torch.mps.synchronize doesn't seem to work #3
Description
I get a insanely high flop throughput (like 17 PF) on my Mac M2 Ultra.
Using mlx seems to produce reasonable numbers synchronizing the GPU with mx.eval(output_marix)
{{ message }}
I get a insanely high flop throughput (like 17 PF) on my Mac M2 Ultra.
Using mlx seems to produce reasonable numbers synchronizing the GPU with mx.eval(output_marix)