RTX Spark is the two-word name Acer wants you to remember from IFA 2026. At its next@Acer press conference in Berlin on September 2, 2026, the company showed off the Acer SFF RTX Spark, a small-form-factor desktop design built around NVIDIA’s RTX Spark superchip. The pitch is bold: up to one petaflop of AI compute, up to 128 GB of unified memory, and the ability to run enormous AI models entirely on a desk, with no cloud subscription in sight.

The RTX Spark reveal was a design showcase rather than a product launch. Acer has not announced pricing or a ship date, and the company says availability of its RTX Spark devices will be confirmed later. Yet the machine on the Berlin show floor — a vertical shell with vented panels, a metallic accent stripe and a fold-out stand — tells us a great deal about where Acer believes desktop computing is heading.

This article walks through everything Acer and NVIDIA have disclosed about the RTX Spark design: the silicon inside it, the local agentic AI workloads it targets, how it compares with the Veriton RI110 mini workstation Acer announced alongside it, and what the wider RTX Spark ecosystem means for creators, developers, gamers and businesses planning their next hardware cycle.

What Acer Showed at IFA 2026: The SFF RTX Spark

acer rtx spark ai desktop future b berlin tv tower sphere spire

Acer used IFA 2026 to stake out a position in a category that barely existed a year ago: the compact personal AI supercomputer. The SFF RTX Spark is Acer’s interpretation of NVIDIA’s reference platform, and the company presented it as a machine for “personal AI agents” that combine content creation, development and high-performance gaming in one small chassis.

The reveal at next@Acer

The announcement came during Acer’s global press conference on the opening day of IFA 2026 in Berlin, Europe’s biggest consumer electronics show and the stage where PC makers traditionally set their agenda for the year ahead. Acer positioned the RTX Spark design as a full-stack AI platform, leaning on NVIDIA’s complete suite of RTX technologies rather than assembling parts from multiple vendors. The message was consistent across the briefing: local execution, personal agents, and data privacy through on-device processing.

A design built to be seen

Unlike most mini PCs, which hide under monitors, the Acer SFF RTX Spark is styled to sit in view. Yanko Design described a vertical shell with vented panels and a metallic accent stripe, plus a fold-out stand that props the unit upright on a desk. The RTX Spark aesthetic borrows more from premium consoles than from corporate towers, which fits the audience Acer named on stage: creators, AI developers and gamers.

A concept with a confirmed platform

It is worth being precise about status. The RTX Spark unit shown in Berlin is a design study on a confirmed NVIDIA platform, not a finished SKU. Acer says availability will be announced at a future date, and no price was given. The silicon platform underneath it, however, is real, already validated by NVIDIA, and shipping in other vendors’ systems from fall 2026.

RTX Spark Specifications: Grace, Blackwell and 128 GB of Unified Memory

acer rtx spark ai desktop future c artist palette thumb hole

The heart of every RTX Spark system is NVIDIA’s superchip: an Arm-based Grace CPU and a Blackwell-generation RTX GPU sharing one pool of memory. Acer’s SFF configuration reaches the top of the platform’s range, and the numbers are striking for a machine this small.

The superchip inside

The Acer SFF RTX Spark pairs up to a 20-core NVIDIA Grace CPU with a Blackwell RTX GPU carrying up to 6,144 CUDA cores. Grace is the Arm processor NVIDIA co-designed with MediaTek — the same partnership we examined in Nvidia’s $3.5B MediaTek bet — and its job here is to feed the GPU without the bandwidth bottlenecks of a conventional CPU-plus-card design.

One petaflop on a desk

NVIDIA rates the platform at up to 1 petaflop of AI compute. That is 1,000 teraflops of low-precision throughput aimed squarely at model inference, dwarfing what any consumer CPU delivers for this class of work. The headline capability that follows: the RTX Spark platform can execute models with up to 120 billion parameters locally, with context windows reaching one million tokens.

The DGX Spark lineage

The recipe will look familiar to anyone tracking NVIDIA’s workstation line. The company already sells the DGX Spark, a developer-focused mini AI machine built on the same formula: a 20-core Grace CPU, a Blackwell GPU with 6,144 CUDA cores, 128 GB of unified memory and a petaflop of compute. What Acer showed in Berlin looks like that platform’s consumer-facing evolution — the same silicon, restyled for the desk rather than the lab, and opened up to the industry’s biggest system builders. That lineage matters for buyers, because it means the architecture is not an unproven first attempt; it has been running real developer workloads since the DGX version arrived.

Why unified memory matters

The up-to-128 GB unified memory pool is arguably the most important RTX Spark specification. Discrete graphics cards strand models in limited VRAM; unified memory lets the CPU and GPU address one large pool, which is what makes 120-billion-parameter models feasible on a desktop. Here is the Acer SFF RTX Spark at a glance:

ComponentAcer SFF RTX Spark (as disclosed)
CPUUp to 20-core NVIDIA Grace (Arm, co-designed with MediaTek)
GPUNVIDIA Blackwell RTX, up to 6,144 CUDA cores
MemoryUp to 128 GB unified
AI throughputUp to 1 petaflop
Model capacityUp to 120B parameters, up to 1M token context
Form factorSmall-form-factor vertical desktop, fold-out stand
Price and availabilityNot announced; to be confirmed by Acer

Why the RTX Spark Desktop Targets Local Agentic AI

acer rtx spark ai desktop future d heat sink block row of fins

NVIDIA CEO Jensen Huang framed the category bluntly: “The PC is being reinvented.” In his telling, the next computer is not a device you launch applications on, but one where agents work on your behalf. The RTX Spark desktop is the hardware expression of that idea — enough compute and memory to keep the agents, and your data, at home.

Personal agents without the cloud meter

An agent that reads your files, drafts your work and monitors your projects has to run somewhere. Doing it in the cloud means paying per token and shipping private data to someone else’s servers. A local RTX Spark box flips that equation: buy the compute once, then run autonomous AI agents around the clock at no marginal cost. We saw the same logic in Perplexity’s Nvidia-based portable computer, which promised a fully local agent with zero token costs.

Privacy as a hardware feature

Acer’s briefing repeatedly stressed data privacy through local processing. For professionals handling client files, contracts or unreleased creative work, an on-device agent that never transmits source material is a materially different proposition from a cloud assistant. The RTX Spark design turns a compliance argument into a product feature, and that argument lands well beyond the enthusiast crowd.

The million-token difference

Context length is where local hardware gets interesting. A model holding up to one million tokens of context can ingest an entire codebase, a season of scripts or years of correspondence in one pass. On the RTX Spark platform, workloads that lean on natural language processing — summarising archives, querying long documents, reviewing repositories — stay on the desk instead of streaming through an API, which changes both the economics and the risk profile.

RTX Spark vs Veriton RI110: Acer's Two Mini AI Desktops

acer rtx spark ai desktop future e bullet train rounded nose

The RTX Spark concept did not travel to Berlin alone. Acer also announced the Veriton RI110 AI mini workstation, a palm-sized Intel-based machine that is much closer to shipping. Reading the two together reveals Acer’s strategy: one attainable mini workstation now, one aspirational petaflop desktop next.

Two machines, one thesis

Both machines make the same bet — that meaningful AI work is moving on-device — but they serve different budgets and timelines. The Veriton RI110 runs an Intel Core Ultra X7 358H with Intel Arc B390 graphics in a 5.45 x 5.17 x 2.05-inch chassis weighing about 1.39 lbs. The SFF RTX Spark chases a far higher ceiling with NVIDIA silicon. Acer’s comparison in one view:

FactorSFF RTX SparkVeriton RI110
SiliconNVIDIA Grace CPU + Blackwell RTX GPUIntel Core Ultra X7 358H + Arc B390
MemoryUp to 128 GB unifiedUp to 96 GB
StorageNot disclosedUp to 4 TB SSD
Model capacityUp to 120B parametersUp to 120B parameters
ExpansionNot disclosedOCuLink (PCIe 4.0 x4, up to 64 Gbps)
Bundled agentNVIDIA AI software stackAcer Qubi Claw, sandboxed local assistant
AvailabilityTo be announcedNorth America Q4 2026, EMEA Q1 2027

The memory gap in one picture

Memory capacity decides which models fit, so the 32 GB gap between the two machines matters more than any clock speed. The RI110’s 96 GB is three quarters of the RTX Spark design’s 128 GB ceiling:

Maximum memory: Acer’s two IFA 2026 mini desktops
SFF RTX Spark, unified 128 GB
Veriton RI110 96 GB

What the RI110 tells us about the RTX Spark roadmap

The RI110 ships with Qubi Claw, Acer’s agentic assistant that installs in one click and runs locally in an isolated sandbox, handling research, travel planning, transcription and translation. That software investment only makes sense as groundwork. When Acer’s RTX Spark hardware arrives, the agent layer will already be field-tested on thousands of RI110 desks.

What Creators, Developers and Gamers Get From RTX Spark

acer rtx spark ai desktop future f bowling pin and ball

NVIDIA and Acer are unusually specific about who the RTX Spark platform serves, and the published workload claims read like three product pitches stapled together. Each audience gets a concrete, measurable capability rather than a vague promise of acceleration.

Creators: heavyweight scenes and 12K video

The platform claims fill a creative professional’s wish list: rendering 3D scenes up to 90 GB in size and editing 12K 4:2:2 video on the desktop. Adobe, Blackmagic Design and CapCut are among more than 100 developers already optimising software for RTX Spark systems, which suggests the creative toolchain will be ready before Acer’s own hardware is.

Developers: frontier-scale models at the desk

For AI developers, the RTX Spark proposition is straightforward: prototype against a 120-billion-parameter model locally, iterate without per-call costs, and keep proprietary training data in the building. The 6,144-core Blackwell GPU also accelerates fine-tuning and evaluation runs that would otherwise queue for shared cloud capacity.

Gamers: 1440p with headroom

The gaming claim is the most conventional: AAA titles at 1440p and beyond, at over 100 fps with ray tracing enabled. That matters strategically, because it makes the RTX Spark desktop a defensible purchase for a household that games in the evening and runs agents by day. The disclosed claims by workload:

WorkloadPublished RTX Spark claim
3D renderingScenes up to 90 GB
Video editing12K 4:2:2 footage
Model inferenceUp to 120B parameters, 1M token context
Gaming1440p+, 100+ fps with ray tracing

The RTX Spark Ecosystem: Vendors, Software and Availability

Acer is joining a platform, not inventing one, and the shape of that platform explains the company’s confidence. NVIDIA has lined up the industry’s biggest system builders behind RTX Spark, with the first wave of devices arriving in fall 2026.

The fall 2026 hardware wave

ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI are all slated to ship RTX Spark systems in fall 2026. Acer’s own timing is the outlier — its availability will be announced separately — which is precisely why Berlin got a design showcase rather than a price tag. The vendor line-up so far:

VendorRTX Spark device status
ASUS, Dell, HP, Lenovo, Microsoft Surface, MSIShipping from fall 2026
AcerDesign shown at IFA 2026; availability to be announced

Software gravity

More than 100 developers optimising for the platform is the number that should reassure buyers. Hardware categories live or die on software support, and RTX Spark is launching with Adobe and Blackmagic Design already on board. NVIDIA’s full RTX technology stack — the same libraries that dominate professional graphics — comes with it.

The contrast with earlier attempts at AI-first PCs is instructive. NPU-equipped laptops arrived before the software that could use them, and buyers noticed. Here the sequence is reversed: the toolchain is being ported first, on hardware that developers already own in DGX form, and the consumer boxes follow into an ecosystem that is warmed up. For a purchase this new, that ordering removes the biggest historical reason to wait for a second generation.

Where Acer can still differentiate

Every vendor gets the same superchip, so differentiation moves to thermals, acoustics, industrial design and bundled software. Acer’s vented vertical shell and its Qubi Claw agent are early answers. A late arrival with a distinctive design may serve Acer better than a rushed box that disappears into a wall of identical fall 2026 launches.

What the RTX Spark Design Means for Businesses

Strip away the show-floor gloss and the RTX Spark announcement is a procurement signal: the price of serious on-premises AI capability is about to fall through the floor of what a rack used to cost, and it will arrive in a chassis that fits on a desk.

A new tier between laptop and server

Until now, organisations wanting local AI had two options: underpowered laptops with NPUs, or servers with enterprise GPUs and enterprise invoices. An RTX Spark class desktop creates a middle tier — petaflop compute per seat — that changes the maths for teams in data center operations, software development and creative production alike. Small firms that could never justify a GPU cluster can now consider one desk-side unit per team.

Questions to ask before you buy

The sensible response is preparation, not a pre-order. Inventory which of your workloads currently pay cloud inference bills; those are your candidates for local execution. Check whether your software vendors are among the 100-plus optimising for the platform. And weigh the RI110-style interim machines against waiting for Acer’s full RTX Spark hardware — the answer depends on whether 96 GB covers the models you actually run.

Budgeting for the petaflop desk

No price has been announced, but the budgeting exercise can start now. List what your team spends monthly on hosted inference, transcription, rendering farms and image generation, then annualise it. If that figure approaches the cost of a high-end workstation, a desk-side unit pays for itself inside its warranty period — and the hardware keeps working after the subscription would have kept billing.

Factor in the quieter costs too: data-handling reviews get simpler when material never leaves the building, and teams stop rationing experiments because every prompt has a price on it. The point of the exercise is not to pre-order anything; it is to know your number before seven vendors start competing for it.

The competitive read

For NVIDIA, spreading RTX Spark across seven system vendors builds a moat around local AI before rivals can respond. For Acer, showing a design first and shipping later is a calculated trade: it cedes the first wave to competitors but buys time to differentiate. For buyers, competition among seven vendors on one platform usually means one thing — prices will move in your favour within two quarters of launch.

RTX Spark FAQ

Is the Acer SFF RTX Spark on sale?

No. The RTX Spark design was shown at IFA 2026 as a concept on NVIDIA’s confirmed platform. Acer says device availability will be announced at a future date, with no pricing disclosed yet.

What silicon powers the RTX Spark design?

An NVIDIA superchip combining up to a 20-core Grace CPU with a Blackwell RTX GPU of up to 6,144 CUDA cores, sharing up to 128 GB of unified memory and delivering up to 1 petaflop of AI compute.

Can an RTX Spark machine really replace cloud AI?

For many inference workloads, yes: it runs models up to 120 billion parameters with a million-token context locally. Training frontier models and serving thousands of concurrent users still belong in the cloud.

How does it differ from the Veriton RI110?

The Veriton RI110 is Acer’s shipping mini workstation: Intel silicon, up to 96 GB of memory, OCuLink expansion and the Qubi Claw local assistant, arriving in North America in Q4 2026. The SFF concept is the bigger swing — NVIDIA silicon, 128 GB of unified memory and petaflop-class compute, on a timeline Acer has not yet committed to.

When do the first RTX Spark systems ship?

ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI are expected to ship systems in fall 2026. Acer’s timeline and pricing are still to be announced.

References