
A new physics benchmark, Warp Core PhysX Bench, has been released with CUDA support and Vulkan - GameGPU
An enthusiast using the pseudonym Warp9-Systems has released a beta version of the standalone physics benchmark Warp Core PhysX Bench v0.99r.
The program includes three main benchmark modes. The first utilizes exclusively CPU resources to calculate collisions in two cameras with 5,000 objects on eight threads in runs of 5, 10, or 25 cycles.
The second mode utilizes NVIDIA CUDA technology to simulate 150,000 cubes in five standard passes of 240 steps each, and the third utilizes the Vulkan API for interaction of 8,192 spheres at maximum frame rate.
In addition to standard scenarios, the benchmark includes an experimental section with extreme stress benchmarks, Neurotix Mode, which simultaneously stresses the CPU, RAM, and GPU.
Performance results
The final performance score in the application is measured in computational steps per second, provided that the variance between runs does not exceed 5%.
In the published benchmark results, the NVIDIA GeForce RTX 5070 Ti graphics card in CUDA mode, while processing 150,000 solids, achieved a peak value of 7,603,105 contact points and 1,872,700 contact surfaces, resulting in a final result of 7,4536 steps per second.
Compatibility and development
At the current stage of development, hardware acceleration GPU PhysX in the utility supports NVIDIA graphics cards based on the Blackwell, Pascal, and Maxwell architectures, with the list of compatible devices expanding in future updates.
The application was developed and debugged on the AM5 platform with a processor. AMD Ryzen 7 9800X3D and graphics accelerators GeForce RTX 5070 Ti and Radeon RX 9070 GRE.
The tool has also already undergone preliminary testing on models GeForce RTX 3070, GTX 1080 Ti, Quadro M6000 and flagship solutions RTX 5080 and RTX 5090, the results of which are published on specialized forums.
