User Profile

Posts

Posts

Post not yet marked as solved
0 Replies
319 Views

How to get the remaining memory on gpu of macOS programmatically?

Hi, I want to know the gpu device available memory by code, so that I could calculate the size of buffer and texture I will create using Metal. I only find the MTLDevice's property currentAllocatedSize that can show the allocated memory for this process.
Posted
by cycle1024.
Last updated
.
Post marked as solved
3 Replies
256 Views

Metal Frame Capture Debugging Tools shader time isn't consistent with CommandBuffer's GPUTime

I profile a metal shader using both Frame Capture Debugging Tools and addCompletedHandler method of command buffer. The frame capture tool shows me that my shader costs 2.7ms (differs between 2.5ms-3ms for executing multiple times). And the command buffer I use following code: [commandBuffer addCompletedHandler:^(id<MTLCommandBuffer> cb) { &#9;&#9;CFTimeInterval executionDuration = cb.GPUEndTime - cb.GPUStartTime; &#9;&#9;// print executionDuration to console }]; The executionDuration shows about 5ms. What the difference between these two time duration?
Posted
by cycle1024.
Last updated
.
Post not yet marked as solved
0 Replies
301 Views

Core ML inference results on A14 device differs a lot between GPU and ANE

I have a model contains Convolution, LeakyReLU as activation layer and one concat layer. I convert it into .mlmodel format for IPhone12 inference. When the model is on GPU, the result is a little different from the result on CPU but still acceptable. When using ANE (Apple Neural Engine), the result differs a lot from CPU and GPU both. I have tried changing LeakyReLU to ReLU, but the problem still exists. The only difference in my CPU, GPU and ANE inference execution is changing compute units between MLComputeCPUOnly, MLComputeCPUAndGPU and MLComputeAll.
Posted
by cycle1024.
Last updated
.