I am getting this repost, from opencl driver selction.
platforms found:
opencl platform: AMD Accelerated Parallel Processing
opencl platform: Intel(R) OpenCL
opencl platform: Intel(R) OpenCL
selecting:
opencl device name: gfx1101
opencl device version: OpenCL 2.0 AMD-APP (3652.0)
opencl device compute units: 30
opencl device local memory: 65536
but the spec say the GPU has twece as many CU?
I have search for answers, wit not avail, the only thing I can get is these moronic
AI generated report from Google,
In summary, while the AMD RX 7800 XT has 60 CUs, if you are seeing fewer CUs reported in OpenCL, it’s likely a software or driver-related issue rather than a fundamental limitation of the GPU hardware .
does any body has that issue, or know how to fix it?
and yes I have all the latest driver and opencl SDKs,
Sorry, I posted this in the wrong thread.
However, I did find the answer to the problem. It turns out that AMD, likely for marketing reasons, reports the number of compute units based on wavefronts (the SIMD-width processors).
But this is misleading because two wavefronts in a processor aren’t independent, they operate as a single unit.
In reality, a compute unit on AMD hardware is 128-wide in single precision, not 64. So while the driver correctly reports 30 compute units, the product box advertises 60 to make it appear comparable to something like an NVIDIA 4070.
Even if that count were technically accurate, there’s no practical way for developers to leverage it, unless you’re using wider workgroups, which isn’t always feasible.
Anyway, at least that mystery is cleared up.
Feel free to close this question.
thanks.