Attached is a selection of results on my M2 MacBook Air.
@07broom.fibulae Thank you so much for your report
Did the machine (MacBook Air M2) not heat up too much ?
Have there been any significant slowdowns or power losses ?
Thanking you very much
You’re welcome. The aluminium part of the case between the keyboard and the screen got a little warm but nothing out of the ordinary. I wasn’t using the machine for anything else at the time. I noticed the frame rate started at around 17.5 but reduced to about 13.5, which is to be expected for a MacBook Air due to not having fans for cooling. The initial test was done with processing memory set to 10% for force the processing on the the GPU cores. I did the test again with processing memory set to 100% so the processing would be done on the Neural Engine. This finished slightly quicker at 7m41. Also, the case remained cooler and the frame rate remained around 15.4 fps. On my Mac Studio with 24 core GPU, I always ensure processing on the GPU cores as it’s always faster (but with much more power consumption). However, as my MacBook Air only has 10 GPU cores, the 16 Core Neural Engine wins in this case on speed and power.
Thank you again for your return. I dare to ask for your opinion, do you think that if I want to do more or less 200 episodes of 20/30 min, the MacBook Air M2 could hold ?
I won’t do it all of a sudden but little by little.
I’m sure that would be fine, if you go for the 24 GB M2 Air and if you make sure the processing is done on the Neural Engine (by setting the processing memory to at least 90% - which is the default anyway). That means you can use the machine for other tasks at the same time if you want to.
Thank you very much for all your feedback and advice. So I will be able to take the MacBook Air M2 without too much problem to do what I want to do I think. Knowing that it will be my heaviest task. And that won’t be every day either.
Once again, thank you for your availability
You’re most welcome.
Beware though that the speeds reported above are for just a 2x upscale. 4x is considerably slower.
But then in my (short) tests there wasn’t really much gain in going to 4k with that material or other models and it also looked quite fine in 2x upsclaed resolution (1416*960).
Also the used model Artemis is the fastest of them all but again still seems very appropriate for that type of input.