Quote from: Bestham on March 07, 2026, 17:31:54The real advantage with these machines is not so much raw performance of graphics rendering, but LLM inference. There is no Nvidia 5060 does not have access to 24 or 48 GB memory with 300 + GB/s bandwidth like the the M5 Pro.

The price of M5 Pro with 64GB is exactly same here as was M1Max/32GB with nice uplift in OpenCL and Metal scores. Valid only if there is no throttling for combined CPU+GPU tasks.
M1 MAX was famous for running both CPU and GPU with max clocks without throttling and fans at 50%

"Comparing an Apple GPU with an Nvidia part in OpenCL performance is hardly fair, considering that the API has long been abandoned on the Mac platform." I guess that explains the title!

Bestham, true, and for prompt processing there might no be even that big of a difference:

QuoteNeedless to say, Nvidia's GeForce RTX 5060 Laptop has a clear upper hand in the OpenCL test, where the GeForce GPU manages a roughly 13% lead over the M5 Pro.

5060 Laptop' 384 GB/s aren't that much faster either and the M5 256-bit APU makes its 307.2 GB/s (= 256-bit * 9600 MT/s / 1000 / 8) more than up with its much more RAM indeed.

Crazy 3600 pricing for only 8 GB VRAM on the Galaxy Book6 Ultra.

The real advantage with these machines is not so much raw performance of graphics rendering, but LLM inference. There is no Nvidia 5060 does not have access to 24 or 48 GB memory with 300 + GB/s bandwidth like the the M5 Pro.