3 ms·

For inference, sure. Not yet for training. Weight quantization is much easier than weight + activation.