Topaz Video AI v5.3.6
System Information
OS: Mac v15.01
CPU: Apple M4 Max 128 GB
GPU: Apple M4 Max 96 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 14.44 fps 2X: 08.83 fps 4X: 02.88 fps
Iris 1X: 14.95 fps 2X: 05.22 fps 4X: 01.94 fps
Proteus 1X: 17.43 fps 2X: 09.78 fps 4X: 03.05 fps
Gaia 1X: 04.09 fps 2X: 03.03 fps 4X: 02.22 fps
Nyx 1X: 06.04 fps 2X: 05.85 fps
Nyx Fast 1X: 14.32 fps
Rhea 4X: 01.13 fps
4X Slowmo Apollo: 16.14 fps APFast: 36.75 fps Chronos: 05.39 fps CHFast: 08.81 fps
16X Slowmo Aion: 16.10 fps
7 Likes
Topaz Video AI v5.3.6
System Information
OS: Mac v15.01
CPU: Apple M4 Max 128 GB
GPU: Apple M4 Max 96 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 3840x2160
Benchmark Results
Artemis 1X: 03.18 fps 2X: 01.82 fps 4X: 00.66 fps
Iris 1X: 02.64 fps 2X: 01.32 fps 4X: 00.42 fps
Proteus 1X: 02.91 fps 2X: 01.99 fps 4X: 00.66 fps
Gaia 1X: 00.95 fps 2X: 00.66 fps 4X: 00.50 fps
Nyx 1X: 00.98 fps 2X: 00.93 fps
Nyx Fast 1X: 02.35 fps
Rhea 4X: 00.24 fps
4X Slowmo Apollo: 03.31 fps APFast: 08.98 fps Chronos: 01.33 fps CHFast: 01.99 fps
16X Slowmo Aion: 05.19 fps
Topaz Video AI v5.3.0
System Information
OS: Mac v14.03
CPU: Apple M3 Max 128 GB
GPU: Apple M3 Max 96 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 12.63 fps 2X: 07.74 fps 4X: 02.67 fps
Iris 1X: 08.20 fps 2X: 04.83 fps 4X: 01.68 fps
Proteus 1X: 12.28 fps 2X: 08.33 fps 4X: 02.37 fps
Gaia 1X: 03.29 fps 2X: 02.33 fps 4X: 01.69 fps
Nyx 1X: 05.39 fps 2X: 05.06 fps
Nyx Fast 1X: 12.34 fps
Rhea 4X: 00.82 fps
4X Slowmo Apollo: 11.04 fps APFast: 41.16 fps Chronos: 03.50 fps CHFast: 06.33 fps
16X Slowmo Aion: 13.43 fps
see post #2
Uh-huh!
So, the new Apple Silicon, M4 series, (as expected) is just an ant size difference in performance (margin of errorā¦ 
) compared to the M3 Max Silicon, eh?!
Who knew?! 
I am surprised itās not more because the neural engine is twice as fast in the M4 compared to the M3.
Exactly what Apple was counting on, sadlyā¦ again!
Interestingly, the gain is a bit model dependent. Iris 1x seems to benefit quite significantly, while others do almost not at all.
Iām not really surprised. This is just what I have skimmed form quick searches of the internet, so I could be wrong. It seems like the Neural Engine in the M chips is only really made to run Transformers as defined by the Transformer Architecture. To me, that sounds like itās more useful for LLM style AIs, where youāre trying to generate something out of something else. Sure, video enhancing could be described with those same words, but itās possible that video is not includedāin a useful way to Topazāin the Transformer Architecture.
Anyway, maybe should research more before spouting continuing to spout such things.
For 5K+, āspoutingā (WTF) for shady marketing, shills (liars) YTers, ect.?!
Hmmā¦
Perhaps you need to PRACTICE what you preachā¦ research more, eh, rather than being this personā¦
I have SKIMMED form quick searches of the internet, so I could be wrong
I suppose I left out the part about knowing loads about how to make AI programs, what kinds of problems they can solve, and what tools are available to the public to run pretrained AI models with. Thatās why the quick search about how Apple decided to āunleashā āAIā to the masses. I just needed to answer if they went with something everyone already knows how to use, or did they reinvent it, Apple style? Would they be Apple if they didnāt?
Firstly, Apple hasnāt been TRULY innovative since the G.O.A.T., Steve Job. 2nd, under Tim Apple, the company has taken a huge hit of great talented researchers, scientist, ect. due to Cook fear of taking risk, ect. 3rdā¦ well, take a read below for starts. I do not want to continue putting you on blast here/derailing this thread, so, we could continue this discussion in PM if you like.
I guess we shouldnāt expect major speed improvements on the M4 Ultra whenever Apple releases next year.
Iām just wondering if Topaz Labs can do something more to optimize their code for Apple silicon
Topaz Video AI v5.3.6
System Information
OS: Mac v15.01
CPU: Apple M4 Max 64 GB
GPU: Apple M4 Max 48 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 14.89 fps 2X: 08.43 fps 4X: 02.95 fps
Iris 1X: 15.08 fps 2X: 05.23 fps 4X: 02.03 fps
Proteus 1X: 20.89 fps 2X: 09.59 fps 4X: 02.68 fps
Gaia 1X: 04.03 fps 2X: 02.90 fps 4X: 02.17 fps
Nyx 1X: 05.70 fps 2X: 05.71 fps
Nyx Fast 1X: 14.25 fps
Rhea 4X: 01.01 fps
4X Slowmo Apollo: 16.14 fps APFast: 42.51 fps Chronos: 04.74 fps CHFast: 08.65 fps
16X Slowmo Aion: 16.07 fps
2 Likes
Apple has made available the tools to convert pre-made AI models to work on their Neural Engine. There are a few small optimizations that can be done along with that conversion, but they mostly are just reducing the floating point accuracy. Since all the examples they post tend to revolve around natural language, Iām guessing reducing the accuracy in videos will have a different effect than just making things faster.
If Topaz has already used these tools, and the speeds are still not great, then itās up to Apple to make the conversion tool better. Besides starting over with Apple Neural Engines as the main AI computers, thereās nothing else Topaz can do.
1 Like
Windows 11 Pro 22h2 debloated
No Antivirus or cleaner programs
ASUS-PRIME Z790-A WIFI (Bios 1663)
i9-13900K + Kraken AIO
4090 drivers + 565.90 studio drivers
128 Gig DDR5-4800 UDIMM 1.1V CL40
Motherboard bios all auto no OC
Running Quick CPU
Core parking 100%
Frequency scaling 100%
Turbo boost 80%
Corsair HX1200 PS
4 onboard Nvme 4th gen drives
- Boot drive 2 TB
- Cache drive 2 TB
- Media drive 4 TB
- Mixdown drive 2 TB
Topaz Video AI v5.3.6
System Information
OS: Windows v11.22
CPU: 13th Gen Intel(R) Core(TM) i9-13900K 127.75 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 720x480
Benchmark Results
Artemis 1X: 189.35 fps 2X: 77.99 fps 4X: 22.98 fps
Iris 1X: 175.94 fps 2X: 112.39 fps 4X: 30.67 fps
Proteus 1X: 222.39 fps 2X: 118.45 fps 4X: 30.84 fps
Gaia 1X: 89.20 fps 2X: 59.71 fps 4X: 27.50 fps
Nyx 1X: 51.53 fps 2X: 46.89 fps
Nyx Fast 1X: 90.83 fps
Rhea 4X: 24.82 fps
4X Slowmo Apollo: 224.82 fps APFast: 355.93 fps Chronos: 170.05 fps CHFast: 197.50 fps
16X Slowmo Aion: 131.08 fps
Topaz Video AI v5.3.6
System Information
OS: Windows v11.22
CPU: 13th Gen Intel(R) Core(TM) i9-13900K 127.75 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 41.94 fps 2X: 18.01 fps 4X: 04.67 fps
Iris 1X: 39.38 fps 2X: 20.01 fps 4X: 04.81 fps
Proteus 1X: 42.33 fps 2X: 20.18 fps 4X: 05.04 fps
Gaia 1X: 15.92 fps 2X: 10.97 fps 4X: 04.50 fps
Nyx 1X: 17.73 fps 2X: 14.63 fps
Nyx Fast 1X: 31.10 fps
Rhea 4X: 04.19 fps
4X Slowmo Apollo: 40.60 fps APFast: 77.02 fps Chronos: 33.06 fps CHFast: 35.72 fps
16X Slowmo Aion: 31.64 fps
Topaz Video AI v5.3.6
System Information
OS: Windows v11.22
CPU: 13th Gen Intel(R) Core(TM) i9-13900K 127.75 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 3840x2160
Benchmark Results
Artemis 1X: 08.81 fps 2X: 04.80 fps 4X: 01.26 fps
Iris 1X: 08.27 fps 2X: 04.35 fps 4X: 01.22 fps
Proteus 1X: 08.74 fps 2X: 04.41 fps 4X: 01.20 fps
Gaia 1X: 03.40 fps 2X: 02.32 fps 4X: 01.02 fps
Nyx 1X: 02.94 fps 2X: 03.61 fps
Nyx Fast 1X: 05.21 fps
Rhea 4X: 00.95 fps
4X Slowmo Apollo: 14.41 fps APFast: 19.22 fps Chronos: 07.20 fps CHFast: 12.64 fps
16X Slowmo Aion: 15.06 fps
One thing Topaz could do very easily for Apple silicon macs is to provide an option to automatically reduce the max memory to 10% when processing standard definition sources, resulting in around 40% speed increase.
You can increase the overall performance even more for SD by simultaneously processing one video at 100% and another at 10% max memory.
But to get the absolute maximum performance for SD is more complicated. This involves segmenting the video, stacking or tiling the segments (to produce the equivalent of an HD source), then upscaling, then cropping the tiles back to segments and joining back to one video. However, this gives around a further 60% speed increase overall on my base spec M1 Max. And this last technique would benefit all computers for SD sources, not just Apple silicon. Although looking at the benchmarks, the benefit would be less for something like a 4090, but still noticable.
However, Iād be very surprised if Topaz did any of these things, even the simplest one.
1 Like
TOPAZ 5.4.0. *****
Topaz Video AI v5.4.0
System Information
OS: Mac v15.01
CPU: Apple M4 Max 128 GB
GPU: Apple M4 Max 96 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 14.11 fps 2X: 08.11 fps 4X: 02.74 fps
Iris 1X: 13.43 fps 2X: 05.22 fps 4X: 01.87 fps
Proteus 1X: 14.71 fps 2X: 09.31 fps 4X: 02.67 fps
Gaia 1X: 04.18 fps 2X: 02.99 fps 4X: 02.16 fps
Nyx 1X: 05.36 fps 2X: 05.70 fps
Nyx Fast 1X: 11.56 fps
Rhea 4X: 01.10 fps
4X Slowmo Apollo: 16.41 fps APFast: 43.47 fps Chronos: 05.39 fps CHFast: 08.64 fps
16X Slowmo Aion: 16.10 fps
Sorry if itās a naive question. Will 48gb ram improve the performance in any significant way for the M4 pro? I have an M4 pro mac mini with base configuration and i am seeing similar performance to the stats posted above.
Hello.
Noā¦ I mean NOā¦ consumer product will ever utilize that much system RAM (unless that software has some serious leaks, very poor optimization, ect going on) for todayās and/or tomorrowās workloads (allocation and usage IS NOT the same), regardless of what the naysayers/shills claim.
These companies Apple, Microsoft, Corsair, G.Skill, ect. ONLY care about profit, not truth.
Apple M4 max (40) (and ALL other Mx) would benefit from full METAL coding, gpu and neural maxed would have similar output to the nvidia 4080ā¦ please please optimize the software.
I get what you are saying. I am confused because ram and gpu are shared on the apple silicon machines. So technically the software could have access to 30-40gb gpu, if I understand correctly.
At the same time, I am seeing reports above that topaz isnāt utilizing the Mac GPU to itās fullest extent. Which is why I am so confused if I should have gotten the 48gb ram config.
