NVIDIA’s RTX Spark PCs went up for pre-order this month, and the pitch is bold: one slim Windows machine that plays modern games and runs a local AI model at the same time, built around the Grace Blackwell GB10 superchip. Wccftech confirmed the pre-order launch on October 7, 2026, with shipping set for October 16. The marketing language is “the best of gaming and AI.” The spec sheet tells a more complicated story, and the gap between NVIDIA’s advertised 100 frames-per-second claim and what a dedicated gaming desktop with a discrete RTX 5090 actually delivers is the part nobody selling these machines wants to dwell on.
That gap matters because RTX Spark is not priced like a budget machine. NVIDIA’s launch release states that RTX Spark laptops start at $2,599, in line with pricing reported by PC Gamer (a later official NVIDIA blog post says $2,399), putting it well above most gaming laptops with a discrete RTX 5070 or 5080 and squarely in enthusiast desktop territory. Buyers weighing a Spark machine against a dedicated gaming rig or a Mac Studio deserve a clearer picture of what the Grace Blackwell architecture actually buys them in a game, not just in a chatbot.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What RTX Spark Actually Is
RTX Spark is NVIDIA’s name for a new class of Windows machines built on the Grace Blackwell GB10 superchip, the same silicon family behind the enterprise-focused DGX Spark. Where DGX Spark runs NVIDIA’s own DGX OS and targets AI developers exclusively, RTX Spark runs standard Windows 11 and is explicitly marketed for gaming, content creation and everyday computing alongside AI workloads, according to The Register. NVIDIA’s own product page lists the top configuration with a 6,144-core Blackwell RTX GPU, up to 1 petaflop of FP4 AI compute and as much as 128GB of unified LPDDR5X memory shared between CPU and GPU.
Eight OEMs are attached to the rollout, a lineup this site covered in detail when retail availability was confirmed: Acer, ASUS, Dell, Gigabyte, HP, Lenovo, MSI and Microsoft Surface. PCGuide reports pre-orders are live now across six of those brands, with units shipping October 16. ASUS told Wccftech that its initial RTX Spark allocation had already been claimed by channel partners before the public pre-order window even opened, a sign of real demand even if it says nothing about how the machines perform once a demanding game is actually running.
The Two Chip Tiers: 5,120 Cores vs. 6,144 Cores
NVIDIA is shipping RTX Spark in two silicon configurations, and the difference is bigger than a simple step up. The entry tier pairs an 18-core Arm CPU with a 5,120-core Blackwell RTX GPU, according to spec breakdowns published by PC Gamer. The flagship tier, often referred to by the N1X designation, pairs a 20-core Arm CPU (10 Cortex-X925 performance cores plus 10 Cortex-A725 efficiency cores) with a 6,144-core Blackwell RTX GPU. Unified memory scales from 24GB on the cheapest builds up to 128GB on the flagship, all soldered LPDDR5X with no user upgrade path, PC Gamer notes.
Pricing follows the same split. ASUS lists its ProArt P14 laptop, built on the lower tier, at $2,599. The ProArt P16, built on the higher tier, lists at $2,799. NVIDIA’s launch release states that RTX Spark laptops start at $2,599. A Morgan Stanley estimate placed the high-end N1X configuration closer to $2,899 in some builds, with entry systems starting around $1,799, though that figure is described as an analyst estimate rather than a confirmed retail price across every OEM.
The 100 FPS Claim, and Why It Needs an Asterisk
NVIDIA’s own marketing for RTX Spark claims roughly 100 frames per second at 1440p resolution in games, using a combination of ray tracing and DLSS. That number is doing a lot of work in NVIDIA’s pitch that this machine is a legitimate gaming PC and not just a repackaged AI developer box. The catch: NVIDIA has not published which games, which DLSS mode, which ray-tracing settings or whether frame generation was active when that figure was measured. No independent reviewer has yet run RTX Spark through a controlled benchmark suite against titles like Cyberpunk 2077, Alan Wake 2 or Black Myth: Wukong.
That absence matters more for RTX Spark than it would for a conventional graphics card launch. DLSS and frame generation can inflate a displayed frame-rate number well beyond native rendering performance, and NVIDIA has leaned on exactly that technique in recent GPU launches. A 100 fps figure achieved with DLSS 4 Performance mode and Multi Frame Generation looks very different from 100 fps rendered natively. Until a third party runs the same test suite it runs on every discrete GPU, that number should be read as a platform target, not a verified result.
RTX Spark vs. a Desktop RTX 5090: Core Count Isn’t the Whole Story
The 6,144-core figure on the flagship RTX Spark chip sounds substantial until it’s placed next to a desktop GPU. A discrete RTX 5090, which this site has tracked through resale price swings running as high as 2.4x MSRP amid ongoing GPU memory shortages (NVIDIA executive Jeff Fisher reportedly said the card has not been discontinued and remains in production, while acknowledging constrained supply), carries a vastly larger die, far higher memory bandwidth, and a power budget many multiples above anything RTX Spark’s chassis can sustain. CUDA core count alone is not a performance ranking between architectures and power classes; it is one input among clock speed, memory bandwidth and sustained thermal headroom, all of which favor a full-size desktop tower.
RTX Spark’s real advantage isn’t raw graphics throughput. It’s the unified memory pool. A desktop gaming PC with an RTX 5090 ships with 32GB of dedicated GDDR7 VRAM on the card, a figure this site examined in earlier RTX 5090 rendering benchmarks. RTX Spark’s flagship configuration offers up to 128GB that both the CPU and GPU can draw from, which is a meaningfully different proposition for anyone trying to keep a large local AI model loaded while also running a game. It is not, on current evidence, a path to beating a discrete flagship GPU at native rendering.
RTX Spark Configuration and Pricing at a Glance
| Tier | CPU | GPU | Unified Memory | Starting Price | Example Model |
|---|---|---|---|---|---|
| Entry | 18-core Arm | 5,120-core Blackwell RTX | 24GB-64GB LPDDR5X | ~$2,599 | ASUS ProArt P14 |
| Flagship (N1X) | 20-core Arm (10 Cortex-X925 + 10 Cortex-A725) | 6,144-core Blackwell RTX | Up to 128GB LPDDR5X | ~$2,799-$2,899 | ASUS ProArt P16 |
| AI compute (both tiers) | – | Up to 1 petaflop FP4 | – | – | – |
| Shipping window | – | – | – | – | October 16, 2026 |
| OEM count at launch | – | – | – | – | 6 (ASUS, Dell, HP, Lenovo, Microsoft, MSI) |
Power and Thermals: What a Slim Chassis Can Actually Sustain
RTX Spark’s gaming case rests on a thermal premise that hasn’t been independently stress-tested yet. The GB10 chip at the heart of the platform carries a 140-watt TDP, with systems drawing power through an external supply rated up to 240 watts, according to a detailed review of the related DGX Spark hardware published by gpusmith.com. That’s a fraction of the power budget a desktop tower running a discrete RTX 5090 can draw, and it directly shapes how long RTX Spark can hold peak clocks once a game has been running for more than a few minutes.
An external PSU and 240-watt ceiling do give RTX Spark more thermal headroom than a typical thin laptop with its battery and power supply crammed into one chassis, which is part of why NVIDIA and its partners describe the design as favorable compared with ultra-compact machines like a Mac mini. But nobody has yet published sustained-load numbers showing what happens to RTX Spark’s GPU clocks after 30 or 60 minutes of a demanding game, nor how much those clocks drop once a local AI model is also resident in memory and drawing on the same shared power and thermal budget. That is the single most consequential test still missing from the public record.
The AI Half of the Pitch: 128GB and 120-Billion-Parameter Models
NVIDIA’s AI case for RTX Spark is more straightforward than its gaming case. With up to 128GB of unified memory, the company says RTX Spark can run local language models with as many as 120 billion parameters and context windows up to 1 million tokens, numbers pulled directly from NVIDIA’s own RTX Spark product listing. That capacity outstrips what almost any gaming laptop with a discrete mobile GPU can hold in VRAM today, since most mobile RTX cards top out well under 24GB of dedicated graphics memory.
The platform supports fifth-generation Tensor Cores, DLSS, NVIDIA Reflex, the ACE framework for AI-driven NPC behavior, G-SYNC, CUDA and the NVFP4 data format, according to NVIDIA’s own listing. That’s a genuinely unusual combination for a Windows consumer PC: the full NVIDIA graphics stack sitting next to enough shared memory to hold a mid-size local LLM, rather than forcing a buyer to choose between a gaming GPU and an AI workstation.
The Hybrid Use Case Nobody Has Tested Yet
The most interesting claim buried inside NVIDIA’s RTX Spark pitch is also the least examined: running a game and a local AI workload at the same time, on the same shared memory pool, without one starving the other. NVIDIA’s own ACE framework is built around exactly this idea, using on-device models to drive NPC dialogue or behavior while a game renders in real time. In theory, a Spark machine could keep a coding assistant or a translation model loaded in the background while a game runs in the foreground, something a conventional gaming laptop with 16GB of VRAM physically cannot do without swapping memory constantly.
Whether that holds up once both workloads are competing for the same 140-watt power budget is an open question. Unified memory removes the hard VRAM ceiling that limits a discrete gaming GPU, but it does not remove the compute or power ceiling. A 6,144-core GPU splitting cycles between ray-traced rendering and a resident language model has to split its thermal and power budget too, and that tradeoff is exactly the kind of benchmark that hasn’t been run publicly yet.
How RTX Spark Stacks Up Against the Alternatives
| System | GPU Cores / Class | Memory | Power Envelope | Primary Strength | Entry Price |
|---|---|---|---|---|---|
| RTX Spark (flagship) | 6,144-core Blackwell RTX | Up to 128GB unified LPDDR5X | 140W chip TDP, 240W external PSU | Large local AI models + Windows gaming in one chassis | ~$2,799 |
| Desktop RTX 5090 | Full discrete Blackwell die | 32GB dedicated GDDR7 | Desktop-class, 500W+ system power | Native rendering and ray-tracing throughput | Reported resale up to 2.4x MSRP |
| Mac Studio (Apple silicon) | Apple GPU cores, unified memory architecture | Configurable high-capacity unified memory | Desktop-class, efficient per-watt | Local AI inference and creative workloads, no native AAA gaming library | Starts well above RTX Spark entry tier |
| DGX Spark (enterprise) | Same GB10-family silicon as RTX Spark | Up to 128GB unified memory | 140W chip TDP, 240W PSU | AI development and training, DGX OS, no gaming focus | $4,999 per prior site coverage |
Historical Context: From DGX Spark to a Windows Gaming Pitch
RTX Spark did not appear out of nowhere. NVIDIA first built the GB10 superchip for DGX Spark, a desktop AI developer box running the company’s own DGX OS rather than Windows. This site’s earlier comparison of DGX Spark against Framework Desktop and Mac Studio found that machine positioned purely as a local AI development tool, with no gaming ambitions attached. The Register reported in early October that NVIDIA had also debuted a $4,999 DGX Spark variant with half the RAM and storage of the original, a move tied to the broader memory-chip supply crunch squeezing the entire industry this year.
RTX Spark represents NVIDIA taking that same GB10 silicon and repositioning it for a much larger Windows consumer audience, rather than the narrow slice of AI developers who would buy a DGX-branded machine. Earlier coverage on this site of the initial pre-order wave tracked pricing across six OEMs before the broader eight-brand rollout now underway. The strategic logic is straightforward: a Windows machine that can plausibly double as a gaming PC reaches a far bigger buyer pool than an AI-only workstation ever could, even if the gaming performance claims haven’t been independently verified.
Market Impact: A New Category Squeezed Between Two Markets
RTX Spark enters a market where buyers already have two well-understood options: a gaming laptop with a discrete RTX GPU for pure frame-rate performance, or a compact AI workstation like a Mac Studio for local model work. RTX Spark is pitched as replacing both, at a price that sits above most gaming laptops and below a maxed-out AI workstation. Dell’s XPS 16 RTX Spark configuration, covered on this site at launch, illustrates the positioning: an 18-core, 99Wh creator-oriented machine priced at $3,799, aimed squarely at buyers who’d otherwise split their budget across two separate machines.
The risk for NVIDIA and its OEM partners is that RTX Spark ends up satisfying neither audience fully. A dedicated gamer who wants maximum frame rates at 1440p or 4K still has every reason to buy a desktop with a discrete RTX 5090 instead, where the thermal and power budget simply isn’t constrained by a laptop chassis. An AI developer who needs sustained, uninterrupted inference throughput may still prefer a workstation built without gaming compromises baked into its cooling design. RTX Spark’s success depends on proving there’s a large enough buyer segment that genuinely wants both in one box, rather than two specialists each doing their job better alone.
What to Watch Before Buying
- Independent frame-rate testing in actual AAA titles, with DLSS mode and ray-tracing settings disclosed
- Sustained clock-speed measurements after 30-plus minutes of continuous gaming load
- Combined gaming-plus-local-AI benchmarks showing how much frame rate drops when a model is resident in memory
- Battery life on laptop configurations when gaming on battery power versus plugged into the external PSU
- Software and driver maturity for Arm-based Windows, including anti-cheat and launcher compatibility
Five Predictions for RTX Spark’s First Six Months
Independent gaming benchmarks will undercut the 100 fps claim in native rendering. Expect reviewers to show that figure is only reachable with DLSS Performance mode and frame generation active, not native 1440p rendering, mirroring the pattern seen with other recent DLSS-reliant marketing claims.
Thermal throttling will show up in extended gaming sessions. A 140-watt chip TDP in a slim chassis is unlikely to hold peak clocks much past 20-30 minutes under sustained ray-traced load, based on the power envelope gpusmith.com documented for the related DGX Spark hardware.
The hybrid gaming-plus-AI use case will prove niche but genuinely differentiated. Few buyers will run a 70B-parameter model while gaming, but the subset who do (streamers using local AI tools, developers testing game-integrated AI features) will find no real alternative to RTX Spark’s unified memory approach.
Pricing will compress within two quarters. With six OEM partners shipping similar GB10-based hardware, competition on the entry 18-core tier is likely to push starting prices below the current roughly $2,599 mark by early 2027, following the pattern of past NVIDIA reference-design launches.
A second-generation chip will target the gaming gap directly. Expect NVIDIA’s next GB-family superchip to push higher sustained clocks and a larger power envelope specifically to close the credibility gap between RTX Spark’s AI performance (already strong) and its gaming performance (currently unverified).
Frequently Asked Questions
Can RTX Spark actually run modern AAA games?
Yes, in the sense that it runs Windows 11 with full NVIDIA RTX graphics support, DLSS and ray-tracing hardware. Whether it runs them well compared with a discrete gaming GPU is unverified, since no independent reviewer has yet published a controlled AAA benchmark suite for RTX Spark.
Is RTX Spark faster than a desktop RTX 5090 for gaming?
Not on current evidence. RTX Spark’s 6,144-core flagship GPU and 140-watt power envelope are not comparable to a full discrete RTX 5090 with vastly higher memory bandwidth and a desktop-class power budget. RTX Spark’s advantage is its large shared memory pool for AI, not raw gaming throughput.
When does RTX Spark ship?
Pre-orders opened in early October 2026 across multiple OEMs, with the first units shipping October 16, 2026, according to PCGuide and Wccftech.
How much RAM does RTX Spark have?
Configurations range from 24GB up to 128GB of unified LPDDR5X memory shared between the CPU and GPU, depending on the chip tier and OEM model.
What’s the difference between RTX Spark and DGX Spark?
Both use the same GB10-family Grace Blackwell silicon, but DGX Spark runs NVIDIA’s own DGX OS and targets AI developers exclusively. RTX Spark runs standard Windows 11 and is marketed for gaming, content creation and general use alongside AI workloads.
Which OEMs are selling RTX Spark PCs?
Eight brands are attached to the launch: Acer, ASUS, Dell, Gigabyte, HP, Lenovo, MSI and Microsoft Surface, with pre-orders live across several of them already.
Does RTX Spark support DLSS?
Yes. NVIDIA’s RTX Spark product listing confirms support for DLSS 5 (including Super Resolution, Ray Reconstruction and Dynamic Multi Frame Generation), fifth-generation Tensor Cores, NVIDIA Reflex and hardware ray-tracing cores, the same core graphics technology stack used across its discrete RTX lineup.
