Dell Builds the Only RTX Spark Mini PC With an SD Card Reader

Dell’s XPS RTX Spark Desktop arrived at Computex 2026 in Taipei as the only compact desktop in the inaugural NVIDIA RTX Spark wave to include an SD card reader. Every competing machine from ASUS, MSI, HP, Lenovo, and Microsoft announced at the same event left the slot out. The desktop runs the RTX Spark SoC (system-on-chip), an ARM-based design pairing a 20-core Grace CPU and a Blackwell GPU on a single TSMC 3-nanometer die, supports up to 128GB of unified LPDDR5X memory, and carries NVIDIA’s stated AI throughput of one petaflop at FP4 (4-bit floating-point) precision. Dell styled the chassis in dark gray aluminum after Apple’s Mac Studio, with ventilation slots along both sides and a removable bottom panel that opens for storage access without full disassembly.

Apple’s Mac Studio has led the compact high-performance desktop tier largely unchallenged since 2022. RTX Spark, announced jointly by NVIDIA and Microsoft at NVIDIA GTC Taipei during Computex on May 31, 2026, is the first Windows platform engineered to contest that space directly. Six OEM partners unveiled competing RTX Spark mini PCs at the same event, and none has disclosed a price.

A Chip Built on the DGX Spark Bloodline

RTX Spark has a direct commercial ancestor. NVIDIA first shipped the Grace Blackwell architecture inside the $3,999 DGX Spark AI mini-workstation, a Linux-based system sold through authorized resellers to researchers and AI developers. The consumer RTX Spark SoC carries the same architectural family, re-scoped for Windows, for mainstream OEM manufacturing, and for thermal envelopes starting around 45 watts. NVIDIA built the chip internally under the codename N1X; the collaboration with MediaTek produced the 20-core Grace processor at the platform’s core, engineered to fit inside laptops as thin as 14 millimeters and compact desktops drawing between 45 and 80 watts.

The GPU block carries the same CUDA core count as the desktop GeForce RTX 5070, connected to the CPU over NVIDIA’s NVLink-C2C chip-to-chip interconnect rather than a standard PCIe bus. Both processor and accelerator draw from a single shared LPDDR5X memory pool. For AI inference workloads that move large model weights between CPU and GPU, that shared architecture eliminates the copy overhead present when the two components draw from separate memory domains. The platform’s top configuration:

• 6,144 Blackwell CUDA cores (same count as the desktop GeForce RTX 5070)
• 20 ARM-based Grace CPU cores, co-developed with MediaTek on a 70-billion-transistor die
• 128GB maximum unified LPDDR5X memory at 300 GB/s bandwidth
• 1 petaflop of AI compute at FP4 precision with sparsity enabled
• 3nm TSMC process node; NVLink-C2C chip-to-chip interconnect

At the same Computex session, Jensen Huang committed to a three-generation RTX Spark roadmap. After the current Grace Blackwell generation, a Vera Rubin Spark switches to LPDDR6 memory; a Rosa Feynman Spark follows. NVIDIA stated the commitment publicly, with OEM partners present, because companies designing new chassis, cooling systems, and retail channels around an unproven processor architecture need more than a single product cycle to justify the investment.

The Port Choice That Set Dell Apart

Walk the Computex floor and each RTX Spark mini PC looks nearly identical from behind: four USB-C connectors at 20 Gbps each, an HDMI output, and a 10-gigabit Ethernet (10GbE) port. That configuration appears on every announced entry in the wave. Dell’s XPS RTX Spark Desktop adds an SD card reader to that stack, making it the only machine in this first cohort with direct card-media access. Dell’s Computex creator lineup post cited the reader as the key differentiator from other RTX Spark mini PCs, noting its absence from competing machines in the same wave.

Dell also positioned the desktop alongside the XPS 16 Creator Edition laptop in its creator lineup, a machine carrying the same RTX Spark chip in portable form with a Tandem OLED display and True Black HDR 600 certification. The desktop runs that chip at higher sustained power than the laptop version because the larger enclosure provides more thermal headroom. A removable lid on the underside opens directly to the storage drive, removing the need for full disassembly to swap the drive.

ASUS has published more hardware detail than any other OEM in the wave. Its ProArt Mini PC fits in a 150 by 150 by 51-millimeter chassis, includes a 140-watt Elite Thermal cooling system, and ships with Wi-Fi 7, Bluetooth 5.4, and a PCIe Gen5 M.2 storage slot. MSI’s EdgeMesa N AI+ is physically the smallest machine in the cohort and supports four external displays via one HDMI and three USB-C connections. Dell has not released chassis dimensions or a cooling figure for its RTX Spark desktop; Computex hands-on accounts noted it stands taller than the GB10 DGX Spark clone designs it sits alongside in the broader market.

The Six-OEM Launch Cohort

The OEM Lineup

Dell is one company in a six-OEM wave. ASUS, HP, Lenovo, MSI, and Microsoft each revealed RTX Spark mini PCs at Computex within days of each other, with NVIDIA naming all six as part of the fall launch cohort. Microsoft’s entry, the Surface RTX Spark Dev Box, ships with 128GB of unified memory and a 100-watt thermal design, positioned at developers wanting to run large AI language models on-device. NVIDIA has stated RTX Spark can host models of up to 120 billion parameters locally without relying on cloud infrastructure, which is the primary use case the Dev Box targets.

The ConnectX-7 Departure

Every machine in the group runs the same chip. One consistent departure from the enterprise DGX Spark lineage is the absence of the ConnectX-7 NIC, a 400-gigabit networking card that let GB10-based workstation systems link multiple nodes for distributed AI work. RTX Spark desktops drop that component, positioning themselves as consumer creative and AI development workstations competing on form factor and price with the Mac Studio and AMD’s Ryzen AI Halo platform. AMD’s Ryzen AI Halo is also a unified-memory APU (Accelerated Processing Unit) design competing for the same creator workstation segment, though from an x86 architecture rather than ARM.

No OEM Has Released a Price

RTX Spark laptops are expected to start at or above $2,000 based on positioning against the MacBook Pro and Surface Laptop Ultra tiers, per coverage of the Computex laptop announcements. Desktop mini PC pricing has no direct consumer reference point in this performance class; the DGX Spark commanded $3,999 for an AI-research audience that required enterprise networking, and the RTX Spark consumer machines are aimed at a different buyer. No OEM has set a retail price for any RTX Spark desktop. Lenovo confirmed an RTX Spark desktop at Computex but has released fewer hardware details than any other OEM in the group; Acer and GIGABYTE also join the fall cohort at a later stage.

Dell has not issued a formal press release for the XPS RTX Spark Desktop. The machine appeared in NVIDIA’s OEM partner materials and Computex floor coverage, but Dell’s own newsroom, which published detailed materials for the XPS 16 Creator Edition laptop at the same show, released nothing equivalent for the mini PC.

Can RTX Spark Match the Mac Studio?

Apple’s Mac Studio ships today in M4 Max and M3 Ultra configurations. The base M4 Max model carries a price of $1,999 and supports up to 128GB of unified memory at its highest tier; the M3 Ultra runs to $3,999 and accepts up to 192GB, a ceiling RTX Spark’s current specification does not reach. An M5-generation Mac Studio carrying M5 Max and M5 Ultra chips has been expected for 2026, but supply constraints affecting high-memory Apple silicon systems have pushed the likely arrival to around October 2026, per Macworld’s tracking of Apple’s M5 silicon pipeline.

On AI compute numbers, NVIDIA rates RTX Spark at one petaflop of FP4 AI performance. Apple’s Neural Engine in current M-series chips measures 38 TOPS (Tera Operations Per Second) under a different testing methodology and precision tier. These figures use incompatible benchmarking conventions; Apple’s Neural Engine handles a different class of machine learning inference than NVIDIA’s Blackwell accelerator, and the raw numbers do not translate directly to real application speed. RTX Spark runs CUDA (Compute Unified Device Architecture) natively, NVIDIA’s parallel computing framework that underlies the majority of AI research and production toolchains in use today. Running CUDA pipelines on Apple Silicon requires conversion to Apple’s Metal API or the use of compatibility shims.

Apple’s M4 Max GPU tops out at 40 cores in Apple’s tile-based shader architecture; RTX Spark’s Blackwell GPU carries 6,144 CUDA cores in a general-purpose compute model built for NVIDIA’s own framework. The Mac Studio supports Thunderbolt 5 at 120 Gb/s for external storage and drives up to eight external displays with the M3 Ultra chip. Neither of those connectivity specifications has a confirmed RTX Spark desktop equivalent at this point. Independent benchmark data from production applications on actual RTX Spark hardware has not been published.

The App List Still Being Written

Moving to ARM means applications compiled for x86 processors either need dedicated ARM ports or run through Prism, Microsoft’s x86-to-ARM translation layer for Windows. Early Windows-on-ARM products from Qualcomm and Microsoft ran into meaningful native software gaps at launch; NVIDIA pushed to secure native ports from major publishers ahead of RTX Spark’s fall arrival, to reduce dependence on emulation from day one. NVIDIA and Microsoft confirmed at their Computex announcement that Prism covers broad x86 compatibility, including Windows games through native anti-cheat implementations from Epic’s Easy Anti-Cheat and BattlEye.

For creative software, several major publishers confirmed native RTX Spark builds at Computex. Adobe announced it is rebuilding both Photoshop and Premiere Pro from the ground up for the platform, claiming 2x improvements in AI and graphics performance. Other confirmed native applications at Computex 2026:

• Blender
• DaVinci Resolve
• Maxon Cinema4D and Redshift
• Topaz Photo AI
• CapCut
• Cubase and Bitwig Studio
• Affinity by Canva
• MATLAB (via Prism)

Final Cut Pro, Logic Pro, and Apple’s professional suite have no Windows version; a creator whose full pipeline depends on those tools would need to migrate to cross-platform alternatives before switching to any RTX Spark desktop. For developers and creators already inside NVIDIA’s software stack, the RTX Spark package includes CUDA, TensorRT, DLSS, Reflex, OptiX, and G-SYNC natively, the same toolset available on discrete desktop RTX GPUs, with existing CUDA workloads carrying over without porting.

Dell has not confirmed a price or ship date for the XPS RTX Spark Desktop. Neither has any other OEM in the wave. NVIDIA and its partners have cited fall 2026 as the arrival window.