Microsoft takes an Arm architecture license

[tangey]

Microsoft announced today that they have taken an Architecture license from ARM, only 4 other companies have done this, Qualcomm, Marvell, Infineon and an un-named one that is highly likely to be Apple.

It raises some interesting questions.

I wonder what this is likely to mean for the MSFT/Tegra tie up in the long term ?
Also, does it signal MSFT considering porting full windows to the Arm processor ?

Or maybe its got nothing to do with the handheld field and its microsoft looking at its data centres and doing a customised Arm server ?
http://dcsblog.burtongroup.com/data_center_strategies/2010/04/microsoft-in-data-center-arm-race.html

 

[fehu]

courier return?
xboy advance edition series?
win embed 7 well done edition?

 

[Exophase]

Sounds a little bizarre. Since when has MS been in the business of designing CPUs? Feels like there's a few intermediate steps they should be taking before that point. MS is still by and large a software company, with a few big hardware products but still consisting of IP from other vendors.

What all those other licensees have in common is that they're using their custom ARM cores in the SoCs they're selling to pretty focused markets (Qualcomm probably making the biggest killing at it)... Infineon doesn't seem to have a custom ARM out yet, not surprising since the license isn't very old, but they're also targeting a particular group with their products.

It's hard to imagine Microsoft selling SoCs, so they're more likely in the market of developing a CPU all for themselves.. which I doubt will give them much of a substantial competitive advantage. I could say the same for Apple, but at least they have more of a known presence of CPU designers in their ranks.

I wonder if an explanation for these licenses is that the companies expect to sell their ARM devices in such large volume that the cost of developing the CPU is offset by the lower royalties paid per CPU (vs using a stock ARM core). I hear the licensing overhead is pretty low for ARM, so I don't know.

 

[rpg.314]

Whoa.

This could be big. It certainly means they have decided it's worth throwing a lot of resources behind this to make it work. If it takes off, this could shake up the CPU market radically.

 

[MfA]

Personally I think they are just trying to copy Apple for their handset plans.

If they want to really get in the CPU business they'd have bought into AMD&GloFo ... fabless is fine for fast moving markets like GPUs, but competing with the IDMs on CPUs doesn't work.

 

[darkblu]

Did not see that coming. Interesting move, ms.

Personally I think they are just trying to copy Apple for their handset plans.

If they want to really get in the CPU business they'd have bought into AMD&GloFo ... fabless is fine for fast moving markets like GPUs, but competing with the IDMs on CPUs doesn't work.

Well, Apple are fabless.

 

[Lazy8s]

Microsoft really wouldn't have much use for PowerPC anymore.

 

[MfA]

Well, Apple are fabless.

I meant desktop/console CPUs.

 

[rpg.314]

Personally I think they are just trying to copy Apple for their handset plans.

If they want to really get in the CPU business they'd have bought into AMD&GloFo ... fabless is fine for fast moving markets like GPUs, but competing with the IDMs on CPUs doesn't work.

I think they are more interested in trying out ARM for servers. Although having a license means competencies developed can later on be leveraged in the embedded sector as well.

 

[MfA]

Seems a lot of money to me for just that, Oak Trail will almost certainly close the power gap with ARM IMO. At that point can they really afford to switch away from x86?

 

[metafor]

Seems a lot of money to me for just that, Oak Trail will almost certainly close the power gap with ARM IMO. At that point can they really afford to switch away from x86?

Doubtful. Even Moorestown, for all its claims, is still nowhere near ARM in terms of power. Intel's slides coincidentally didn't show power numbers when the CPU is being used but its TDP is still around 2W compared to 500mW of a typical ARM chip.

 

[Exophase]

I don't think Oak Trail is going to match Cortex-A9 in CPU core perf/watt when CPU utilization is high, the kind of scenario that fits server usage. Intel has been optimizing idle consumption, peripheral integration, and fixed function hardware more than it has been bringing down core consumption.

This is also a completely moot point, because Cortex-A9 (or any other ARM Ltd CPU cores) has nothing to do with Microsoft's licensing. Microsoft wants to develop a new CPU entirely, and instead of comparing existing CPUs we should be contemplating why they want to do this. That they chose to use the ARM architecture is not much more than an implementation detail... and it's a fairly good choice (in a lot of ways, much better than doing a new x86 CPU, although I don't think they'd have to get a license for that)

 

[Squilliam]

I don't think it so, but does this have anything to do with their Xbox consoles? Would it be at all effective sticking a huge number of ARM cores into a home console?

 

[Mindfury]

I don't think it so, but does this have anything to do with their Xbox consoles? Would it be at all effective sticking a huge number of ARM cores into a home console?

I don't think it would have anything to do with Xbox,MS will stick with PPC to keep backward compatibility.

 

[brain_stew]

I don't think it would have anything to do with Xbox,MS will stick with PPC to keep backward compatibility.

Has any other division within Microsoft come anywhere near to "designing" their own CPU/hardware as the Xbox division? With the 360 they moved far away from the "off the shelf" approach of the original Xbox so that they could get costs down with regular shrinks and have a more direct influence on the cost of their box. Designing their own CPU entirely would be a natural progression from this and there's definitely been murmurings about such a move in the past.

What other hardware device can Microsoft feel confident in selling ~50 million units? Surely any project that justifies its own CPU design has to have projected sales in that sort of region? Would emulating/virtualising 360's simple IO PowerPC cores really be that difficult if any such ARM design had that as one of its initial project goals? I know we don't see ARM chips released with 3ghz clocks but in a home platform on an even smaller process perhaps that's not unrealistic, isn't the CortexA9 said to be capable of reaching 2ghz already?

I don't know, I certainly never expected them to move away from PowerPC but I'm just struggling to think of any other Micrsoft hardware that could justify the development of a new CPU. Interesting and surprising news what ever comes of it though, that's for sure.

 

[rpg.314]

Has any other division within Microsoft come anywhere near to "designing" their own CPU/hardware as the Xbox division? With the 360 they moved far away from the "off the shelf" approach of the original Xbox so that they could get costs down with regular shrinks and have a more direct influence on the cost of their box. Designing their own CPU entirely would be a natural progression from this and there's definitely been murmurings about such a move in the past.

What other hardware device can Microsoft feel confident in selling ~50 million units? Surely any project that justifies its own CPU design has to have projected sales in that sort of region? Would emulating/virtualising 360's simple IO PowerPC cores really be that difficult if any such ARM design had that as one of its initial project goals? I know we don't see ARM chips released with 3ghz clocks but in a home platform on an even smaller process perhaps that's not unrealistic, isn't the CortexA9 said to be capable of reaching 2ghz already?

I don't know, I certainly never expected them to move away from PowerPC but I'm just struggling to think of any other Micrsoft hardware that could justify the development of a new CPU. Interesting and surprising news what ever comes of it though, that's for sure.

What do they have to gain by designing their own ARM uArch that is worth breaking BC, and taking on massive costs and engineering efforts, while the alternative is to write a cheque to IBM?

AFAICS, it is for server market. There are enough people doing ARM for mobile that it is simply a matter of writing a cheque for MS to get whatever mobile core it wants.

 

[Squilliam]

Has any other division within Microsoft come anywhere near to "designing" their own CPU/hardware as the Xbox division? With the 360 they moved far away from the "off the shelf" approach of the original Xbox so that they could get costs down with regular shrinks and have a more direct influence on the cost of their box. Designing their own CPU entirely would be a natural progression from this and there's definitely been murmurings about such a move in the past.

Well I suspect it may be worthwhile if they can amortize the cost over 3 products to make their own architecture. Why not turn Microsoft surface into an embedded ARM platform and use hardware designed for the next Xbox to run it? Then in addition to that bring their modified ARM into their server platform offerings and become an integrated software/hardware company there as well by providing both at once. Im rambling and I know nothing of ARM or this business...

So how does one Arm core compare to say one PPC core in Xenos? What kind of architecture could they make by combining ARM with other technologies? Could it be something like 3PPC + 10 Arm + ALU array? With the ARM cores acting as controls over a single SIMD array.

 

[Exophase]

Well I suspect it may be worthwhile if they can amortize the cost over 3 products to make their own architecture. Why not turn Microsoft surface into an embedded ARM platform and use hardware designed for the next Xbox to run it? Then in addition to that bring their modified ARM into their server platform offerings and become an integrated software/hardware company there as well by providing both at once. Im rambling and I know nothing of ARM or this business...

So how does one Arm core compare to say one PPC core in Xenos? What kind of architecture could they make by combining ARM with other technologies? Could it be something like 3PPC + 10 Arm + ALU array? With the ARM cores acting as controls over a single SIMD array.

We're talking about CPU implementations, not architecture. I don't think MS will design an ARM CPU that ends up being ideal for handhelds, gaming consoles, and servers, although I could be wrong and it might just be a matter of scaling to the right number of cores. This is more likely to hold for servers than gaming consoles, though, where single core performance is probably still important, especially if they want something more powerful than XBox 360.

ARM vs PPC is really a pretty mild argument, it depends much more on what kind of CPUs Microsoft wants to make. Combining PPC and ARM in any way where the user is expected to program all of them would be a little crazy, although I suppose for the sake of BC they might (you can't really say hardware BC is that important for MS in the first place though)

 

[aaronspink]

Microsoft announced today that they have taken an Architecture license from ARM, only 4 other companies have done this, Qualcomm, Marvell, Infineon and an un-named one that is highly likely to be Apple.

You missed two: DEC and Intel.
And I'm fairly sure that Samsung has one as well.

 

[aaronspink]

Doubtful. Even Moorestown, for all its claims, is still nowhere near ARM in terms of power. Intel's slides coincidentally didn't show power numbers when the CPU is being used but its TDP is still around 2W compared to 500mW of a typical ARM chip.

That partially has to due with a difference in strategy. do >2x the work in <1/2 the time and go back into sleep vs burn less power for longer periods of time. So far all data points to hurry up and sleep as having overall better power efficiency.