NVIDIA now sells two very different machines under the NVIDIA Spark personal AI computer banner, and confusing them costs money. One is the DGX Spark, a deskside developer box that shipped in late 2025. The other is RTX Spark, the consumer superchip platform NVIDIA unveiled at Computex 2026 to put a Blackwell GPU and a Grace-class Arm CPU inside slim Windows laptops and tiny desktops. Same word, two roadmaps, different buyers.

Quick Answer

DGX Spark is the developer-focused deskside AI computer that launched first and targets people building and fine-tuning models locally. RTX Spark is the Computex 2026 consumer platform, an Arm-plus-Blackwell superchip with up to 128GB of unified memory, landing in laptops and compact desktops from major brands in the second half of 2026. SA buyers can build a capable AI PC today and watch RTX Spark hardware arrive later.

DGX Spark: the developer machine that shipped first

DGX Spark is NVIDIA's deskside personal AI supercomputer, built around the Grace Blackwell architecture and aimed squarely at developers, researchers and data scientists who want a model-fine-tuning rig next to the desk instead of a cloud bill. It pairs a large pool of coherent memory with serious tensor throughput so you can prototype, run inference on sizeable models, and train smaller ones locally.

The point of DGX Spark was never gaming. It is a compute appliance: load big models, iterate, and keep your data on your own hardware. For an SA studio or lab worried about cloud costs and data residency, that local-first pitch lands hard.

Key DGX Spark specs

The GB10 Grace Blackwell Superchip at its core combines a 20-core Arm CPU with a Blackwell GPU. The memory pool is 128GB of unified LPDDR5x, fully coherent across CPU and GPU, paired with a 4TB NVMe SSD. Networking includes 10 Gigabit Ethernet, Wi-Fi 7, Bluetooth 5.3, and a ConnectX-7 port that lets two units be paired directly for models too large to fit in a single 128GB pool. The DGX Spark launched at a global retail price of around USD 4,699.

RTX Spark: the consumer superchip from Computex 2026

At Computex 2026 NVIDIA pulled the wraps off RTX Spark, a superchip co-developed with MediaTek that fuses two chiplets into one SoC. The GPU side is a large Blackwell design with 6,144 CUDA cores and fifth-generation Tensor Cores running FP4 precision, rated at one petaflop of FP4 AI performance, linked over a high-bandwidth NVLink-C2C interconnect to a 20-core Grace Arm CPU. Memory tops out at 128GB of unified memory with up to 300 GB/s bandwidth. The 70-billion-transistor die is built on TSMC's 3nm process. Across the broader family, configurations range from 16GB to 128GB, so entry-level RTX Spark laptops will land at a lower price point than the flagship spec.

What that means in practice

This is NVIDIA's move to make Windows an agentic AI platform, where local models handle assistant and creative tasks on-device rather than round-tripping to a server. NVIDIA rates the platform for running 120-billion-parameter language models locally, editing 12K video, and delivering roughly 100 FPS at 1440p gaming through DLSS 4.5 and Multi Frame Generation. The same silicon is meant to scale from all-day-battery slim laptops to small ultra-efficient desktops. Morgan Stanley's Computex channel checks pegged RTX Spark flagship N1X systems at around USD 2,899.

When you can buy it

RTX Spark laptops and compact desktops are slated for the second half of 2026 from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI, with Acer and GIGABYTE to follow. NVIDIA expects over 30 laptops and over 10 desktops in the initial wave. NVIDIA also outlined a multi-generation roadmap beyond the first chip, moving to a Vera CPU and Rubin GPU on LPDDR6 in a future generation, so this is a platform, not a one-off product.

DGX Spark vs RTX Spark: which is for you

If you are a developer who needs to fine-tune and run large models locally right now, DGX Spark is the deskside answer and it already exists. If you are a creator, gamer or professional who wants on-device AI baked into a normal Windows laptop or small desktop, RTX Spark is what you wait for, arriving later in 2026.

For most SA readers the practical path is simple: you do not have to sit idle until RTX Spark ships. A current AI-capable PC handles local image generation, coding assistants and inference today. Comparing the AI PC range at Evetech against your workload now is smarter than holding out for a launch window, and if the GPU is the part you care about most, the GPU best sellers list shows what is moving locally.

SA availability and the sensible plan

NVIDIA launch dates are global, and SA retail timing usually trails the announcement window while distribution and pricing settle. So treat the second-half-2026 RTX Spark date as a target, not a promise for local shelves. The low-risk plan is to spec an AI PC around the work you do today, then evaluate an RTX Spark machine once real units and Rand pricing land here.

It is also worth noting that Windows on Arm software compatibility was a genuine friction point on earlier Qualcomm-based Windows machines. RTX Spark runs a custom Arm architecture from NVIDIA rather than a Snapdragon, and the software story hinges on how well the broader Windows on Arm ecosystem matures through the rest of 2026. If you need specific professional applications and they have not confirmed Arm compatibility, that is a real reason to wait for compatibility reports from early adopters before committing.

Frequently Asked Questions

Is the DGX Spark the same as the RTX Spark?

No. DGX Spark is the earlier deskside developer machine. RTX Spark is the consumer superchip platform announced at Computex 2026 for laptops and small desktops. They share a name and an architecture family but target different users.

When does RTX Spark hardware ship?

NVIDIA targets the second half of 2026 for the first wave of RTX Spark laptops and compact desktops from major brands. SA retail availability typically follows the global window once distribution settles.

How much memory does RTX Spark have?

The platform ranges from 16GB to 128GB of unified memory, with a Blackwell GPU and a 20-core Grace-class Arm CPU on one SoC. Exact configurations will vary by laptop and desktop model.

Should I wait for RTX Spark or buy an AI PC now?

If your AI work is happening today, a current AI-capable PC does the job and you avoid an open-ended wait. If you specifically want on-device AI in a slim Windows laptop with ARM-native performance, that is the case for waiting on RTX Spark.

Is Spark meant for gaming?

DGX Spark is a compute machine, not a gaming box. RTX Spark is positioned for AI, creation and gaming on consumer Windows devices, so the consumer line is the one with gaming in mind.

Want local AI power without waiting on a launch window? Spec a machine around the work you do today. Explore the AI PC range at Evetech and get building now.