# Nvidia confirms RTX Spark laptops for October 16, and its compact desktops move to November

URL: https://onticpost.com/nvidia-rtx-spark-pc-release/
Published: 2026-10-16
Updated: 2026-10-15
By: Hill, Onticpost

> Nvidia says RTX Spark laptops arrive October 16, but its compact desktops slipped to November and Microsoft says the GPU cannot use all 128 GB of memory.

Nvidia said on October 13, 2026, that the first RTX Spark laptops will be available on Friday, October 16, confirming the date Microsoft and Nvidia set at a Windows event in San Francisco on October 7. Microsoft opened pre-orders that day for RTX Spark laptops, including its own Surface Laptop Ultra. The chip pairs an Nvidia Blackwell graphics processor with a 20-core Nvidia Grace processor and up to 128 GB of unified memory, Nvidia says. As of October 15, the laptops have a date but have not yet reached customers, and the compact RTX Spark desktops that Nvidia said in September would come in October are now set for November.

**Takeaway points**

- Nvidia’s October 13 blog says RTX Spark laptops will be available on Friday, October 16, the date Microsoft and Nvidia gave on October 7.
- Microsoft’s footnote says the graphics chip can address less than the full 128 GB of memory, and Nvidia’s compact desktops have moved from October to November.
- We have not tested any RTX Spark machine, and no maker has published a test that we have checked.

## What did Nvidia and Microsoft confirm about the RTX Spark release date?

They confirmed that RTX Spark laptops are due on Friday, October 16, and nothing in the makers’ own documents says any have been delivered before that. Nvidia’s blog of October 7, published after the Windows AI and Surface event at Dogpatch Studios in San Francisco, says laptop pre-orders opened that day. Its blog of October 13 repeats that “RTX Spark laptops will be available on Friday, Oct. 16.”

Microsoft’s Windows blog of the same day names the machines that opened for pre-order: models from ASUS, Dell, HP, Lenovo, MSI and Microsoft’s own Surface line. It says all of them “begin shipping October 16.” The post lists the Dell XPS 16 Creator Edition, the HP OmniBook Ultra 16, the Lenovo Yoga 9n 2-in-1, the MSI Prestige N16 Flip AI+, the ASUS ProArt P16 and P14, and the Surface Laptop Ultra.

| Date | Document | What it says about timing |
| --- | --- | --- |
| May 31, 2026 | Nvidia press release, GTC Taipei | Laptops and compact desktops “available this fall” |
| September 2, 2026 | Acer press release, IFA | Acer availability “will be announced at a future date” |
| September 3, 2026 | Nvidia blog, IFA | New compact RTX Spark PCs “coming in October” |
| September 3, 2026 | Lenovo press release, IFA | Pricing and availability “announced at a later date” |
| October 7, 2026 | Nvidia blog | Laptops available October 16, compact desktops in November |
| October 7, 2026 | Microsoft Surface blog | Surface Laptop Ultra from October 16, Dev Box in November |
| October 13, 2026 | Nvidia blog | Laptops available Friday, October 16 |

Microsoft’s Surface posts are the only maker documents we found that state prices: $2,599 for the Surface Laptop Ultra and $5,999 for the Surface RTX Spark Dev Box, a small desktop. Neither Lenovo’s release nor Acer’s gives a price.

## What is inside the RTX Spark chip?

RTX Spark is one package that joins a graphics processor and a main processor, and Nvidia’s own figures for it have stayed the same since May. Its press release of May 31, issued at GTC Taipei, describes a Blackwell graphics processor with 6,144 CUDA cores, joined by Nvidia’s NVLink-C2C link to a 20-core Grace processor. MediaTek worked with Nvidia on the design of that processor, the release says. Nvidia’s October 7 blog puts the link between the two at 600 GB/s.

| Part | Nvidia’s figure | Where it is stated |
| --- | --- | --- |
| Graphics processor | Blackwell, up to 6,144 cores | May 31 release, October 7 blog |
| Main processor | Grace, up to 20 cores | May 31 release, October 7 blog |
| Link between them | 600 GB/s | October 7 blog |
| Memory | Up to 128 GB, unified | All Nvidia documents |
| AI compute | 1 petaflop at FP4 | May 31 release, October 7 blog |

The petaflop figure has a condition. A footnote in Microsoft’s Surface post says it is “Based on theoretical FP4 performance of 1 petaflop using the sparsity feature.” FP4 is a 4-bit number format, and sparsity is a method that skips some of the arithmetic. None of the Nvidia documents we saved carries that note.

The machines run Windows on Arm. Nvidia’s October 13 blog lists the first PyTorch build with CUDA for Windows on Arm and says 97% of the top 15,000 Python wheels are ready, with Nvidia working on the rest.

## Is 128 GB of memory on every RTX Spark PC?

No. The 128 GB is the most the chip can carry, and the makers’ documents show smaller amounts in some models. Lenovo’s IFA release gives the Yoga Pro 9n up to 128 GB and the Yoga 9n 2-in-1 up to 64 GB. Microsoft’s post says of the Surface Laptop Ultra that “Memory varies by configuration.”

The same footnote limits what the graphics chip can use. In a unified design the main processor and the graphics processor share one pool. Microsoft writes that the maximum the graphics processor can address depends on the system and the workload “and is less than the total.” It does not say by how much.

As a fact about capacity only, 128 GB is 80 GB more than the 48 GB of unified memory in our test machine, an Apple M4 Pro, or about 2.7 times as much. We have written about what that 48 GB holds in [two models at once on a 48 GB Mac: an idle second model cut the first from 49.4 tokens a second to 14.5](https://onticpost.com/two-models-at-once-48-gb-mac/). We make no comparison of speed between the two kinds of machine.

## Did Nvidia’s compact RTX Spark desktops move from October to November?

Yes, by Nvidia’s own dates. Its IFA blog of September 3 said “New compact NVIDIA RTX Spark Windows PCs are also coming in October.” Its October 7 blog says “Compact desktops will be available for sale in November.” Neither post explains the change.

Microsoft’s dates agree with the later one. Its Surface post says the Dev Box will begin shipping to customers in November. Its Windows post says the “dev boxes and mini desktops will be coming later this year.”

The makers’ lists of partners do not match either. Nvidia’s October 7 blog says systems are coming from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte. Microsoft’s list of PCs open for pre-order that day leaves out Acer and Gigabyte. Nvidia’s May release had already said models from those two would follow. Acer’s own release from IFA calls its machine, the Acer SFF RTX Spark, a design. Acer says it carries up to 128 GB of memory and that availability “will be announced at a future date.”

The documents also differ on when RTX Spark was first shown. Nvidia’s press release is dated May 31. Microsoft’s Windows post of October 7 says Microsoft and Nvidia introduced RTX Spark “In June.”

## What are Nvidia’s llama.cpp, vLLM and PAIR claims based on?

Nvidia’s speed claims for llama.cpp and vLLM were measured on other Nvidia hardware, not on RTX Spark. Its IFA blog says llama.cpp gives “up to 1.9x higher throughput” on a GeForce RTX 5090. It says vLLM gives 1.2x on an RTX PRO 6000 Blackwell Workstation Edition and up to 1.4x on two DGX Spark machines. The blog says the gains reach users of LM Studio and Ollama too, two apps we compared in [Ollama vs LM Studio on an M4 Pro: the same Gemma 4 file runs at the same speed](https://onticpost.com/ollama-vs-lm-studio-speed-m4-pro-gemma-4/). It does not say what the 1.9x is measured against.

Microsoft’s comparison is with an Apple laptop. Its Windows post claims gains over a MacBook Pro 16-inch with M5 Pro of up to 2.1 times faster time to first token, 4.3 times faster image generation and 6.2 times faster video generation. The footnotes say the machines tested were preproduction RTX Spark PCs with 64 GB, not 128 GB, against a MacBook Pro with 64 GB. The first figure comes from testing Microsoft commissioned in September 2026, using llama.cpp with Qwen3.5 27B and an 8,192-token prompt. The other two come from Nvidia’s own testing in ComfyUI. These are the makers’ claims, and we have not repeated them. How a speed figure can mislead is the subject of [how to measure tokens per second in LM Studio without the cache fooling you](https://onticpost.com/measure-tokens-per-second-lm-studio/).

PAIR, Nvidia’s Personal AI Router, is software that spreads AI work across the PCs on a home network. Nvidia’s IFA blog launched it as a beta and said it worked with Ollama and LM Studio. The beta supported Apple M4 or newer chips as well as Nvidia graphics cards. The October 13 blog says PAIR “is officially moving out of beta today.” It adds llama.cpp support, downloads from Hugging Face and RTX Spark support. Nvidia says it is free and open source, and runs on Windows, Linux and macOS.

Nvidia also named a model for the machines. The October 13 blog says Nemotron 3.5 Bolt 74B, a 74 billion parameter mixture-of-experts model with 9 billion active, fits in 24 GB using 2-bit quantization and is due on Thursday, October 15.

## What has not been published about RTX Spark?

No maker’s own document we found gives a price for a 128 GB configuration, a price outside the United States, or a ship date for Acer or Gigabyte. Lenovo’s release from September 3 left its pricing for later, and we did not find a Lenovo document with a price that we could save. Microsoft has not said how much of the 128 GB the graphics chip can address. Nvidia has not said why the compact desktops moved from October to November.

There is no independent test yet either. The makers’ speed figures come from preproduction machines with 64 GB. Nvidia’s llama.cpp and vLLM gains were measured on other hardware. Delivery itself has not happened as of October 15, the day before the laptops are due.

We have not tested any RTX Spark machine, and this piece contains no speed, memory or quality figure of our own. Every figure above is the maker’s, taken from documents by Nvidia, Microsoft, Lenovo and Acer that we saved on October 15, 2026.

## Sources

- [Nvidia blog, All Systems Go: Global PC App Ecosystem Ready for NVIDIA RTX Spark, October 13, 2026](https://blogs.nvidia.com/blog/local-ai-rtx-spark-laptops-software-updates/)
- [Microsoft Surface blog, Pre-order our most powerful Surface devices ever, October 7, 2026](https://blogs.windows.com/devices/2026/10/07/pre-order-our-most-powerful-surface-devices-ever/)