Old NVIDIA Geforce GTX 1080 is about 8 times faster than Apple Mac Studio M1 when unmix

When Apple silicon support for AI processing devices comes?
Old NVIDIA Geforce GTX 1080 is about 8 times faster than Apple Mac Studio M1 when unmix. :grimacing:

1 Like

I’ve been discussing with Apple recently about this, its’ not that the M1/M2/M3 are not capable of some interesting AI acceleration, it’s rather that the Apple software stack to do it is… incomplete and quite experimental to say the least. I’ve been providing them with several AI models from SL10 and the upcoming SL11 to help debugging their AI tools, hopefully the situation will get better over time.

7 Likes

The Unmix function is probably not used heavily by many people. There may be development in the matter when the time is right, or something more important can be done with the help of AI. Thank you for the information.

edit:
Of course Unmix is important. Every function is. :))

1 Like

Interesting. I tried an M3 two days ago and I was sadly unimpressed with its performance. My 5 year old 1950x is faster, even with its RX580. Windows has come a long way.

1 Like

Apple don’t have Nvidia CUDA, there’s no way Silicon can compete with it, the same is true for GPU audio dsp and gaming, there’s a reason people use Intel/AMD and Cuda for this work and it’s because its powerful and well-developed/evolved over decades of X86 CPU and GPU improvements.

2 Likes

If anyone is interested, Here is video explaining how GPU drivers and optimization actually work (from one the former chief engineers)

About Cuda, I actually found out that AMD secretly had ported Cuda for years (through Zluda) but scrapped it for unknown reasons.

2 Likes

A little off-topic but still related:

Here’s another video explaining video compression and decompression and encoding and decoding. So informative. A must watch