Should have written SoC, because it combines the major part of current Mediatek IP blocks with the GPU parts being replaced by Blackwall.
What Nvidia is trying to do here is a laptop CPU (with a huge NPU and graphics, of course, almost a SOC if you will.) to run Windows in a corporate environment. There is absolutely ZERO chance it works without Windows, it would require that organization to completely replace their software stack at huge expense. It's a very risky venture IMO, high risk, high reward, (and Snapdragon has at least a 1 year lead.)
I guess the confusion comes from project DIGITS or what's now called DGX Spark. That seems to be the SoC Mediatek and Nvidia seemingly want to grow into a full family of SoCs to address everything from laptop to the workstation size of a DGX.
That is evidently coming out this month with Linux, while the Windows variant has been pushed backward. And while that SoC has battled hardware issues according to SemiAccurate, it's now the software issues on Windows that hold things back there.
And while your focus may be on corporate IT, that's only a small part of what Mediatek/Nvidia may want to address.
Again, apple is out of the picture. This is Nvidia hardware. One supposes, later someone may worry about getting MacOS running on third-party hardware, but why? I guess we did have Hackintoshes for a while.
Apple has pushed the industry hard with energy efficiency and their 1/2/4 scaled unified SoC design. That's why we see things like Snapdragon Orion, Lunar Lake, Strix Halo but also Nvidia DIGITS.
I obviously don't expect the Fruity Cult to port to DIGITS, I was mostly commenting on theoretical portability.
Microsoft has been working on Windows ARM for a decade now. I'm guessing this chip is requiring them to do something different. It's not so much Windows anymore in most cases.
Porting Windows to ARM is a lot of different things. Porting the OS or kernel code as such may be the smallest part, especially since Windows originally used to be multi architecture. The application part is another issue, but the biggest problem seems to be driver support.
On x86 graphics drivers are delivered by vendors in a form and process that's well established and doesn't requrire a lot of effort from the device OEMs or where at least support procedures and staff are well established.
In the case of the Snapdragon, neither Qualcomm nor Microsoft were said to be any help, resulting not only in very fragile initial graphics drivers, but also in practically zero maintenance since. Laptop OEMs aren't in the habit of maintaining OS drivers, at best they deliver BIOS updates.
And then there is the issue of all those other small devices and sensors, for which Snapdragon has IP blocks built in, but offers no integration support to OEMs, who are used to having their own stuff to differentiate.
Again for x86 processes and staff have been well established for decades, for Snapdragon (and other ARMs) it's a brick wall.
I take my Orange Pi 5+ and my Raspberry PI5 mostly as comparison: they face similar issues on Linux in that it's not just about compiling the kernel, but the drivers needed. You can run Windows for ARM on a VM there much more easily than on the physical hardware, because the effort for developing native Windows drivers for each bespoke SoC is insane.
Applications compiled for Windows Arm are the real issue. Otherwise, like with all emulation, there are prices to pay. If you go to Amazon and look at ANY Snapdragon laptop and you'll find the dreaded, "Frequently Returned Item" label. That tells me that there are enough problems that it's not ready for consumers, let alone ultra-conservative corporate buyers.
Users can't tell if applications crash because applications fail to be emulated properly or if drivers fail with applications. Just look at how failing games were originally attributed to the games, then the GPU drivers when actually it was Raptor Lake that was defective.
If somebody paid me for that, I'd run the crashing applications on an RDP session (or in a VM) just to see if it's not actually the graphics drivers that cause the issues.
Emulation is obviously still an issue, especially with all those ISA extensions being crucial to get top performance, but it's not the whole picture. And then again, I don't see corporate buyers as being the biggest beef in the laptop market or the most interested in ARM designs.
I don't think you understand. Microsoft has been begging manufacturers to make these things for a LONG time now. Microsoft is all-in. They really don't have to be pressured. More likely they did the work for Snapdragon and now are having to add some compatibility stuff if it's installed on Nvidia's processor. But no, Microsoft has wanted Windows on ARM longer than anyone. No ARM-twisting required!😜
Microsoft wants a bigger part of the PC pie. ARM was a tool to fend off the Fruity Cult, Intel and whoever wants any $ M$ believes belongs to them.
But being $ oriented, Microsoft didn't want to pay for that themselves, so they talked Qualcomm into doing most of the work for them. And then left them hanging, when they couldn't do it alone. Microsoft will happily go back to x86 or with Nvidia, if it means hardware costs less and more of the $ go to them.
Been watching them for over 4 decades.
45 years here.
They had some niche things in corporate environments. Back in the late 80's and 90's pretty much every print magazine used Apple. Hollywood for rendering animation, but not much else. Apple's problem is they have always wanted to own the whole stack, and priced themselves out of the business market early. I recall a business ad they ran in the mid-80's for the Apple Lisa. The machine in the ad cost $17000 in the day the IBM PC cost $2000. No chance a corporate buyer can justify that.
Nope, when I ran my Apple ][ it was still open and the IBM PC the Apple ][ successor in spirit. It's with the Lisa and the Mac that the wrong Steve got his way.
The problem with Windows 11 was security. It really was. The problem is that security came with a performance penalty on any machine that didn't have the security processor.
You're obviously corporate IT
No, the security processor doesn't do any acceleration and in fact any machine running Windows 11 without a TPU is likely to run it faster.
The security processor basically grew out of a service processor much like you had a Z80 on IBM mainframes to diagnose, initialize, load the µ-code, and bootstrap the hardware before IPL.
Likewise x86 has featured a service processor, 80486 based and running Minix on Intel and ARM based on AMD to do things like hardware initialization, µ-code loading etc.
This then came to include PKI cryptography and certificate management to protect that µ-code and UEFI firmware against tampering but with all that it still remains a relatively weak piece of hardware. The brunt of the security work is always done with symmetric cryptography and the service processor or TPU is just responsible for managing those symmetric keys using assymetric cryptography for key management pretty much like an HSM.
And that brunt, including transparent RAM encryption is handled by the CPU or its memory controller, not by the TPU.
And while that may cost perhaps 5% overhead on memory access, a TPU can't do anything to reduce that. I only enables it because it can manage the key, nobody wants to type in a memory encryption key every time a system boots, while storing it anywhere where the CPU can see it, means it's easily compromised.
And Bitlocker for storage encryption is pretty similar to memory encryption.
So turning a TPU off (or not having one) actually has your sytem run faster, because it disables memory encryption (and Bitlocker), because without HSM like key management, both aren't very secure..
But most of the overhead from security comes from things like HVCI and that runs Windows in a VM, with plenty of ring transition overhead and hypervisor calls. It exists on both Windows 10 and 11, the only difference is the default on normal desktop editions: Windows 11 normally turns it on, quite independently of a TPU being present, because that doesn't play a role, except perhaps for checking driver signatures. But server and IoT LSTC editions allow you to turn it off, which has Windows 11 run just as fast as Windows 10, which at least initially didn't turn it on by default.
This put Microsoft in a position where they could release a new OS that demonstrably performed worse than the previous one on some machines. It would have been a PR disaster. So they set the hardware requirements, then allowed those with machines that didn't have the hardware work around it. Can't blame MS if you used a work-around can you?
Sorry, but as explained above you got it all wrong, and there is nothing new in Windows 11 than the defaults. The functionality existed in Windows 10 and only the presets differ.
I'm guessing Nvidia sees and age with less employees and more AI. Not sure if they're right either, as I said, it's a risky thing.
Interestingly here, x86 doesn't seem to be quite dead. Intel's very first chip after the Snapdragon hoopla was more performant, had 90% of the battery life, and none of the compatibility issues. So I don't think we'll see the end of x86 in the corporate world anytime soon, it's easier for them to stick with X86 than switch to ARM. Snapdragon and Nvidia don't have to equal x86, they need to far surpass it to get those stodgy folks interested.
CPU efficiency isn't just about the ISA.
Snapdragon Elites did worse than the Fruity Cult's Mx chips in that regard, which helped reduce the gap already. And then Lunar Lake might draw equal in terms of battery life, but only at half the multithreaded CPU power.
And a US (or even Europe) based corporate IT view no longer reflects where revenue is made in the future.
Most growth (instead of just replacement) is actually in mobile, tablets and perhaps now portable consoles and mostly outside the old West, in Asia, Africa and LatAm.
And there governments dictate against continued US dominance in their own IT, and campaign heavily even in corporate markets against the risks of US hardware and software being weaponized against them, while consumers just want to have fun with games and content.
The US have dominated global IT for 8 decades, but growth potential dictates who leads. That potential is much smaller in the US saturated market than elsewhere and the tide is turning. All the faster the more US actors try to push against it.
Facebook
Twitter
Instagram