에이서의 컴팩트 폼 팩터 비교 결과, 인텔 기반 Veriton RI110 모델만 올해 출시되며 RTX Spark 컨셉 롤아웃은 지연된다.

Acer arrived at IFA 2026 in Berlin with two compact AI desktops and a clear directive regarding availability: the Veriton RI110 AI Mini Workstation ships in North America in Q4 2026 on confirmed Intel specifications, while the Acer SFF RTX Spark—the petaflop-class NVIDIA concept that dominated stage attention—remains a design showcase. The Spark carries no announced price, no confirmed ship date, and an underlying AI performance claim that independent reviewers have yet to validate on actual hardware. Both systems were unveiled during Acer’s next@Acer global press conference on September 2, 2026, as part of the company’s largest IFA presentation in its 50-year history. IFA 2026 opens to the public in Berlin on September 4.

This distinction is critical for buyers evaluating compact AI workstations in the current purchase window. One machine is production-ready; the other presents a compelling architectural argument for the trajectory of desktop computing and underscores the importance of monitoring NVIDIA’s fall 2026 retail launch before committing capital. The two systems do not compete directly; they target different buyer timelines and budget ceilings. Acer positioned the Veriton RI110 as the immediately available solution for enterprise and small-to-midsize business customers unwilling to wait for NVIDIA silicon, while the SFF RTX Spark serves as Acer’s strategic design statement for when that silicon reaches retail.

The Veriton RI110 is the concrete product. It houses an Intel Core Ultra X7 processor 358H and Intel Arc B390 graphics within a chassis measuring 138.5 × 131.3 × 52.1 millimeters (5.45 × 5.17 × 2.05 inches) and weighing approximately 0.63 kilograms (1.39 pounds)—compact enough to mount behind a monitor. The system supports up to 96 GB of LPDDR5X memory and up to 4 TB of PCIe Gen 4 SSD storage. Acer states it can handle local inference, design work, and content creation using AI models with up to 120 billion parameters. An OCuLink port delivering up to 64 gigabits per second provides a post-purchase external GPU expansion path, confirming that the RI110 functions as an upgradeable platform rather than a fixed-spec device. It ships with Windows 11 Pro and includes Acer Qubi Claw, an agentic AI assistant that executes local workflows within an isolated sandbox to keep data off external servers.

The SFF RTX Spark remains a concept. Its vertical enclosure, mounted on a fold-out stand, leans aesthetically toward high-end audio equipment rather than traditional workstation design. Inside sits NVIDIA’s RTX Spark Superchip: a configuration featuring up to a 20-core NVIDIA Grace ARM CPU paired with up to a 6,144-core Blackwell RTX GPU and up to 128 GB of unified LPDDR5X memory in a single coherent pool. Acer also confirmed an OCuLink port on the Spark design, offering up to 64 Gbps for external GPU or storage expansion. During the press conference, Acer targeted creators, AI developers, and gamers, though pricing and availability remain unannounced pending a future confirmation.

The Spark’s architectural promise hinges on NVIDIA’s NVLink-C2C chip-to-chip interconnect, a technology originally engineered for data center superchips. Understanding its function is essential to evaluating the 1-petaflop headline. In conventional desktop workstations, the CPU communicates with system RAM via one bus while the GPU accesses dedicated video memory through another. When AI workloads like large language models exceed the GPU’s VRAM capacity, the system must page data between pools, creating bandwidth bottlenecks and latency penalties. This limitation explains why desktop RTX 5090 cards with 32 GB of VRAM cannot run 70-billion-parameter models without quantization: the model simply exceeds physical memory constraints.

The RTX Spark Superchip eliminates that boundary. The Grace CPU and Blackwell GPU reside in a single package—a 2.5D chiplet design fabricated on TSMC’s 3-nanometer process—and connect via NVLink-C2C. Both processors share a single 128 GB unified pool as their native memory, with LPDDR5X delivering approximately 300 gigabytes per second of memory bandwidth. This is roughly five times the bandwidth of PCIe Gen 4 and meaningfully faster than PCIe Gen 5, which peaks at approximately 128 GB/s for an x16 lane. While the internal chip-to-chip bandwidth runs higher, it remains substantially below the 900 GB/s NVLink-C2C found in NVIDIA’s data center Grace-Hopper systems.

Practically, NVIDIA states the RTX Spark can run 120B-parameter models with up to 1 million tokens of context entirely in local memory—a workload that would otherwise require a $4,699 NVIDIA DGX Spark or a dedicated data center GPU under conventional architecture. The Spark’s Blackwell GPU utilizes NVIDIA’s NVFP4 format, a 4-bit floating-point precision scheme that generates the 1-petaflop figure. At standard FP32 precision, the same silicon delivers approximately 31 teraflops, a competitive metric for a compact desktop but a necessary calibration for buyers comparing this number against competitors citing higher-precision outputs.

Before the Spark’s AI performance claim influences purchasing decisions, a critical data point must be noted: as of today, NVIDIA’s 1-petaflop CUDA AI inference capability has not been independently validated on any shipping RTX Spark hardware. In July 2026, a tech reviewer who obtained a pre-production Surface Laptop Ultra equipped with the top-tier RTX Spark N1X configuration published benchmark data spanning more than a month of hands-on testing. The most relevant finding for AI buyers was that CUDA-based workloads failed to complete on both tested driver versions—the original 591.33 drivers shipped with the engineering sample and the newer 616.00 developer preview supporting CUDA Toolkit 13.4. The reviewer was forced to revert to CPU and Vulkan Compute paths, which offer substantially slower performance. Tom’s Hardware, TechSpot, VideoCardz, and Notebookcheck independently corroborated these results.

This does not indicate the hardware cannot deliver on its AI promises at retail; driver development for new platform architectures is inherently iterative, and engineering samples run pre-production software explicitly not intended to reflect final performance. Rather, it means the 1-petaflop figure remains an architectural target backed by NVIDIA demonstrations, not yet a verifiable metric for independent buyers. NVIDIA has released CUDA 13.4 for Windows ARM64 as a Developer Preview, marking tangible progress in the software stack, though practical validation on the platform remains pending. For buyers evaluating the SFF RTX Spark primarily for local AI workloads—the core use case Acer is marketing—the responsible approach is to trust the architecture while verifying performance claims with independent benchmarks once retail units arrive.

Timing context further matters for purchase planning. NVIDIA confirmed at SIGGRAPH 2026 that ASUS and MSI will lead the initial fall 2026 launch of RTX Spark devices, with Acer and GIGABYTE following in a second wave. The first wave also encompasses Dell, HP, Lenovo, and Microsoft Surface. Across both waves, NVIDIA’s original Computex announcement outlined more than 30 laptop designs and approximately 10 compact desktops. What Acer’s IFA 2026 showcase established is its design direction for both phases: a vertical SFF chassis with a fold-out stand for the RTX Spark desktop, and the Intel-based Veriton RI110 as the immediate offering for enterprise and SMB buyers unable to wait for NVIDIA silicon.

Morgan Stanley analysts placed N1X-class configurations at approximately $2,899, with lower N1 configurations near $1,799, though NVIDIA has not officially announced pricing for any RTX Spark device. The Veriton RI110’s pricing remains unannounced, but its Intel Core Ultra positioning suggests it will fall meaningfully below the RTX Spark’s estimated entry point.

Finally, the SFF RTX Spark runs Windows 11 on ARM, making software compatibility a specific consideration before assuming gaming or legacy application support. Acer and NVIDIA have confirmed native support for a specific set of titles: Fortnite, Valorant, League of Legends, and PUBG—all ported to ARM64. The platform also supports Adobe, Blender, CapCut, and ComfyUI for creative workflows. These represent confirmed-working applications, not representative of the full legacy software ecosystem.