5 ms·
Which is fine in this class of device (i.e. no access to proper compute shaders) as long as it's faster or more power efficient than doing it any other way on t
by zamadatix 5y ago
Which is fine in this class of device (i.e. no access to proper compute shaders) as long as it's faster or more power efficient than doing it any other way on the device.
- varelse 5y agoThat's going to be very task dependent but probably okay for matrix multiplication and streaming kernels which is most of AI.