mlx-examples
mlx-examples copied to clipboard
Speed Benchmarks vs NVidia GPUs
Given that this is the only official package endorsed by Apple, would it be possible to post benchmarks against some of the Nvidia GPUs available in the market so that we can understand where Mac M1/M2/M3 Max->Ultra system stand in terms of raw speed with respect to training and inference ?
This could be a good starting point.
Could you please share some details what sizes (model, parameter count,..) the numbers in the graph relate to!? Guess it must be pretty small if results for 6GB (and ancient GTX) are included.
When I read about MLX on Apple Silicon and all memory shared between CPU and GPU two weeks ago, I ran out and bought a MacBook Pro Max M3 w/ 36GB and I'm trying to get a grasp on what I can run w/ it.
I also got a brand new Xeon 3435X w/ 256GB and 2x RTX 4000 w/ 20GB each to compare Apples to (Intel) plums and (Nvidia) spinach. ;-)
Thx G.
M1 Ultra here, can help benchmark if scripts are provided
M2 Ultra, same comment as above ;-)