ChimbaBenchCL

This OpenCL benchmark allows you to test and compare different graphics cards. Thanks to the availability of pre-compiled executable files, the benchmark can be easily transferred and used on different computers with different operating systems. Test results are automatically saved to a file located in the same directory as the program.

Test suite (v1.1): VRAM Bandwidth – memory bandwidth. VRAM Latency – memory random-access latency. FP16-H8 – half-precision (half8). FP32-F8 – single-precision (float8). FP32-FMA – single-precision using FMA (fused multiply-add) instructions. FP32-MAD – single-precision using MAD (multiply-add) instructions. FP64-D4 – double-precision (double4). INT8-U16 – low-precision integer test (uchar16). INT32-I4 – standard-precision integer test (int4). INT64-L2 – high-precision integer test (long2). Image Blur – simulation of graphics processing in real-world tasks (1920x1080, Box Blur).

A typical benchmark run for a single device takes about 30 seconds (v1.1). If the test takes significantly longer, you might be using an Intel UHD 600 GPU on Linux. In such cases, you can disable the problematic device using commands like -disable-gpu-intel (or similar). You can view a list of available commands by running the program with the -help or --help argument.