Jump To:

We test the RTX Spark Superchip in Microsoft's Surface Laptop Ultra — Nvidia's Windows on Arm platform impresses in local AI, but gaming isn't its strong suit

News
By
Published

Laying the foundation for the reinvention of the PC in the agentic AI era

Microsoft Surface Laptop Ultra
(Image credit: Tom's Hardware)
Jump To:
14

Nvidia CEO Jensen Huang announced the RTX Spark platform with the grand ambition to “reinvent the PC” back at Computex 2026. By pairing the GB10 Superchip with a version of Windows on Arm tailored for the secure and contained execution of local AI agents, Nvidia and its partners at Microsoft, Dell, HP, MSI, Lenovo, and Asus are all hoping to pave the way to a future where the PC is an active, autonomous companion for work and creation.

The hardware foundation for that vision of the future is now in place. Last week, we flew out to San Francisco to get all the details on the RTX Spark platform, and we returned home with Microsoft’s Surface Laptop Ultra in top-end trim to put it all to the test.

Latest Videos FromTom's Hardware
Jeffrey Kampman
Jeffrey Kampman
Senior Analyst, Graphics

As the Senior Analyst, Graphics at Tom's Hardware, Jeff Kampman covers everything that has to do with graphics cards, gaming performance, and more. From integrated graphics processors to discrete graphics cards to the hyperscale installations powering our AI future, if it's got a GPU in it, Jeff is on it. 

14 Comments Comment from the forums
  • cknobman
    Microsoft and Nvidia, 2 of the most greedy anti consumer tech companies out there.

    Hard pass.
    Reply
  • excalibur1814
    Apple and nVidia, 2 of the most greedy anti consumer tech companies out there.

    Pass hard.
    Reply
  • sseemaku
    Is superchip a brand name or everything nvidia makes is called superchip these days?
    Reply
  • bit_user
    sseemaku said:
    Is superchip a brand name or everything nvidia makes is called superchip these days?
    It bugged me when Nvidia tried to call their little SoCs "supercomputers", in the past. But, I'll grant that they have a legit argument, here. A small one, in this day and age, but I can see the point.

    It stems from the fact that RTX Spark is comprised of two (mostly) independent dies. It basically works out to there being a CPU and a GPU die, integrated into a single package. So, it's multiple chips which function as an integrated whole. I think the modern definition of a super computer is: multiple computers that can function as an integrated whole. So, there's an argument that's more than mere marketing fluff.

    The same language is extended to their server-based 2-chip modules, which pair CPU and GPU chips. So, you'll find them also using the term to describe their Grace-Hopper, Grace-Blackwell, or Vera-Rubin solutions, where Grace and Vera are the CPUs and Hopper, Blackwell, and Rubin are GPUs.

    That said, I'd rather they not call it a super chip. But, this is Nvidia and they like to hype everything up to the nth degree.
    Reply
  • Vicecrow
    I like the price tag. Cheap cheap.
    Time to find another hobby
    Reply
  • GenericUser2001
    bit_user said:
    It bugged me when Nvidia tried to call their little SoCs "supercomputers", in the past. But, I'll grant that they have a legit argument, here. A small one, in this day and age, but I can see the point.

    It stems from the fact that RTX Spark is comprised of two (mostly) independent dies. It basically works out to there being a CPU and a GPU die, integrated into a single package. So, it's multiple chips which function as an integrated whole. I think the modern definition of a super computer is: multiple computers that can function as an integrated whole. So, there's an argument that's more than mere marketing fluff.

    The same language is extended to their server-based 2-chip modules, which pair CPU and GPU chips. So, you'll find them also using the term to describe their Grace-Hopper, Grace-Blackwell, or Vera-Rubin solutions, where Grace and Vera are the CPUs and Hopper, Blackwell, and Rubin are GPUs.

    That said, I'd rather they not call it a super chip. But, this is Nvidia and they like to hype everything up to the nth degree.
    Having multiple dies that function as a single package is nothing special, and has been common for years now. AMD has been doing multiple dies for one package since Zen 2 came out in 2019. Apple & Intel have also been doing it for quite some time now. So by that standard a ton of modern computers would also have to be called "supercomputers".
    Reply
  • bit_user
    GenericUser2001 said:
    Having multiple dies that function as a single package is nothing special, and has been common for years now.
    That's not quite it, though. If you look at a supercomputer, it's multiple computers that work together to tackle larger tasks - it's a computer that's made up of smaller computers. In a similar sense, RTX Spark isn't just multi-die. Because its CPU die and GPU die are largely independent, you can argue that they're a similar setup of integrating multiple independent dies to form a larger whole.

    I need to add an asterisk to that, because the memory controller is contained in the CPU/SoC die. So, the GPU die isn't truly independent. However, in their Grace-Hopper, Vera-Rubin, etc. superchips, those CPUs and GPUs are truly independent, yet they're more closely tied together by their NVLinks than they are with any other chip in the system.

    GenericUser2001 said:
    AMD has been doing multiple dies for one package since Zen 2 came out in 2019.
    By contrast, if you look at something like AMD Ryzen, the chiplets in it are highly interdependent. The CCDs are nothing without the I/O die and the I/O die exists pretty much for the sole purpose of supporting the CCDs.
    Reply
  • GenericUser2001
    I would argue that the way Strix Halo works (which came out over a year ago) is functionally very similar to what Nvidia is doing with these RTX Spark machines - Strix Halo has one or two CPU chiplets, and then a die with the big iGPU/NPU/memory controller and i/o stuff. Only real difference layout wise is that Nvidia put the memory controller & i/o stuff on the CPU chip, while AMD put it on the iGPU chip. Or if you look at a breakdown of Apple's M5 Pro/Max chips, or Intel's Patherlake, you see similar tiles/chiplets being combined.
    Reply
  • bit_user
    GenericUser2001 said:
    I would argue that the way Strix Halo works (which came out over a year ago) is functionally very similar to what Nvidia is doing with these RTX Spark machines
    Functionally, but not structurally.

    GenericUser2001 said:
    Strix Halo has one or two CPU chiplets, and then a die with the big iGPU/NPU/memory controller and i/o stuff.
    If the GPU were in its own die, then I'd agree that it would fall in the same category. Being integrated into the I/O die makes it just like the AM5 chiplet-based Ryzens, where the GPU is part of the I/O die.

    The way I see it is that Nvidia's claim about superchips makes more sense for the server modules and they basically extended the analogy down to RTX Spark, where it's a little bit of a stretch since the GPU doesn't have its own memory controller. However, the GPU is connected via NVLink D2D, which gives it a greater degree of decoupling than you have in most chiplet/tile-based processors.
    Reply
  • alan.campbell99
    'Laying the foundation for the reinvention of the PC in the agentic AI era'

    That'll be a no from me.
    Reply