So, not a perfect comparison, but I have an Intel Mac i7 that I’ve been using Topaz Video Enhance on to do a pet project of upscaling some old TV show DVDs to put on my home video server. Using CPU rather than GPU because the latter on the mini is poor Using one of the Dione models, I get about 1.6 seconds per frame, which results in around 16+ hours to process a 30 minute video. I’ve been living with that reality for a couple months, but then it dawned on me last night to try Video Enhance on my partner’s M1 Mac mini. The same video that took 16 hours to upscale from 480i to 1080P took just over an hour on the M1—0.1 seconds per frame. I processed 5 videos overnight. Needless to say the M1 means completing an entire season in a weekend (so long as I’m allowed to continue borrowing the M1 ). I imagine the M1 Pro or Max would beat that time significantly. YMMV.
I don’t have a M1 Pro on hand anymore but I do have have the M1, M1 Max 24 GPU, M1 Max 32 GPU and M1 Ultra 48 GPU but I did test out the M1 Pro 14 GPU before I returned it. For VEAI the Pro, Max and atm the Ultra perform nearly identical. I am assuming VEAI needs to be updated to target the second batch of neural engines on the Ultra but that is just a guess on my part. I also have a slew of other GPUs I can run comparisons on but have a PC with a 3080ti next to me atm.
I generally upscale 1080p to 4k. For all of the M1 Pro+ machines with Artemis, Proteus they run consistently at about ~.25-.3 sec/frame. Same video with the same settings for 3080ti runs .45 sec/frame. The one exception where the 3080ti outperforms is Gaia at a similar .45sec frame. Gaia model on the M1 Pro+ devices does seem to show a difference between the M1 Max 24, 32 and Ultra 48 GPUs scaling almost linear to the amount of GPU cores so that model must be using more GPU than the neural engines. This is running running 2.6.3 for all of the machines.