, so I don't think the comparison is representative of anything.

With all that being said: if you want to highlight an ARM SoC why not go for Apple's A9 and the upcoming A10? Or Exynos 8890? Kirin 955 (awful GPU though)? These are better than the 820 in most cases. And would better support your claim. The Tegra X1 is also crippled by the process node and the A57 cores but is helped by the TDP to achieve and sustain a higher performance level than would be otherwise possible.
I'm throwing you a bone here.

Lastly: rumors are rumors and very unreliable. 4K/2K 60 Hz what?

As for the whole debate about RISC vs CISC. Modern chip design is coalescing. The differences aren't as clear as they used to be. x86 processors are leaner than they used to be and ARM processors are fatter - so to speak. Also, x86 processors translate code to RISC before executing so the discussion is moot.

 

[Prima.Vera]

Softbank = worst company in Japan

 

[prtskg]

Softbank = worst company in Japan

Reason?

 

[Brusfantomet]

At the present day and time ATTiny is a thing of the past. Cortex M0/M0+ already costs as much as barely usable ATTiny2313. I did a few automation projects for my previous work, and I'm telling you from the personal experience that even a relatively simple task requires more memory than most ATTiny MCUs can provide. I had to write firmware in AVR Assembly, in order for it to fit into a tiny ROM space, and it was real pain in the ass. Especially when you do a highly-branched algorithm or a state machine, and hit the jump limitation, space limitation or hardware limitation few days into development.
Then we switched to Cortex M0 for the rest of our projects.
8-bit AVR architecture is pretty much dead. I still occasionally use ATTiny85 and ATTiny13 for very small stuff, but it is only because I have a cheap MCU supply from local manufacturer's overstock (and I can buy small bundles of 5-10 chips). Otherwise I would choose something like NXP LPC800-series or Freescale Kinetis L-series for these tasks (at almost the same price).

Hmm, guess having a free supply of any AVR did screw my view of them, they were handing out development boards in Trondheim a few years ago.

 

[Prima.Vera]

Reason?

Most expensive mobile contracts, shitty offers (like if you want to pay more than 1GB of traffic you need to PAY A LOT more), charging extra on your contract for various reasons without giving any explanation, tech support and also general support is the worst of any Japanese company - they will never give you an exact explanation on what and why; always try to deceive and cheat the customers, ZERO English support and very racist towards non Japanese natives; charging ridiculous amounts of money for repairing a faulty device - replacing is a negotiation nightmare; keeping all their phones locked into their network; if you want it unlocked you need to pay a ridiculous amount of cash for that, and you cannot do it in the first 6 months anyway, even if the new Japanese law specifically prohibits locking the phone exclusively to 1 carrier; etc, etc, etc. I can go on like this all day.

 

[remixedcat]

were softbank the ones that bought out sprint...? sprint sucks.