# System requirements (/docs/getting-started/requirements) ## The short version [#the-short-version] * **Windows 10/11 only.** There is no macOS, Linux, or Android version today. * **Local generation needs an NVIDIA GPU, RTX 20-series or newer.** AMD and Intel GPUs are not supported for local generation (the engine is CUDA-based). * **Cloud generation works on any PC.** It runs on SimpliGen's GPUs, so your graphics card doesn't matter. If your GPU is below the requirements, Cloud is the intended path. Every cloud-enabled preset runs on our GPUs with credits, and API presets (Google, OpenAI) run on the provider's servers. Only local generation needs an NVIDIA card. ## What your GPU can run locally [#what-your-gpu-can-run-locally] | Your NVIDIA GPU | Images | Video | | --------------- | ----------------------- | --------------------------------------------------------------- | | 4-6 GB VRAM | Most image presets work | Not realistic; use Cloud | | 8 GB VRAM | Everything | Lightweight / low-VRAM video presets only, at lower resolutions | | 12 GB VRAM | Everything | Most video presets at moderate settings | | 16 GB+ VRAM | Everything | Everything, including the heavier video models | Also plan for: * **RAM:** 16 GB minimum, 32 GB recommended for video. Video models spill into system RAM heavily; more RAM genuinely helps. * **Disk:** models are large (single video models can be 10-30 GB). Set your storage location in **Settings -> Storage** before installing big presets. * **Windows page file:** heavy video generation can need a large page file. If generations crash near the end, see [Out of memory](/docs/troubleshooting/out-of-memory). Many presets ship several **variants** (for example fp8 and GGUF builds of the same model). The app picks a variant that fits your GPU, and the preset card lets you choose; a smaller variant trades some quality for fitting into less VRAM. ## How long does a generation take? [#how-long-does-a-generation-take] Local generation takes minutes, not seconds. That is normal. Rough expectations: * **Images:** seconds to a couple of minutes on most supported cards. * **Video on a 16 GB+ desktop card:** a few minutes for a 5-second 480p clip; high resolutions take much longer. * **Video on an 8 GB card (especially laptops):** 30-60 minutes for a 5-second clip on a lightweight preset is within normal range. Start at 480p and 5 seconds to validate a prompt, then scale up. Every resolution tier and extra second multiplies VRAM use and time. Cloud generation is much faster because it runs on datacenter GPUs. ## Cards and setups that do NOT work locally [#cards-and-setups-that-do-not-work-locally] * **AMD and Intel GPUs**: not supported. On the wishlist, but with no timeline. * **NVIDIA GTX 10-series and older** (GTX 1080, 1060, 1050 Ti...): the PyTorch version the engine uses no longer supports these architectures. You may see `CUDA error: no kernel image is available for execution on the device`. * **Workstation cards from the same era** (Quadro M6000, P5000...): same architecture limitation. * **Macs**: the installer is a Windows `.exe`; it will not run on macOS. * **Phones / tablets**: no. If the installer or engine reports **UNSUPPORTED\_HARDWARE**, your GPU falls into one of the groups above. ## Virtual machines and rented GPUs [#virtual-machines-and-rented-gpus] Running SimpliGen inside a VM or on a virtualized GPU (Azure, some rented cloud desktops) is hit-or-miss: virtual GPUs often lack memory-management features the engine uses. If the engine crashes on a vGPU, enable **Virtual GPU compatibility** in Settings. Renting a GPU server (RunPod and similar) to run SimpliGen remotely is not a supported setup; use built-in Cloud generation instead. ## Multiple GPUs [#multiple-gpus] SimpliGen uses one GPU. If you have several, the app picks the most capable one. Multi-GPU pooling is not supported.