NVIDIA says RTX Spark laptops are on sale. The petaflop is NVIDIA's figure.
NVIDIA's October 16 release says RTX Spark laptops are on sale from six makers, with up to 128 gigabytes of memory. The petaflop figure is NVIDIA's.
The short version
The release says RTX Spark laptops are on sale from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. It says they combine a Blackwell GPU and a Grace CPU, with up to 128 gigabytes of unified memory and up to one petaflop of FP4 AI performance. Those upper figures are NVIDIA's.
The availability sentence
NVIDIA's newsroom post, dated October 16, 2026, says RTX Spark laptops are available from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. The summary says the machines bring Blackwell graphics and NVIDIA's software stack to slim Windows PCs, with up to 128 gigabytes of unified memory and up to one petaflop of FP4 AI performance. Jensen Huang is quoted on creating, gaming, and local agents. Yusuf Mehdi of Microsoft is quoted on a Windows PC that can run locally and still reach the cloud. Both quotations are in the same release.what a process node measures.
Up to is a ceiling
A laptop on the list can sit under 128 gigabytes and under a petaflop and still be an RTX Spark machine as the release defines the family. The product page describes building and running models on the PC with the CUDA stack. It does not, in the lines this brief relies on, print a single street price. A ceiling without a price is still a ceiling. It is not a receipt. The document named in the sources is the record this page will keep.the hardware desk.
One chip, two names
The release says the foundation is a single superchip combining a Blackwell RTX GPU and a Grace CPU, and that MediaTek worked with NVIDIA on the CPU. Unified memory is the reason the release gives for running a large model and a creative file at the same time. Shared memory is an architecture claim. It is not a measurement of a particular project's frame time. Asked directly, Does the release say the work never touches a cloud? It quotes a Microsoft executive on hybrid use: local when that fits, and the cloud when it is needed. NVIDIA's local-model sentence and that hybrid sentence are both in the release. They describe different moments.Futureweb.
October 19 is the wire's echo
engineering.com's October 19 item introduces the laptops for local AI and accelerated computing. It is the line that put the product on a hardware wire two days after NVIDIA's release. A second account can confirm the announcement traveled. It does not add a benchmark this page will adopt. Asked directly, What ceilings does NVIDIA print? Up to 128 gigabytes of unified memory, and up to one petaflop of FP4 AI performance. The word up to means a configuration, not a promise that every laptop in the list carries the maximum.an earlier denial was a different card.
Local agents are the use NVIDIA names
The release says buyers can run agents that edit, generate assets, and write code, and it names creative applications and games as day-one workloads. Those are examples in a product announcement. An example is not a test log. The Microsoft sentence beside them keeps the cloud in the picture for the moment when local compute is not the one the user wants. A public scheduler is a different stack. The document named in the sources is the record this page will keep.Run:ai's public source.
What this page will not benchmark
It will not convert one petaflop of FP4 into a training time. It will not invent a starting price the release's availability list does not print in the passages used here. It will not treat a hybrid quote as a claim that the model weights stay on the SSD. The record is the October 16 release, the product page, and the October 19 retelling. Asked directly, What does engineering.com add? An October 19 account that the laptops are for local AI and accelerated computing. It is a second telling of the availability announcement, not a teardown of a unit this desk bought.
What a review unit would change
A teardown with a measured memory figure, a street price, or a workload timed on one of the six brands would replace the ceiling with a configuration. Until then, the dependency is NVIDIA's availability claim and the upper bounds the company attached to the family. Asked directly, Which brands does the release name for laptops on sale? ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with further systems to follow. The date on the release is October 16, 2026.
A different GeForce story stays filed
This paper already filed a denial about a GeForce RTX 5090 interview. RTX Spark is a laptop family with a Grace CPU and a Blackwell GPU in one module, sold as a Windows PC. The names share a brand and not a document. The October 16 release is the document for this page. Asked directly, Who does NVIDIA say designed the CPU with it? MediaTek, described as a collaborator on the custom CPU. The GPU generation named in the release is Blackwell, and the CPU name in the release is Grace.
What stays on the record
October 16, six laptop brands, Blackwell and Grace, up to 128 gigabytes, up to one FP4 petaflop, a Microsoft sentence about hybrid use, and an October 19 account of the same announcement. The petaflop is NVIDIA's ceiling, not a stopwatch. The short answer this page will stand on is this: The release says RTX Spark laptops are on sale from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. It says they combine a Blackwell GPU and a Grace CPU, with up to 128 gigabytes of unified memory and up to one petaflop of FP4 AI performance. Those upper figures are NVIDIA's.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
Which brands does the release name for laptops on sale?
ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with further systems to follow. The date on the release is October 16, 2026.
What ceilings does NVIDIA print?
Up to 128 gigabytes of unified memory, and up to one petaflop of FP4 AI performance. The word up to means a configuration, not a promise that every laptop in the list carries the maximum.
Who does NVIDIA say designed the CPU with it?
MediaTek, described as a collaborator on the custom CPU. The GPU generation named in the release is Blackwell, and the CPU name in the release is Grace.
What does engineering.com add?
An October 19 account that the laptops are for local AI and accelerated computing. It is a second telling of the availability announcement, not a teardown of a unit this desk bought.
Does the release say the work never touches a cloud?
It quotes a Microsoft executive on hybrid use: local when that fits, and the cloud when it is needed. NVIDIA's local-model sentence and that hybrid sentence are both in the release. They describe different moments.