NVIDIA RTX Spark specs look impressive because they combine several ideas normally discussed separately: a Blackwell RTX GPU, an Arm CPU, unified memory, local AI performance, RTX graphics, and Windows platform work from Microsoft. The important part is understanding what each spec means and what it does not prove yet.
Direct answer
The key RTX Spark specs are up to 6,144 Blackwell RTX cores, up to 20 CPU cores, up to 1 petaflop of FP4 AI performance, and up to 128 GB of unified memory. NVIDIA presents RTX Spark as a superchip for slim Windows laptops and small desktop PCs, with CUDA, RTX, DLSS, TensorRT, NVIDIA Studio tools, Reflex, and G-SYNC support.
These are platform-level claims. A final device may ship with different power limits, cooling, displays, storage, and memory configurations. Specs are useful for understanding the direction of the platform, but they are not the same as real-world reviews.
Why people search it
People search for RTX Spark specs because the platform sounds like a new category of Windows PC. Searchers want to know whether this is a GPU, a CPU, a full system-on-chip, a laptop platform, or a compact desktop platform. They also want to know how it compares with Apple Silicon, DGX Spark, and existing RTX laptops.
The confusion is reasonable. RTX Spark is not a normal graphics card announcement. It is closer to a complete platform story for AI, graphics, memory, and Windows optimization.
RTX Spark specs table
| Spec area | What NVIDIA and Microsoft have described |
|---|---|
| GPU | Up to 6,144 Blackwell RTX cores |
| CPU | Up to 20 power-efficient Arm CPU cores |
| AI performance | Up to 1 petaflop FP4 AI performance |
| Memory | Up to 128 GB unified memory |
| Chip link | NVIDIA describes Blackwell RTX GPU and Grace CPU connected through NVLink-C2C |
| Software | CUDA, TensorRT, RTX, DLSS, NVIDIA Studio, Reflex, G-SYNC |
| Device types | Slim Windows laptops and small desktop PCs |
| First systems | Fall 2026 window described by NVIDIA and Microsoft |
For the larger context, read What Is NVIDIA RTX Spark?.
Blackwell RTX GPU
The Blackwell RTX GPU is central to RTX Spark’s pitch. NVIDIA says RTX Spark brings RTX graphics, ray tracing, DLSS, Tensor Cores, and local AI acceleration into a slim PC platform. That makes the GPU relevant for creators, gamers, and AI developers.
For creators, the GPU can matter for rendering, video workflows, AI-assisted editing, streaming tools, and 3D applications. For developers, the GPU can matter for local inference, model prototyping, TensorRT, and CUDA workflows. For gamers, the GPU can matter for ray tracing, DLSS, Reflex, and Windows PC game support.
20-core Arm CPU
RTX Spark also includes an Arm CPU. This is important because Windows on Arm is part of the platform story. Microsoft says it has worked on scheduling, thermal, unified memory, and Prism emulation optimizations for RTX Spark.
The CPU spec alone does not tell you whether every app will feel fast. Native Arm apps, emulated x86 apps, background tasks, cooling, and software optimization all matter. That is why final reviews should test both native and emulated workloads.
Unified memory
Unified memory means CPU and GPU workloads can access a shared memory pool. NVIDIA and Microsoft describe RTX Spark systems with up to 128 GB of unified memory. That can be useful for local AI models, creative projects, and workloads that need large memory access.
But unified memory is not automatically faster for every task. Memory bandwidth, latency, operating system behavior, app optimization, and model size all matter. Treat the 128 GB figure as an important capacity signal, not a universal performance guarantee.
FP4 AI performance
NVIDIA describes RTX Spark as reaching up to 1 petaflop of FP4 AI performance. FP4 is relevant for lower-precision AI workloads, especially when models and frameworks are built to use it well.
The key phrase is “when supported.” AI performance depends on model format, quantization, framework support, drivers, memory capacity, and software maturity. A headline number can be meaningful, but it should not be used alone to decide whether a system is right for a specific model.
What specs cannot prove
RTX Spark specs do not prove final price, battery life, fan noise, sustained performance, gaming frame rates, compatibility, or real creator workflow speed. They also do not prove that every AI framework will be ready on day one.
Before using RTX Spark specs to compare devices, verify:
- The exact laptop or desktop configuration.
- Price and availability in the buyer’s country.
- Whether the app is native Arm, emulated, or GPU accelerated.
- Independent battery life and performance tests.
- Driver and framework support for the workload.
For buying and comparison discipline, use our guide to how to compare AI tools before turning specs into recommendations.
Sources to check
Primary sources include the NVIDIA RTX Spark product page, the NVIDIA Computex 2026 RTX announcement, and the Microsoft Windows Blog announcement.
FAQ
How much memory does RTX Spark support?
NVIDIA and Microsoft describe RTX Spark systems with up to 128 GB of unified memory. Final configurations may vary by device.
Is RTX Spark an Arm chip?
Yes, the platform uses Arm CPU cores and is tied to Microsoft’s Windows on Arm work.
What does 1 petaflop FP4 mean?
It is a low-precision AI performance metric. It can matter for supported AI workloads, but it does not replace real model and app testing.
Does RTX Spark support CUDA?
NVIDIA says CUDA runs natively on RTX Spark. Check framework and app support before relying on a specific workflow.
Are RTX Spark specs enough to compare laptops?
No. You also need power limits, cooling design, display, storage, memory configuration, price, and independent testing.
Does RTX Spark replace regular RTX laptops?
Not necessarily. RTX Spark is a new platform for certain Windows laptops and small desktops. Traditional RTX laptops will still matter for many buyers.