The recommended system specification for running SVL Simulator (locally) are as follows:
- at least 4 GHz Quad core CPU
- NVIDIA GTX 1080 (8GB memory) or higher
- Windows 10 (64-bit), Ubuntu 18.04 (64-bit), or Ubuntu 20.04 (64-bit)
While lower-specification hardware may be able to run SVL Simulator, there may be issues with the performance required to interface properly with a user’s System Under Test.
SVL Simulator is currently supported for Windows (64-bit) and Linux (64-bit). For optimal performance, Windows is recommended.
Currently, the full functionality of SVL Simulator in Developer Mode (in Unity Editor) is supported on Windows only. End-to-end automatic simulations using PythonAPI Runtime template or Visual Scenario Runtime template are supported on Linux only.
On Windows, SVL Simulator requires a graphics card that supports DirectX 11. On Linux, SVL Simulator requires a graphics card that supports Vulkan 1.1.
If running Apollo or Autoware on the same system as the Simulator, it is recommended that the GPU have at least 10GB of memory.
If running Apollo or Autoware on a different system as the Simulator, a gigabit connection between the systems is required (a gigabit switch is sufficient, gigabit internet is not required).
System and graphics performance varies tremendously from lightweight gaming laptops to high end graphics workstations. There are many different hardware factors that can influence overall performance including CPU model, clock frequency, number of cores, system RAM, GPU model and available GPU memory. In addition, the Simulator configuration (environment, vehicle, number and type of sensors) can affect performance, as well as the autonomous software configuration (e.g. enabled modules).
Keep an eye on CPU load, system memory, and GPU memory. Running out of any of these can cause a variety of problems such as low frame rates or autonomous software modules being unable to function. The minimum goal for real time simulation should be 15fps (or the desired sensor frame rate) since that will avoid dropping frames for critical sensors like camera and LiDAR sensors. For complex multi-sensor simulations, refer to the Distributed Simulation docs for information on multi-machine (multi-GPU) distributed simulation.
While it is possible to run sophisticated autonomous software stacks like Apollo or Autoware on the same machine that runs SVL Simulator, it will challenge even the highest performing systems. For best results, the Simulator should be run on a separate machine from the autonomous software.
If it is not practical or possible to run autonomous software on a separate machine then you might want to consider using ground truth sensors in place of perception and traffic signal modules, as documented in Modular Testing. This will greatly reduce the CPU and GPU requirements of an autonomous stack such as Apollo and increase the likelihood that you can run it on the same machine with SVL Simulator.