OPENCL implementation on VIM3

[jpreisig]

I am starting to use the GPUs on the VIM3. The clinfo info command indicates that the GPU is configured as 1 device with 2 compute units and can support a maximum work group size of 384 with a work item dimension of 3.

Few questions:

I was under the impression that there were 4 GPU cores. Is this incorrect?

Do the two compute units each map to a GPU core? If not, what is the mapping from compute units to cores?

Does the work group size of 384 map to a number of shader cores in each compute unit?

How does the work item dimension of 3 map to a platform/physical core model for your implementation of OPENCL on the VIM3?

Regards,

Jim

[RDFTKV]

Hello, Welcome to the community.
The Khadas Team is on holiday and will return on 06 May. Hopefully one of the other users can answer your question before then.
In the meantime, there is quite a bit of documentation on the A311D in the downloads section, such as the Datasheet Hopefully something there can help.

[Electr1]

@jpreisig

Do the two compute units each map to a GPU core? If not, what is the mapping from compute units to cores?

regd. this please refer to this topic:

Which chip has the best GPU?

This data was from Wikipedia and some Arm forums, I am not sure about how true this data is, but this was what I was Able to get from them The VIM3 aka the A311D, Has 2 pixel cores aka the Shader cores with a more fluid name 4 SIMD lanes which is the “4” in MP4 6EE are the 6 execution engines, Since we have 2 Shader cores, 6/2 = 3 execution engines per core @numbqq please verify this Information, thank you I believe this is the correct quote.

the “mp” notation is outdated by today’s standard, and is wrongly listed on the spec sheet (it is supposed to be mentioning the number of execution engines in the GPU)