That combination could appeal to developers, AI experimenters and creators whose projects are constrained by memory capacity as much as raw graphics throughput. But the GR1X remains an announced product rather than a reviewed one. ASUS has not stated a retail price or fixed launch date, and there are no independent results yet for sustained performance, fan noise, power consumption, local-model speed, or the practical compatibility of a professional Windows software stack on this system.
What ASUS has announced
ASUS announced the ProArt GR1X alongside its ProArt P16 and P14 laptops on September 2, 2026. All three use NVIDIA RTX Spark, but the GR1X is the desktop-oriented implementation rather than a laptop derivative in a different shell.
The listed GR1X configuration combines a 20-core NVIDIA Grace CPU with a 6,144-core NVIDIA Blackwell RTX GPU. NVIDIA lists the desktop RTX Spark configuration with a 140W TDP, while ASUS specifies a 240W power supply for the mini PC. Those are useful platform and system power figures, but they do not establish the wall-power draw of a finished retail configuration under any particular workload.
Physically, the system measures 150 × 150 × 51 mm, weighs 1.48 kg, and occupies about 1.1 liters. That is a meaningful size constraint for a machine intended to carry workstation-class aspirations. It may be attractive where desk space, portability between work locations, or the ability to place a compute node near a fast network matter. It also makes cooling behavior a central unanswered question. The available material does not provide independent evidence of how quietly the GR1X runs during long renders, model inference, compilation, or simultaneous GPU and CPU loads.
ASUS currently lists Black and Silver finishes. A separate report described Black and White models, but ASUS’s own specification table is the stronger basis for the present expectation.
The crucial detail: up to 128GB is not every configuration
The headline specification needs careful reading. ASUS lists up to 128GB of LPDDR5X unified memory and also says 64GB unified-memory configurations will be offered. Buyers should therefore not assume that every GR1X includes 128GB, or that the lower-memory version will have identical suitability for large AI or graphics workloads.
“Unified” has a specific practical consequence here. The LPDDR5X memory is shared by the Grace CPU and Blackwell GPU through NVLink-C2C, rather than being divided into conventional system RAM plus a separate pool of dedicated graphics VRAM. In principle, that design makes a single large memory pool available across CPU and GPU work. For workloads that require large datasets, sizable models, or complex assets to be accessible to accelerated code, capacity may matter more than a familiar distinction between RAM and VRAM.
That does not mean all 128GB is automatically free for a given application. Windows, background software, the active workload and its data all use the same overall resource. Nor does the specification alone tell us how memory allocation, software optimization, or real sustained throughput will compare with a conventional x86 workstation paired with a discrete GPU. Those are questions for application-specific testing.
Still, the capacity is the GR1X’s clearest differentiator in this physical size class. A system with 64GB shared memory could be a sensible compact machine for many development and creative jobs. A 128GB configuration may be the more consequential option for users specifically trying to keep larger workloads local. Until ASUS publishes a full SKU list and pricing, however, it is impossible to judge the premium attached to that step up.
A desktop that emphasizes network and display connectivity
The networking and I/O specification makes clear that ASUS is positioning the GR1X for a desk rather than merely as a small personal computer. The confirmed connections include:
- 10GbE through an RJ45 port
- Wi-Fi 7 and Bluetooth 5.4
- One HDMI 2.1 port
- Three USB 3.2 Gen 2×2 Type-C ports with DisplayPort Alt Mode
- One additional USB-C port for power-delivery input
ASUS says the machine can connect up to four displays. The three high-speed USB-C ports with DisplayPort Alt Mode are particularly relevant to the ProArt audience, where multiple monitors, fast external storage, capture hardware, docks, and color-critical display arrangements can matter as much as peak compute specifications.
The 10GbE port is equally notable. On a suitable network, it could make the GR1X easier to use with shared project storage, a NAS, or a broader studio workflow than a typical compact PC limited to lower-speed Ethernet. The specification does not establish actual network throughput in every environment, since that depends on the network infrastructure, cabling, storage and drivers. But including 10GbE avoids an immediate connectivity bottleneck for buyers who already work with large local files.
Storage is less fully explained. ASUS lists M.2 NVMe PCIe 5.0/4.0 SSD storage slots, yet has not specified the number of slots, factory capacities, or end-user upgrade terms. That omission matters. A machine pitched for local AI assets and large creative projects needs more than fast storage in principle; prospective buyers need to know how much can be installed, whether it can be expanded, and whether internal access affects support.
Windows 11 on Arm is an advantage and a qualification
RTX Spark supports Windows 11, but this is an Arm-based platform, not a conventional x86 Windows desktop. That distinction will shape the buying decision more than the compact enclosure or the AI throughput claim for many professionals.
Microsoft says Prism, its emulator for running both 32-bit and 64-bit x86 applications on Windows on Arm, will be present and optimized for RTX Spark-powered PCs. This substantially broadens the potential Windows application library: software does not have to be native Arm code simply to have an available route to run.
But an emulation route is not the same as a blanket certification of every workflow. It does not demonstrate that an application is native Arm, that its plug-ins work, that specialized peripherals have suitable drivers, or that performance under emulation meets a professional user’s expectations. It also does not establish that a given program has been independently tested on the GR1X specifically.
For a general-purpose home or office machine, that may be an acceptable uncertainty if the key applications are known to work. For a creator, engineer or developer buying a compact workstation, it deserves more caution. The safest approach is to identify the exact software chain before purchase: primary applications, plug-ins, input devices, external hardware, codecs, security tools, development toolchains and any organization-managed software. Native Arm support, successful x86 emulation and full workflow compatibility are related but separate questions.
That is not a verdict against the GR1X. Windows on Arm’s emulation layer is precisely intended to make broader compatibility possible. It is instead a reminder that the transition cost is borne differently by each user. Someone running a focused, validated set of applications may see the compact unified-memory design as compelling. Someone dependent on obscure extensions or hardware-specific utilities may reasonably wait for compatibility reports and retail units.
Treat performance headlines as vendor capability claims
NVIDIA and ASUS promote up to 1 petaFLOP of FP4 AI performance for this platform. ASUS also presents examples including local operation of LLMs up to 120 billion parameters, rendering 90GB 3D scenes, and AAA gaming at 1440p above 100 frames per second.
These claims describe the vendors’ intended capabilities, not independently measured GR1X results. The distinction is particularly important for the petaFLOP figure. FP4 is a low-precision format used for certain AI workloads, and the number is not directly comparable with traditional FP32 computing figures or the gaming performance of a GeForce GPU. Recasting it as a general measure of workstation speed would be misleading.
Likewise, a claim that a system can run a 120B-parameter model does not answer the questions that determine whether it is useful in practice: the model format and quantization, context length, available memory after the operating system and other software, token generation rate, prompt processing speed, and whether the chosen software is optimized for the platform. A model that technically fits is not necessarily responsive enough for every interactive workload.
The same discipline applies to rendering and games. The published material does not provide independent GR1X benchmarks, game settings, image-quality settings, frame-time data, thermals, or test methodology. Buyers should resist treating a creative mini PC’s advertised AI format throughput as a shortcut to gaming expectations.
Who should watch the GR1X—and who should wait
The GR1X looks most interesting for users with a clearly defined need for local accelerated workloads, abundant shared memory, small physical dimensions and fast networking. Potential examples include developers experimenting with local AI tools, teams that want a compact desk-side compute system, and creators whose workflows can benefit from a multi-display machine connected to rapid shared storage.
The 64GB and 128GB choices will matter greatly. Buyers should match the configuration to actual working sets rather than buy on the assumption that any RTX Spark badge guarantees the same model capacity or project headroom. They should also account for storage needs, since internal configuration information remains incomplete.
Conversely, users should wait if their decision depends on a known price, a fixed retail date, verified application behavior, or independently measured sustained output. As of September 8, 2026, ASUS’s product page presents the GR1X as upcoming, while broader reporting on RTX Spark availability does not supply a confirmed GR1X launch schedule. A general expectation that some RTX Spark devices may ship in October cannot be converted into a promise for this particular ASUS mini PC.
The missing price is especially important because the GR1X will need to compete against larger conventional desktops that may offer different upgrade paths, established x86 compatibility, and discrete graphics configurations. Its small footprint and shared-memory architecture could justify a premium for the right workflow, but value cannot be assessed until ASUS discloses retail configurations and pricing.
The real test comes after retail hardware arrives
ASUS has given the ProArt GR1X a technically distinctive foundation: NVIDIA Grace and Blackwell silicon, up to 128GB of unified LPDDR5X memory, 10GbE, Wi-Fi 7, multi-display connectivity and a compact chassis. It is more specific and potentially more useful than a generic “AI PC” announcement.
Yet its appeal rests on issues that specification sheets cannot settle. Reviews will need to test sustained performance in the actual enclosure, thermals and acoustics over long workloads, local-model responsiveness, storage configuration, and the reliability and speed of important native and emulated Windows applications. Until then, the defensible conclusion is promising but conditional: the GR1X could become a highly capable compact Windows workstation for carefully validated Arm-compatible or Prism-supported workflows, but its price, availability and real-world behavior remain the facts that will determine whether it earns that role.