Describes OpenCL property CL_DEVICE_MAX_WORK_ITEM_SIZES and if newer Intel graphics products have better performance for parallel computing.
- Does Intel® Iris® Xe Graphics have more work-items or threads than Intel® UHD Graphics?
- Looking for the best Intel® Graphics to perform parallel computations.
- Typing the command "clinfo" in Linux with OpenCL:
- Intel Iris Xe Graphics has 96 Execution Units (EUs) or threads in total, four times more than the Intel UHD Graphics with 24.
- The max work item sizes from the image above is a value of a property called CL_DEVICE_MAX_WORK_ITEM_SIZES from the OpenCL 2.1 specifications:
Maximum number of work-items that can be specified in each dimension of the work-group to clEnqueueNDRangeKernel.
Returns n size_t entries, where n is the value returned by the query for CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS.
The minimum value is (1, 1, 1) for devices that are not of type CL_DEVICE_TYPE_CUSTOM. - This does not mean that the machine can have 256*256*256 = 16777216 work-items in a workgroup and can execute 16777216 threads in parallel.
- The CL_DEVICE_MAX_COMPUTE_UNITS property indicates that the machine has 24 Execution Units (EUs) in total.
- Each thread is capable to execute in either SIMD8, SIMD16, or SIMD32 lanes, SIMD depends on kernel and how it was built by the compiler.
- Each single EU has 7 hardware (HW) threads (for 6th Generation Intel graphics).
- The theoretical max number of OpenCL threads for this platform is 24*7*32 = 5376.
- It will be handled by 168 HW threads and given that any particular kernel is built using SIMD32 mode.
The number of EUs for a specific product can be found on our graphics specifications page.
For example, 96 EU is seen in the Processor Graphics section for i7-11370H processor.