NVIDIA has released CUDA Toolkit 13.4 as a developer preview, marking the first time the company has shipped an official CUDA development package that natively targets Windows Arm64 machines. This move closes a critical gap that made Windows-on-Arm essentially invisible to the AI and scientific computing ecosystem, since every major AI framework and tool that relies on CUDA had no practical way to run on Windows Arm devices until now.

CUDA is NVIDIA's proprietary parallel computing platform, and it powers the vast majority of the world's AI development and inference work. By extending CUDA to Windows Arm64, NVIDIA is not just filling a technical hole; it is extending its software advantage into the one remaining Windows device category where it had no presence. RTX Spark laptops and mini PCs will be the only Windows-on-Arm machines where the full production AI software stack runs without workarounds or emulation layers.

Why Does CUDA on Windows Arm Matter?

Windows on Arm has struggled for years because each hardware launch arrived before the software ecosystem was ready. Developers faced missing tools, incompatible libraries, and frustrating workarounds. NVIDIA is deliberately reversing that sequence with RTX Spark. By shipping the CUDA toolkit and developer driver before retail hardware hits the market, the company is giving developers months to prepare their applications. This means when RTX Spark laptops arrive in fall 2026, users will find a mature software ecosystem waiting for them, not a barren landscape of compatibility issues.

The timing is strategic. CUDA is the exclusive computing substrate for most of the world's AI work. AMD ROCm and Intel OneAPI offer alternatives, but the overwhelming majority of AI frameworks default to CUDA. By establishing CUDA as the native development platform for Windows Arm before competitors can respond, NVIDIA is cementing its position in a new device category.

What Tools Are Available for Developers Right Now?

CUDA Toolkit 13.4 gives developers multiple paths to build Windows Arm64 applications immediately:

• Native Arm64 Build: Compile CUDA applications directly on a Windows Arm64 machine using NVCC, NVIDIA's LLVM-based C/C++ compiler, producing code that runs without any emulation layer.
• Cross-Compilation: Build Windows Arm64 CUDA binaries from a conventional x86-64 Windows machine using the existing toolkit, allowing developers to work from their current desktop while generating output for the new architecture.
• Hybrid Arm64EC Mode: Use Arm64EC (Emulation Compatible) to run native Arm64 code and emulated x64 code inside the same process, letting developers convert applications piece by piece and ship hybrid builds that run CUDA-accelerated paths natively while keeping legacy x64 components running through Windows' Prism emulation layer.

The toolkit includes the full core library set available on other platforms: NVCC 13.4.46, Nsight Compute, Nsight Systems, cuBLAS 13.7.0, cuFFT, cuRAND, cuSOLVER, cuSPARSE, and NPP. Developers do not need RTX Spark hardware to begin porting; any existing Windows-on-Arm development system, such as a Surface Pro X or Snapdragon X Elite laptop, can be used today to start the process.

What Is RTX Spark, and Why Is It Significant?

RTX Spark is a two-chiplet system-on-chip fabricated on TSMC's 3-nanometer process. A Blackwell GPU die carries 6,144 CUDA cores across 48 streaming multiprocessors, the same shader count as a desktop GeForce RTX 5070, alongside fifth-generation Tensor Cores supporting FP4 (4-bit floating point) precision. A MediaTek-designed CPU die houses 20 Arm cores: ten Cortex-X925 performance cores and ten Cortex-A725 efficiency cores, a configuration drawn from high-end mobile silicon and scaled for PC workloads.

The two dies connect via NVIDIA's NVLink-C2C chip-to-chip interconnect, running at 600 GB/s of bidirectional bandwidth between CPU and GPU, approximately five times the bidirectional bandwidth of PCIe Gen 5. This coherent interconnect allows the CPU and GPU to share the same memory address space directly, which is the architectural underpinning of unified memory design.

The unified memory pool reaches up to 128 GB of LPDDR5X, with the memory subsystem delivering approximately 300 GB/s of bandwidth to the chip. For large language model inference, which is fundamentally a memory-bandwidth-limited workload, the 300 GB/s figure governs token generation speed. NVIDIA rates the platform at 1 petaFLOP of AI compute, but that figure is measured at FP4 precision with sparsity enabled, representing a best-case theoretical ceiling rather than a sustained general-compute figure.

NVIDIA claims the 128 GB memory pool is sufficient to run local AI reasoning models with up to 120 billion parameters without relying on cloud inference, a figure specific to the laptop SKU. The DGX Spark desktop module, which uses the same GB10 die, targets 200 billion parameter models at its full 128 GB configuration.

When Will RTX Spark Hardware Arrive, and Who Is Building It?

RTX Spark systems are expected to ship in fall 2026 across premium laptops and compact desktop mini PCs, with confirmed OEM partners including ASUS, Dell, HP, Lenovo, and MSI. Microsoft has also built kernel-level Windows 11 optimizations specifically for RTX Spark, changes that were not extended to Qualcomm's Snapdragon platforms. Microsoft is co-developing OpenShell, a pre-1.0 agentic AI runtime that provides kernel-level sandboxing, a YAML policy engine, and a Privacy Router for masking personally identifiable information before routing to cloud models.

What Do Developers Need to Install to Get Started?

CUDA Toolkit 13.4 marks a structural change in how NVIDIA ships its developer tooling. Starting with this version, the GPU driver is no longer bundled with the CUDA Toolkit on Linux; it ships separately as NVIDIA's standalone Developer Driver (R616 and above). On Windows, this decoupling started earlier with CUDA 13.1, but the Linux change is new in 13.4.

The practical implication is significant: developers cannot simply run the standard CUDA Toolkit installer and expect a working GPU environment. They must separately download and install an R616-series Developer Driver (version 616.00 or later) to access CUDA 13.4 features and the newly enabled Windows Arm64 platform. The preview driver, version 616.00, is already available through the NVIDIA Developer portal and is initially targeted at Microsoft Surface RTX Dev Box hardware.

Existing CUDA 13.x applications will continue to run on any driver version at or above 580 through CUDA minor-version compatibility, but new platform-specific features require the R616 series.

How to Prepare Your AI Application for Windows Arm64

• Download CUDA Toolkit 13.4: Access the developer preview from NVIDIA's official portal and install the toolkit on your current development machine or a Windows Arm64 device if you have one available.
• Install the R616 Developer Driver: Separately download and install Developer Driver version 616.00 or later from the NVIDIA Developer portal to enable full CUDA 13.4 functionality and Windows Arm64 support.
• Choose Your Build Path: Decide whether to use native Arm64 compilation, cross-compilation from x86-64, or hybrid Arm64EC mode based on your codebase structure and development workflow preferences.
• Test on Available Hardware: If you have access to a Surface Pro X, Snapdragon X Elite laptop, or other Windows Arm64 device, begin porting and testing your CUDA applications immediately without waiting for RTX Spark hardware to arrive.

The release of CUDA Toolkit 13.4 represents a deliberate strategic move by NVIDIA to establish software dominance in the Windows Arm64 space before the RTX Spark hardware ecosystem matures. By compressing the typical platform launch sequence and providing developer tools months before retail hardware arrives, NVIDIA is ensuring that when RTX Spark laptops and mini PCs reach consumers in fall 2026, they will have a mature, ready-to-use AI software ecosystem waiting for them.