Question Another potential big blow to Intel

[chrisjames61]

Microsoft going the Apple route and using their own ARM cpu's in Surface laptops and desktops and hardware for their data centers.

Intel falls on report Microsoft plans to design own chips for PCs and servers

Intel has famously had a long-running partnership with Microsoft as the primary processor maker for Windows PCs.

www.cnbc.com

 

[podspi]

I have to admit, I am surprised. I was expecting them to go with a semi-custom AMD chip - guess they've decided its time to get serious about ARM.

 

[VirtualLarry]

Microsoft... ARM... their own laptops and servers? Hmm. Guess that version of Windows for ARM is finally ready?

 

[NTMBK]

What will Microsoft do to differentiate their custom chip from Qualcomm? I just don't see it making sense.

 

[moinmoin]

I don't think Microsoft will sell the ARM chips, it's likely just pulling an Amazon in creating its own chips for use in it own products (Surface) and services (Azure). I wouldn't expect significant customization.

 

[MalVeauX]

Whatever pushes competition to increases superior releases, drive prices down and increase innovation.

Very best,

 

[lyonwonder]

I wonder if MS (unlike Apple) will also sell these ARM chips to OEMs like Dell, HP and Lenovo so they can expand support for their chips throughout the computer ecosystem?

 

[MalVeauX]

Realistically lots of things are moving to mobile platform more and more.

Very best,

 

[chrisjames61]

I wonder if MS (unlike Apple) will also sell these ARM chips to OEMs like Dell, HP and Lenovo so they can expand support for their chips throughout the computer ecosystem?

I would think mass adoption would be the plan.

 

[fluffmonster]

Microsoft will be able to purchase in a volume that custom-build can make sense for even small changes around the edges because semiconductor manufacturing capacity and specialization is at the point where customization for different clients is just a normal part the business. Expect to see more of this kind of thing from high-volume purchasers that are more focused around software than hardware (like Microsoft and Apple).

 

[DisEnchantment]

Finally I might be able to get that fanless Windows Laptop which some people on here were very dismissive about my concerns.
If it has no fan, produces less heat, has more than 4 cores, can run Linux (in VM at least, dual boot would be awesome) and most importantly not Apple, I am sold.
Running engineering SW does not have to burn the laps of engineers, people.
However I would rather see new Software first before I see new HW from MS. Although Windows is not my main OS, I use it a lot for development. It needs a lot of polishing, revitalized UI, orchestrated system libraries...

That being said, I am not sure engineering software will get ported any time soon.
This is no joke, in one extreme case one of our SW supplier refuse to port their tool chain (which generate calibration data for the DSPs we use) because the certification is too extensive. To overcome this we had to run their software in a Windows XP VM.

 

[chrisjames61]

However I would rather see new Software first before I see new HW from MS. Although Windows is not my main OS, I use it a lot for development. It needs a lot of polishing, revitalized UI, orchestrated system libraries...

That is an understatement. It hasn't in over 30 years so I would assume it will never happen. The same story since at least 1990.

 

[beginner99]

What will Microsoft do to differentiate their custom chip from Qualcomm? I just don't see it making sense.

Custom accelerators transparently supported by windows (only), just like Apple. Better core than the default ARM one? So I agree I'm not sure that forth the R&D cost.

 

[Doug S]

What will Microsoft do to differentiate their custom chip from Qualcomm? I just don't see it making sense.

Pretty sure they would use standard ARM cores like Qualcomm, but there is a ton of room for differentiation with everything else on the SoC. The ARM CPU cores are a small part of the overall SoC, and Microsoft would have full control over how all the rest of that area gets used. They will decide what hardware they consider important, and how powerful they need it to be, and set their own baseline for their products like Apple did with the M1 that sets a baseline expectation of a pretty powerful NPU for example. We have yet to see what developers can do if they have something like that at their disposal and can assume it is available in all models.

Also important, Microsoft won't be beholden to Qualcomm. That should not be underestimated. Even Apple has had major battles with them, Microsoft's size will not help them if Qualcomm wanted to squeeze them by raising prices once Microsoft becomes dependent on them or try to push them in a particular direction like making cellular mandatory on all Surface devices.

 

[Kenmitch]

I have to admit, I am surprised. I was expecting them to go with a semi-custom AMD chip - guess they've decided its time to get serious about ARM.

AMD does have a custom ARM offering on the back burner. Maybe a joint venture between AMD/Microsoft and Samsung as the fab? Samsung does have access to RDNA already....Guess time will tell in the end.

 

[Mopetar]

I'll admit that some kind of Frankenstein big.LITTLE design using x86 and ARM cores together would certainly be interesting, but I don't know how much practical advantage it has outside of a few niche market segments.

 

[amrnuke]

AMD does have a custom ARM offering on the back burner. Maybe a joint venture between AMD/Microsoft and Samsung as the fab? Samsung does have access to RDNA already....Guess time will tell in the end.

AMD even has an architectural license from Arm. Would be a smart thing to keep a finger in each pot. The Xilinx acquisition may fit into this a bit... I think a lot of people focused on the fact that Xilinx have a ton of experience in interposers, 2.5D/3D design, HPC, etc. But I think it's also notable that Xilinx already use Arm SoCs on their ACAPs and have experience integrating a wide variety of ISAs from PowerPC to Arm to x86 in their solutions.

 

[DrMrLordX]

What will Microsoft do to differentiate their custom chip from Qualcomm? I just don't see it making sense.

Probably nothing. I would expect MS to ape Amazon at first by designing their own in-house cloud server CPUs for Azure. Anything consumer will be handled by Qualcomm for the foreseeable future. Or I guess AMD but I am skeptical of that.

 

[gdansk]

I was hoping it was a continuation of their earlier EDGE/E2 effort but it seems like a new initiative to make more conventional designs. Interested to see if it will have unique features or be a bog standard ARM design. In the latter case, perhaps Qualcomm was unwilling to build what they wanted or maybe it was simply about cost.

 

[moinmoin]

AMD even has an architectural license from Arm. Would be a smart thing to keep a finger in each pot. The Xilinx acquisition may fit into this a bit... I think a lot of people focused on the fact that Xilinx have a ton of experience in interposers, 2.5D/3D design, HPC, etc. But I think it's also notable that Xilinx already use Arm SoCs on their ACAPs and have experience integrating a wide variety of ISAs from PowerPC to Arm to x86 in their solutions.

Indeed AMD has, and indeed ARM is potentially becoming more relevant again with the merger with Xilinx. Xlinix just joined the Confidental Computing Consortium, in its introduction to it it writes (my bolding):

"Confidential Computing can be achieved by assembling a TEE entirely in hardware. The three major CPU platform vendors: Intel, AMD, and ARM, all support a TEE. Intel has produced Software Guard Extensions (SGX), AMD’s offered up Secure Encrypted Virtualization (SEV), and ARM has TrustZone. Developers can leverage these TEE platforms, however, each is different, meaning code written for SGX will not work on an AMD processor. So, where does Xilinx fit in? Our objective is to understand how we can extend a TEE into an accelerator card or provide a method to securely hand off data and code between a host TEE and one executing within the accelerator card.

At this point, our Data Center Group (DCG) is exploring two paths. First, through our strong alliance with AMD, and we are exploring SEV to better understand how it might map to DCG’s future accelerator product plans. The second path involved our licensing of ARM core designs, which are included in many of our chips to handle control plane tasks. ARM has several other research projects underway that they’ve proposed to CCC that further extend TrustZone in ways that might make it much easier for us to secure an accelerator card’s execution environment. We’ve already begun discussions with the ARM team and hope to learn more over the coming months as we start to formulate our security plans for the future."

Xilinx Customer Community

forums.xilinx.com

And just to note again, in all Zen chips AMD uses ARM for the control plane tasks as well, including offering ARM's TrustZone.

 

[podspi]

I'll admit that some kind of Frankenstein big.LITTLE design using x86 and ARM cores together would certainly be interesting, but I don't know how much practical advantage it has outside of a few niche market segments.

I wonder if it would be possible to treat x86 (or ARM) as something to be accelerated. So you have a custom SoC with 6 powerful ARM cores, and 2 x86 cores off to the side like a GPU. Could we then have our ARM and x86 too?

 

[DrMrLordX]

I wonder if it would be possible to treat x86 (or ARM) as something to be accelerated. So you have a custom SoC with 6 powerful ARM cores, and 2 x86 cores off to the side like a GPU. Could we then have our ARM and x86 too?

Altera is already using CCIX to connect sockets on a 2P platform. Only sticky wicket would be getting an OS to handle two different uarches at once without using a VM/container approach.

 

[amrnuke]

Altera is already using CCIX to connect sockets on a 2P platform. Only sticky wicket would be getting an OS to handle two different uarches at once without using a VM/container approach.

Apple seem to have taken a step towards that with the M1.

 

[DrMrLordX]

Apple seem to have taken a step towards that with the M1.

Not really? There's only one uarch in the M1 - ARMv8 - and the OS has binaries that have been translated to ARMv8 instructions. A "true" hybrid x86/ARM machine would require an OS that could handle binaries compiled for different uarches natively. It might require multiple kernels.

 

[DisEnchantment]

Indeed AMD has, and indeed ARM is potentially becoming more relevant again with the merger with Xilinx. Xlinix just joined the Confidental Computing Consortium, in its introduction to it it writes (my bolding):

"Confidential Computing can be achieved by assembling a TEE entirely in hardware. The three major CPU platform vendors: Intel, AMD, and ARM, all support a TEE. Intel has produced Software Guard Extensions (SGX), AMD’s offered up Secure Encrypted Virtualization (SEV), and ARM has TrustZone. Developers can leverage these TEE platforms, however, each is different, meaning code written for SGX will not work on an AMD processor. So, where does Xilinx fit in? Our objective is to understand how we can extend a TEE into an accelerator card or provide a method to securely hand off data and code between a host TEE and one executing within the accelerator card.

At this point, our Data Center Group (DCG) is exploring two paths. First, through our strong alliance with AMD, and we are exploring SEV to better understand how it might map to DCG’s future accelerator product plans. The second path involved our licensing of ARM core designs, which are included in many of our chips to handle control plane tasks. ARM has several other research projects underway that they’ve proposed to CCC that further extend TrustZone in ways that might make it much easier for us to secure an accelerator card’s execution environment. We’ve already begun discussions with the ARM team and hope to learn more over the coming months as we start to formulate our security plans for the future."

Xilinx Customer Community

forums.xilinx.com

And just to note again, in all Zen chips AMD uses ARM for the control plane tasks as well, including offering ARM's TrustZone.

While not inaccurate in that article, TEE and SEV are totally different things. You will find TEE in all modern ARM SoCs used in mobile, automotive etc.
TEE allows a secure enclave which an application can access to do certain things without accessing the underlying secrets. A very common usage of TEE for example is storing private keys, API keys etc. For example, an application request a chunk of data to be signed using the private keys from the enclave but have no access to the private keys in question, but only gets the signed data. Therfore, very common use of TEE is in DRM, payment methods etc. It protect secrets on the host not the app. If you are developing consumer devices in some way or the other you would have likely needed to use TEE in some way.
A comparable implementation of TEE for Windows desktop would be the upcoming Microsoft Pluton IP.

SEV on the other hand protects VMs (and also processes) from one another by implementing control of context registers, process memory pages etc from within the process. SEV-SNP extends this to IOMMU and others.
In SEV-SME, the memory is encrypted using keys owned by the process, the host cannot access it. WITH SEV-ES host cannot access the paged memory, or influence DMA, or read registers owned by the process. A HW exception is triggered when such things happen and the guest will have to handle these exceptions which they can either terminate or log or send info somewhere else.
There is no equivalent to SEV as far I know. SGX and FME attempts to do part of it, but SEV is far more comprehensive. SEV-SNP is taking this even further.
SEV is unique tech, and for those who are in the server world, who are running their apps in public cloud knows how important this is.

AMD's PSP can aid the implementation of TEE, but by itself does much more than just that. It checks microcode signature, firmware signature, authenticate nodes attached to a system fabric etc. PSP is not involved in SEV in anyway. The SEV logic is built in the core itself.