LogoBottleneckCalcs
Back to Blog
CPUGPUWindows on ARMNvidiaHardware

RTX Spark Launch Guide: Leaked Benchmarks

Adam Roy
Adam Roy
September 18, 2026
12 min read
RTX Spark Launch Guide: Leaked Benchmarks

Nvidia's RTX Spark platform, the chip most people still call N1X, has moved from spec sheet to shipping product. On September 3, Nvidia confirmed an October 2026 launch window and released the final configuration details. A week later, the first leaked benchmarks from a prototype device gave us the first real look at how it performs, and the picture is more complicated than Nvidia's marketing slides suggest.

If you read our Nvidia N1X (RTX Spark) Complete Guide back in June, this is the update: what actually shipped, what it costs, and what the leaked numbers mean if you're deciding whether to wait for one.


The Two Configurations, Officially Confirmed

Nvidia is launching RTX Spark in two tiers, both pairing an Arm-based Grace CPU with a Blackwell GPU on a single die:

  • RTX Spark (flagship): 20-core Grace CPU, 6,144-core Blackwell GPU, 24GB to 128GB of unified LPDDR5X memory. Built for laptops and compact desktops.
  • RTX Spark (base): 18-core Grace CPU, 5,120-core Blackwell GPU, 24GB or 32GB of unified memory. Laptop only at launch.

Both share a 300GB/s memory bus, USB4/Thunderbolt support, and a TDP that scales up to 80W depending on the chassis. The unified memory design means CPU and GPU share one pool instead of separate system RAM and VRAM, which is the same architecture Nvidia used in the DGX Spark AI workstation released last year.

SpecificationRTX Spark (Base)RTX Spark (Flagship)
CPU Architecture18-core Arm Grace CPU20-core Arm Grace CPU
GPU Architecture5,120-core Blackwell GPU6,144-core Blackwell GPU
Unified Memory24GB or 32GB LPDDR5X24GB to 128GB LPDDR5X
Memory Bandwidth300 GB/s unified pool300 GB/s unified pool
Platform TDPScales up to 80WScales up to 80W
ConnectivityUSB4 / ThunderboltUSB4 / Thunderbolt
Form FactorsLaptops only at launchLaptops & compact desktops

Who's Actually Shipping Devices, and for How Much

Eight manufacturers have confirmed RTX Spark systems: Acer, Asus, Dell, Gigabyte, HP, Lenovo, Microsoft, and MSI. The most detailed listings so far come from Lenovo:

  • Lenovo Yoga Pro 9n (15.3-inch): Up to 128GB of memory, 80W thermal limit, 92.5Wh battery
  • Lenovo Yoga 9n 2-in-1 (16-inch): Up to 64GB, 70W thermal limit, 99.9Wh battery
  • Microsoft Surface Laptop Ultra: Confirmed, but full specs not yet public
  • Acer compact desktop: Confirmed, details pending

Thin silver RTX Spark laptop prototype with lid open

Acer and Gigabyte have committed to the platform but haven't announced specific models or dates yet, so expect those to trail the October wave.

No manufacturer has published official pricing. The closest thing to a real number comes from Morgan Stanley analyst estimates, which put base N1 systems around $1,799 and flagship N1X configurations near $2,899. Treat that as informed speculation, not a price you can hold Nvidia to.


What the First Leaked Benchmarks Actually Show

This is the part the June guide couldn't cover, because the hardware didn't exist yet. An engineering prototype of the Surface Laptop Ultra, running the 20-core chip with 24GB of memory, produced these Cinebench numbers:

RTX Spark engineering prototype board on a testing bench

Benchmark TestSingle-Core ScoreMulti-Core Score
Cinebench 20241231,386
Cinebench 20265405,771

Single-core performance reads as competitive. Multi-core is the weaker story: for a 20-core design, reviewers running the leak expected a bigger jump, and the numbers suggest the CPU isn't scaling with core count the way a mature high-end laptop chip normally would.

The Thermal Picture & Early Driver State

Macro shot of laptop cooling vent under sustained thermal load

The thermal picture adds context:

  • The prototype's platform power caps at 80W sustained (95W burst), but Nvidia's own Max-Q profile restricts the GPU to roughly 50W of that budget.
  • CPU temperatures reportedly hit 98–100°C under sustained load, and pushing power past 44W produced almost no extra clock speed — a sign the chip is running into a thermal wall rather than a power wall.
  • Game testing on the prototype showed periodic stuttering, and some CUDA workloads reportedly failed to complete.
  • Firmware and drivers were still early (an internal 616.00 developer build), so some of this should improve before retail units ship. Not all of it will.

Separately, comparisons against Apple's M5 Max, the chip Nvidia is most directly positioning RTX Spark against, have shown RTX Spark trailing in the benchmarks run so far.


About That "100fps at 1440p" Claim

Nvidia's September marketing materials advertise 100fps at 1440p with ray tracing enabled, paired with DLSS 5. That figure leans heavily on DLSS 5's frame generation, which Nvidia has said can multiply frames up to 6x.

It's not a native rendering number, and FP4 precision (which DLSS 5 uses) is more prone to rounding artifacts than the FP16/FP32 math used in older comparisons. Take the headline number as a best-case marketing figure until independent reviewers can test it directly, which won't happen until devices ship in October.


The Windows on Arm Catch

RTX Spark runs Windows on Arm. Games and apps built for x86 run through Microsoft's Prism emulation layer.

Nvidia says the performance cost should be minor in GPU-bound titles, but CPU-intensive games — which is most simulation, strategy, and open-world titles — could see a larger hit. This matters more for RTX Spark than it did for past Arm laptop attempts, because Nvidia is explicitly marketing this as a gaming-capable machine, not just a productivity chip.


What This Means if You're Comparing It Against a Normal Build

RTX Spark doesn't fit our calculator's usual CPU-plus-discrete-GPU model, because the CPU and GPU are one fused chip with shared memory rather than two separate parts you'd pick and pair. If you're weighing an RTX Spark laptop against a traditional gaming laptop or a desktop with a discrete RTX 5070-class card, the honest comparison right now is:

  • Buy RTX Spark for: Local AI workloads, unified-memory-heavy creative apps (Blender, Adobe, ComfyUI all have day-one support), and thin-and-light portability.
  • Stick with a traditional CPU/GPU build for: Competitive gaming, CPU-heavy titles, and anything where you need proven multi-core performance today rather than a promise that drivers will mature.

📌 Bottleneck Calculator Database Note
We'll add RTX Spark configurations to the calculator's database once retail units ship and independent reviewers publish full benchmark suites. Leaked prototype numbers aren't reliable enough to build accurate bottleneck scores around yet.


Frequently Asked Questions

When does RTX Spark actually ship?

October 2026, per Nvidia's own September 3 announcement. No exact date has been given.

How much will it cost?

Nvidia hasn't announced official pricing. Analyst estimates (Morgan Stanley) suggest roughly $1,799 for base configurations and $2,899 for flagship models, but treat this as unconfirmed.

Is RTX Spark good for gaming?

Unclear until retail units are tested. Nvidia's 100fps/1440p claim depends heavily on DLSS 5 frame generation, and the only leaked benchmarks so far show weaker multi-core CPU scaling and gaming stutter on early firmware.

Does RTX Spark replace a discrete GPU?

No. It's a fused CPU/GPU chip with shared memory, aimed at thin laptops and compact desktops running local AI workloads, not a discrete graphics card replacement for a gaming desktop.

What's the difference between RTX Spark and DGX Spark?

DGX Spark, released in 2025, is a $3,999 AI workstation running Linux. RTX Spark is the consumer laptop and compact-desktop version, running Windows, aimed at a lower price point with gaming support layered on top.

Adam Roy
Written By

Adam Roy

Founder of BottleneckCalcs and hardware performance expert. Adam specializes in PC architecture, web optimization, and creating data-driven tools for the hardware community.

View Full Profile

Optimizing your build?

Use our Bottleneck Calculator to ensure your CPU and GPU are perfectly matched for your next upgrade.

Check Your Bottleneck