3 ms·
Local ML isn't a CPU workload. The NPUs in mobile processors (both laptop and smartphone) are optimized for low power and low precision, which limit how much me
by wtallis 2y ago
Local ML isn't a CPU workload. The NPUs in mobile processors (both laptop and smartphone) are optimized for low power and low precision, which limit how much memory bandwidth they can demand. So as I said, demand for more memory bandwidth depends mainly on how powerful the GPU is.