ambidextrous computing; AMD project skyla..skybridge!

[monstercameron]

AMD will launch project Skybridge, a pair of pin-compatible ARM and x86 based SoCs. Leveraging next generation Puma+ x86 cores or ARM's Cortex A57 cores, these SoCs form the foundation of the next phase in AMD's evolution where ARM and x86 are treated like equal class citizens. As I mentioned in today's post however, both of these designs really aim at the lower end of the performance segment. To address a higher performance market, AMD is doing what many ARM partners have done and is leveraging an ARM architecture license to design their own microarchitecture.

source: http://www.anandtech.com/show/7990/amd-announces-k12-core-custom-64bit-arm-design-in-2016

Next year, AMD will release a low-power 20nm Cortex A57 based SoC with integrated Graphics Core Next GPU. The big news? The 20nm ARM based SoC will be pin compatible with AMD's next-generation low power x86 SoC (using Puma+ cores). The ARM SoC will also be AMD's first official Android platform.

source: http://www.anandtech.com/show/7989/...bridge-pincompatible-arm-and-x86-socs-in-2015

amd arm core "k12"

 

[witeken]

ARM, Samsung, Qualcomm, Apple, Nvidia, now AMD,... it seems that everyone's designing their own ARM ISA cores.

Not sure why you'd want that, though. One good x86 core should be enough for everything from Android to Windows. I don't know why AMD doesn't simply use its Puma core.

 

[NTMBK]

ARM, Samsung, Qualcomm, Apple, Nvidia, now AMD,... it seems that everyone's designing their own ARM ISA cores.

Not sure why you'd want that, though. One good x86 core should be enough for everything from Android to Windows. I don't know why AMD doesn't simply use its Puma core.

It does kind of make sense from the "semicustom" point of view- add ARM cores to the menu of IP that you can offer a customer, and let them pick and choose what they want integrated into their SoC.

 

[flash-gordon]

http://www.media-server.com/m/p/762ectcz

Live now...

 

[Phynaz]

Am I missing something?

I'm not seeing on the slides what the OP is implying in the thread title - a CPU that is a combination ARM/x86 CPU.

 

[Jodell88]

Am I missing something?

I'm not seeing on the slides what the OP is implying in the thread title - a CPU that is a combination ARM/x86 CPU.

http://www.anandtech.com/show/7990/amd-announces-k12-core-custom-64bit-arm-design-in-2016

 

[monstercameron]

Am I missing something?

I'm not seeing on the slides what the OP is implying in the thread title - a CPU that is a combination ARM/x86 CPU.

i couldnt get a screenshot fast enough...oops didnt mean integrated

 

[Arachnotronic]

ARM, Samsung, Qualcomm, Apple, Nvidia, now AMD,... it seems that everyone's designing their own ARM ISA cores.

Not sure why you'd want that, though. One good x86 core should be enough for everything from Android to Windows. I don't know why AMD doesn't simply use its Puma core.

For a company running a semi-custom business, a custom ARM core for higher performance applications makes a lot of sense.

 

[monstercameron]

they were asked why not just leave x86 for arm fulltime...they gave a PR-ish response but I believe they just like the abuse.

 

[WaitingForNehalem]

They're co-developing ARM and x86 IP cores for inclusion in semi-custom APUs. There will not be an x86 core with an ARM core side by side on the same die for compute purposes. AMD does have a Cortex A5 on some of their more recent CPUs for trusted computing but that is about it.

 

[Headfoot]

There must be some value to a x86-ARM dual core to some parties. What that value is, whether large or small, is the real debate. But AMD is very uniquely positioned to be the only company currently able to deliver a product which has both capabilities. So whatever the worth of that market, AMD is the only one that can currently tap it. So this is a great move for them, given that it gives them better entry into the burgeoning ARM market and taps a "new" market. The new management team seems to have much better focus.

 

[bullzz]

I too dont understand the need for pin compatible ARM and x86 core for client. I think AMD is doing it so OEMs can us the same mb for both windows and android tablets/phones. isnt it easier to port android to puma than develop a ARM soc and then port android for it?

 

[MisterMac]

Am I missing something?

I'm not seeing on the slides what the OP is implying in the thread title - a CPU that is a combination ARM/x86 CPU.

....pin-compatible meaning socket socket compatible or am i misreading it?


I don't see the advantage for except small volume custom design.


Can some EE heads like IDC answer - if it could save design time by creating a 1 socket fits all design strategy for SoCs?

 

[Vesku]

I too dont understand the need for pin compatible ARM and x86 core for client. I think AMD is doing it so OEMs can us the same mb for both windows and android tablets/phones. isnt it easier to port android to puma than develop a ARM soc and then port android for it?

Yes, cost savings for potential customers and some packaging cost savings for themselves. The real question would have been "why NOT make them pin compatible".

 

[Headfoot]

....pin-compatible meaning socket socket compatible or am i misreading it?


I don't see the advantage for except small volume custom design.


Can some EE heads like IDC answer - if it could save design time by creating a 1 socket fits all design strategy for SoCs?

I think it makes more sense to look at it from the customer perspective than the design perspective. Because AMD is now doing "semi-custom design" for any customer who wants it (like IBM used to/still kinda does), they are likely offering it as a convenience for their customers. At the very least is gives customers another option they can't get anywhere else.

A theoretical example would be a dual-boot tablet that can go from Android to Windows (these already exist, see http://www.asus.com/AllinOne_PCs/ASUS_Transformer_AiO_P1801/) but having it in a single die would cut costs to the OEM as they don't have to place two sockets, buy two processors, equip them with two sets of memory etc.

 

[Headfoot]

Yes, cost savings for potential customers and some packaging cost savings for themselves. The real question would have been "why NOT make them pin compatible".

Exactly this!

 

[sontin]

I'm the only who's shocked that they will need two years for their own ARM core? Looks like somebody pushed the panic button a few days ago...

Even nVidia is beating them in the ARM race and they are a GPU company...

 

[Phynaz]

I think it makes more sense to look at it from the customer perspective than the design perspective. Because AMD is now doing "semi-custom design" for any customer who wants it (like IBM used to/still kinda does), they are likely offering it as a convenience for their customers. At the very least is gives customers another option they can't get anywhere else.

A theoretical example would be a dual-boot tablet that can go from Android to Windows (these already exist, see http://www.asus.com/AllinOne_PCs/ASUS_Transformer_AiO_P1801/) but having it in a single die would cut costs to the OEM as they don't have to place two sockets, buy two processors, equip them with two sets of memory etc.

They aren't talking about an integrated ARM/x86 CPU. That would be far, far too costly.

 

[MisterMac]

They aren't talking about an integrated ARM/x86 CPU. That would be far, far too costly.

Ding.

I imagine it would also need an absurd amount of validation in the many years range - before decoders for both would be effective and not be IPC handicapped.

 

[ShintaiDK]

Ding.

I imagine it would also need an absurd amount of validation in the many years range - before decoders for both would be effective and not be IPC handicapped.

The memory mapping alone kills it.

 

[bullzz]

@vesku - yes I understand they want to save costs by having the same package. but my real ques is why not just port android to puma instead of developing a whole new soc for it

@sontin - that too. 2016 seems a bit far when qualcomm will have 64bit A57 soc by end of this year. yes it not for servers but I am not sure if it will abt 2 yrs

 

[monstercameron]

The memory mapping alone kills it.

you mean like huma?

 

[Vesku]

@vesku - yes I understand they want to save costs by having the same package. but my real ques is why not just port android to puma instead of developing a whole new soc for it

@sontin - that too. 2016 seems a bit far when qualcomm will have 64bit A57 soc by end of this year. yes it not for servers but I am not sure if it will abt 2 yrs

1. Mainly because it will cost millions in software development, where as they are already spending money to produce ARM cores which will support Android with very little tweaking.

2. That's for custom ARM, AMD has stock ARM 64 bit aimed for late this year: http://www.theregister.co.uk/2014/0...64bit_arm_server_chip_to_launch_in_late_2014/

Skybridge is just about having design compatible ARM and x86 SoCs, but it would be interesting if they did include one or two low power ARM cores in an x86 SoC so Linux/Android can run simultaneously with Windows. But then again can almost do the same thing with virtual machining if Android gets official x86 support.

 

[AtenRa]

One implementation of SkyBridge is seemicro servers. You get the same fabric, same socket, same platform etc and you only change the APU.
Skybridge is more about Embedded, semi-custom designs and server than consumer.

 

[Idontcare]

I'm the only who's shocked that they will need two years for their own ARM core? Looks like somebody pushed the panic button a few days ago...

Even nVidia is beating them in the ARM race and they are a GPU company...

The 2yr time frame was most likely selected on the basis of targeting the HVM timeline for a specific process node currently under development (10nm perhaps) combined with needing at least 2yrs head start on getting customers aligned and setup to start developing products around the chip so that everything comes together at the same time.

They aren't talking about an integrated ARM/x86 CPU. That would be far, far too costly.

Seems like if they can do it for their APU's with the GPU tech, they should also be able to do it with x86+ARM as well.

The specific graphics element that they've chosen for their powerpoint presentation depicting the "2016+" product looks to be chosen to communicate "co-processor product is coming".

Will there be a market for it though?