CUDA-Z is a free diagnostic tool for macOS that reads and displays detailed hardware and performance information from NVIDIA CUDA-capable graphics cards and general-purpose GPU computing devices.
What is CUDA-Z?
CUDA-Z is a lightweight GPU inspection utility that surfaces the raw technical specifications of NVIDIA CUDA hardware — everything from compute capability and driver version to memory bandwidth and core counts — in a single, readable window. Think of it as a GPU-flavoured System Information panel, but one that speaks fluent CUDA rather than generic PCIe nomenclature.
If you've ever tried to verify whether a GPU actually supports a specific CUDA compute capability before kicking off a training run or a compute-heavy simulation, you know how frustrating it is to chase that information across driver release notes and spec sheets. CUDA-Z surfaces it instantly, without you having to write a single line of CUDA C.
What does CUDA-Z do best?
CUDA-Z excels at giving GPU-compute practitioners a fast, authoritative read on their hardware before any serious workload begins. The tool reports the device's compute capability version, total and free VRAM, memory bus width, theoretical memory bandwidth, number of CUDA cores, warp size, maximum thread dimensions, and the active driver and runtime library versions — all without requiring you to fire up Xcode, the CUDA toolkit, or any third-party benchmarking suite.
I keep it around specifically for those moments when I spin up a new machine or swap GPUs and need to confirm the environment before pointing PyTorch or a compiled CUDA binary at it. Where GPU-Z (Windows-only) and nvtop (terminal, Linux-first) each cover part of this ground, CUDA-Z is one of the few tools that targets the CUDA abstraction layer directly on macOS — useful for the narrowing window of Mac hardware that still pairs with an eGPU running an NVIDIA card.
- Instant compute-capability version readout (e.g. 8.9 for Ada Lovelace)
- VRAM total, free, and reserved breakdown
- Driver and CUDA runtime version side-by-side
- Theoretical peak bandwidth and core/clock figures
- Multi-GPU systems: each device listed and individually inspectable
Who should use CUDA-Z?
CUDA-Z is squarely aimed at GPU-compute practitioners: ML engineers validating a new workstation, researchers confirming eGPU capability before a training job, and systems programmers debugging CUDA compatibility mismatches. It is decidedly not a consumer tool — if you're looking to benchmark frame rates or monitor GPU thermals while gaming, reach for GPU Monitor Pro or iStatistica instead.
Data scientists running local inference who need to know whether their card can handle a specific precision mode, or developers chasing down a "no CUDA-capable device detected" error in a PyTorch stack trace, will find CUDA-Z the fastest path to a definitive answer. The interface asks nothing of you and gives back exactly the spec sheet you need.
Is CUDA-Z free?
Yes — CUDA-Z is free and open-source, distributed under an open licence and hosted on SourceForge. There are no in-app purchases, no nag screens, and no account required. Installation via Homebrew Cask takes a single command.
The trade-off for zero cost is that development cadence is slow; this is a community-maintained project rather than a commercially backed product. The core functionality remains solid for supported hardware, but don't expect rapid updates for brand-new GPU architectures on day one.
What are the best CUDA-Z alternatives?
For macOS, the honest answer is that direct alternatives are thin. nvtop gives you live utilisation graphs in the terminal but requires a Linux host or Rosetta workarounds. GPU Benchmark — Video Card Test from the Mac App Store covers gaming-oriented metrics but ignores CUDA compute specifics entirely. On Windows, GPU-Z by TechPowerUp is the gold standard and far more frequently updated. If your primary need is live utilisation monitoring rather than static spec enumeration, Activity Monitor's GPU History panel (macOS 10.15+) or iStatistica Pro serve the everyday use case without touching CUDA at all.
For developers who need programmatic access rather than a GUI, running nvidia-smi from the CUDA toolkit or querying cudaGetDeviceProperties() directly gives even richer data — but CUDA-Z remains the fastest no-code path to the same information.
How does CUDA-Z compare to GPU-Z?
GPU-Z is Windows-only, actively updated, covers AMD and Intel GPUs in addition to NVIDIA, and includes a real-time sensor panel with fan speed, temperature, and power draw. CUDA-Z is macOS and Linux compatible, focuses exclusively on CUDA-capable NVIDIA devices, and reports compute-layer specifics (compute capability, warp size, thread limits) that GPU-Z surfaces only partially. They complement rather than replace each other — if you run a dual-boot setup, GPU-Z owns the Windows side; CUDA-Z handles the Mac.
