When you use neural analytics (see General information on Neural analytics), the hardware platform must meet the following requirements:

1. Due to the features of NVIDIA SDK, the neural analytics can operate only on Windows Server 2019 OS and Windows 10 OS and newer.

2. The neural network analytics supports the following devices: CPU, GPU NVIDIA, VPU (Intel NCS).
   

   To connect the Intel NCS, plug in the device to the USB port and make sure that Windows detects it as a USB device with one of the following names: Movidius, Myriad X, VSC Loopback Device. You can use Intel NCS on any computer that meets the hardware requirements of Axxon PSIM (see General requirements for PCs). We don't recommend using more than one Intel NCS device on the same server.

3. If you use CPU or Intel GPU for the neural analytics operation, then meet the following requirements:
   1. Supports the following CPUs:
      1. 6th–10th Generation Intel® Core™ processors;
         

         If you use a CPU below 6th Generation Intel® Core™ processors, the operation of detectors isn't guaranteed.

      2. Intel® Xeon® v5 family;
      3. Intel® Xeon® v6 family;
      4. Intel® Movidius™ Neural Compute Stick;
      5. Intel® Neural Compute Stick 2;
      6. Intel® Vision Accelerator Design with Intel® Movidius™ VPUs.
   2. The OpenVino toolkit must support the used Intel CPU (see https://software.intel.com/content/www/us/en/develop/tools/openvino-toolkit/system-requirements.html).
   3. A processor must support the AVX2 or AVX512 instruction set (see https://ark.intel.com/content/www/us/en/ark/search/featurefilter.html?productType=873&1_Filter-InstructionSetExtensions=3533&1_Filter-Family=595).
      

      If you use a processor with Intel AMX instructions along with Windows Server 2019 OS on the server, we recommend updating the OS to Windows Server 2022 for successful operation of neural analytics.

4. NVIDIA GeForce 1050 Ti GPU or newer. Requirements:
   
   1. At least 2 GB of memory.
   2. CUDA 11.1 − 11.4.

   3. Compute Capability 3.5 − 8.6.

      CUDA is compatible with generations of computers on architectures: Kepler (partially), Maxwell, Pascal, Volta, Turing, Ampere (partially), see CUDA. When you use an NVIDIA GPU, we recommend installing the latest driver from the NVIDIA official website. You can check the version of GPU Compute Capability on the manufacturer's website.

When you use a GPU, a single neural network consumes 500 MB of video memory. For example, to run any number of fire and smoke detection channels based on neural networks, you must use a graphics card with 1GB of memory or more. You can use multiple video cards in your system.