The release of the cross-platform performance test ChimbaBenchCL 1.2 has taken place. It is designed for testing the performance of graphics cards and processors using the OpenCL API. The project is written in C and is distributed under the MIT license. Pre-compiled executable files are available for 64-bit (x86_64) operating systems including Linux (Glibc 2.7+), FreeBSD, and Windows 7+. A 32-bit (x86) version is available for Windows XP+ and Linux (Glibc 2.7+).
ChimbaBenchCL requires a device that supports the OpenCL 1.0 API or higher and has at least 512 MB of video memory (GeForce 8000+, Radeon HD 4000+, Intel Gen7+). Testing one device takes on average 20-30 seconds. Launch parameters 2-disable-gpu*2 and 2-disable-cpu2 are provided to exclude problematic or malfunctioning hardware. A full list of options is available with the command 2—help2.
The set of core tests in version 1.2 includes:
- PCIe Bandwidth — a test of PCI Express bus bandwidth.
- VRAM Bandwidth — a test of memory bandwidth.
- VRAM Latency — a test of random access memory latency.
- FP16-H8 — a test for computations using half8 type.
- FP32-F8 — a test for computations using float8 type.
- FP32-FMA and FP32-MAD — single precision tests using FMA and MAD instructions.
- FP64-D4 — a test for computations using double4 type.
- INT8-U16, INT32-I4, INT64-L2 — tests for computations with various integer types.
- Image Blur — a simulation of graphics processing in real tasks (Box Blur, 1920×1080).
Compared to other OpenCL tests such as ProjectPhysX OpenCL-Benchmark, Intel Compute Benchmarks, Spector, Sycl-bench, Mixbench, and LuxMark, the ChimbaBenchCL project is positioned as a lightweight alternative free from EULA restrictions and does not impose burdens on the test results obtained.
Source: opennet.ru
