Could we have SLP benchmarks integrated in the benchmark tool? This is the most interesting model to me at this time. The benchmark would report the # of seconds required to enhance one frameā¦
I ran 2 videos of approximately the same complexity on 2 different systems:
Mac Mini M4 10 CPU 10 GPU 16GB RAM ~14.3s per frame for a 328p video
MacBook Pro M5 10 CPU 10 GPU 24GB RAM ~10s per frame for a 336p video
I understand the videos are not exactly the same and I have more RAM on MacBook Pro, but even with 24GB of RAM, I see 8-9GB used by the Neuroserver⦠and both systems are actively cooled.
So this would mean M5 is ~40% than M4 for SLP? If true, this is quite impressive!
My observations.
For Starlight Precise 2.5, the encoding rate (SpF) is fairly proportional to the frame size in pixels.
On the RTX 5080, SpF is a bit greater (x1.04) than the frame size in megapixels.
The frame size of an NTSC DVD, scaled to square pixels (720x540), has a size of 720x540x2^2 ā 1.56 Mpx when scaled by a factor of 2 using SLP.
The resulting speed is 1.6 SpF. With a threefold increase, it is 3.6 SpF.
Card performance can be measured in SpF/Mpx (the less, the better). The RTX 5080 has a performance of 1.04 SpF/Mpx.