A 1 petaflop AI chip is coming to ordinary laptops this month
Key takeaways
- Nvidia RTX Spark pairs a Blackwell GPU, rated at up to 1 petaflop of AI performance, with a 20 core Grace CPU
- Up to 128GB of unified memory is shared between the CPU and GPU
- Microsoft, Asus, Dell, HP, Lenovo and MSI ship devices this autumn, with Acer and Gigabyte later
- No pricing has been announced, so treat any laptop list as provisional
One petaflop of AI performance is a figure that used to belong to a server rack. This month Nvidia is putting it in a laptop. The new RTX Spark platform pairs a Blackwell GPU with a 20 core Grace CPU and up to 128GB of unified memory, and the first Nvidia RTX Spark PCs are due to ship this autumn.
What an Nvidia RTX Spark PC actually is
RTX Spark is a platform rather than a single chip, and it runs Windows on Arm. The Blackwell GPU is rated at up to 1 petaflop of AI performance. The CPU is a 20 core Grace design, and both halves share one pool of memory that tops out at 128GB. That last detail matters for anyone who runs language models at home, and we explain why in our piece on what 128GB of shared memory means for local AI.
Microsoft, Asus, Dell, HP, Lenovo and MSI are shipping devices this autumn. Acer and Gigabyte follow later. Microsoft holds an event on 7 October to show an RTX powered Surface, which should be the first look at how the chip behaves in a finished machine.
Will your apps run?
Most Windows software is still built for x86, so RTX Spark machines run it through Microsoft's Prism emulation layer. Emulation usually costs some speed, which is why the native list is the part worth reading. Blender, DaVinci Resolve, Photoshop and Premiere are due to get native support, and those are the heavy creative tools where a laptop is judged.
If you are weighing an Arm laptop, our Windows 11 upgrade checklist for old PCs covers the compatibility checks that apply to any new Windows machine. Games and niche drivers are the usual weak spots on Arm, so check the specific titles and peripherals you rely on.
What nobody has said yet
There is no pricing. Nvidia has not named a laptop price, and a petaflop-class chip with 128GB of memory will not be cheap. The headline numbers also describe the top configuration, not every model. Specifications for lower-end models have not been published.
It is also worth remembering that the petaflop figure is a peak AI rating, which measures a narrow kind of maths at low precision. It tells you what the chip can do on a good day, not how a laptop feels when you edit video on battery. For the manufacturing side of why chips like this keep shrinking, see what a 2nm chip means.
Watch the Surface event on 7 October for the first price, the first battery claim and the first honest benchmark. Those three numbers will say more than the petaflop.
