Is there a way to know the number of kernels that a GPU has?

[laurentlavigne]

That’s for tuning the number of kernels in ComputeShader.Dispatch.

[c0d3_m0nk3y]

First I thought you are asking for the number of #pragma kernel lines in a compute shader.
Then I thought, you are asking for the thread group size of a kernel as it’s defined in the shader.
But, now I’m thinking, you would like to know what the best thread group size is to choose.

I’m not aware that there is a way to query that with any API. The usual recommendation is to use a power-of-two size between 64 and 256. If you don’t use group-shared memory, it’s usually better to use a smaller number. If you do use group-shared memory, it’s usually better to use a bigger number. But it depends on the shader, so you have to experiment.

This has already been answered many times before:

Compute shader thread dispatching

Hey I am confused what numbers i should use for threads for my compute shader. I want to process a buffer of 4096 float2 elements. So what should i be using as my threads and calls in the C# code? I asked on stack overflow and go no information. Do i just do: [1,1,1] in C# and [4096,1,1] in the compute shader? I can't find a good explanation on this stuff, its a bit of a random guess work at the moment. Can any one explain sufficiently how to pick good number combinations ? Thank…

Optimizing GPU occupancy and resource usage with large thread groups

What is the algorithm to determine optimal work group size and number of workgroup

https://www.intel.com/content/www/us/en/develop/documentation/iocl-opg/top/optimizing-opencl-usage-with-intel-processor-graphics/work-group-size-recommendations-summary.html
https://www.reddit.com/r/vulkan/comments/u3i6we/what_does_the_local_workgroup_size_in_compute/

[laurentlavigne]

@c0d3_m0nk3y

First I thought you are asking for the number of #pragma kernel lines in a compute shader.
Then I thought, you are asking for the thread group size of a kernel as it’s defined in the shader.
But, now I’m thinking, you would like to know what the best thread group size is to choose.

I’m not aware that there is a way to query that with any API. The usual recommendation is to use a power-of-two size between 64 and 256. If you don’t use group-shared memory, it’s usually better to use a smaller number. If you do use group-shared memory, it’s usually better to use a bigger number. But it depends on the shader, so you have to experiment.

This has already been answered many times before:
https://discussions.unity.com/t/804846
https://gpuopen.com/learn/optimizing-gpu-occupancy-resource-usage-large-thread-groups/
https://stackoverflow.com/questions/10096443/what-is-the-algorithm-to-determine-optimal-work-group-size-and-number-of-workgro/10098063#10098063
https://www.intel.com/content/www/us/en/develop/documentation/iocl-opg/top/optimizing-opencl-usage-with-intel-processor-graphics/work-group-size-recommendations-summary.html
https://www.reddit.com/r/vulkan/comments/u3i6we/what_does_the_local_workgroup_size_in_compute/

Oh thanks, I was afraid of using too many threads with 64 on Xe but that seems to be the sweet spot.