Posts

Post not yet marked as solved
1 Replies
624 Views
Hi,here is my config:macOS CatalinaVersion 10.15 Beta (19A526h)MacBook Pro (13-inch, 2017, Four Thunderbolt 3 Ports)3.1 GHz Intel Core i58 GB 2133 MHz LPDDR3Intel Iris Plus Graphics 650 1536 MBI have the Mac hooked up to the Sonnet eGPU AMD Radeon RX570.The eGPU is recognized, it is showing in the top bar. When I lanuch create ML after the eGPU is connected, it runs the object detector I have put in place, about 4 times faster than on the intel GPU.Now, the big issue is that the results are completely different. When I run on a reasonnable set (500 images, 2000 annotations) I am getting a reasonable result on Intel (above 70% accuracy on my data). When it is running on the eGPU, I am getting 2-10% accuracy on the same data, which is confirmed in my testing afterwards.I am at a loss trying to understand this - is the eGPU using the same maths as the Intel GPU? Arre Metal/Accelerate tested/validated for this GPU?Thanks,David.
Posted
by dmendels.
Last updated
.