Comparison of VEAI 2.6.4 in Windows 10, Intel Mac Pro 2019 (Monterey) and Apple Silicon M1 Max

VEAI Tested on Three Computers

The clip used for these tests is a 1 min 15 sec DV interlaced clip shot in 1993 with a Sony VX3 camcorder. In all cases the following settings were used in VEAI. Interlaced, Medium Quality, scaled to HD (1920X1080), h. 264 output, Crop To Fill Frame turned off, Grain turned off. An external Thunderbolt 3 Samsung X5 SSD was used for the Source and Output. Time to complete processing was determined by hovering over the thumbnail in VEAI after processing was completed.

The first test was run using a Late 2019 Mac Pro (12 cores, 192 GB memory, AMD Pro Vega II video card with 32 GB video ram). The Mac Pro was booted natively into Windows 10 using Bootcamp. Note that in this configuration the Mac Pro is not running Windows 10 by emulation but natively. No other applications were running during the test. VEAI was set to make maximum use of the GPU.

The second test was run with the above Late 2019 Mac Pro running Mac OS Monterey (12.0.1). No other applications were running during the test. VEAI was set to maximum use of ram in VEAI preferences.

The third test was run on my M1 Max (10 core cpu, 32 core GPU, 64 GB of Unified Memory running Mac OS Monterey (12.0.1). No other applications were run during the test. VEAI was set to make maximum use of ram in preferences.

Test 1: Windows 10 On Late 2019 Mac Pro and VEAI V. 2.6.4

  1. Using the Dione Interlaced Robust Model V. 4. Total Processing time = 3 min 8 sec at 0.08 sec/frame.
  2. Using the Dione Interlaced DV ModelV. 3. Total Processing Time = 6 min 16 sec. at 0.08 sec/frame.
  3. Using The Dione Interlaced TV Model. Total Processing time = 5 min 54 sec at 0.08 sec/frame.

Using Windows 10 Task Manager showed a cpu usage of ca. 35-40% and a gpu usage of ca. 50%.
ETA’s given in VEAI were accurate.
All processed videos had no sound.

Test 2: Late 2019 Mac Pro Running Monterey and VEAI V. 2.6.4

  1. Using the Dione Interlaced Robust Model, V. 4. Total Processing Time = 3 min 10 sec at 0.08 sec/frame.
  2. Using the Dione Interlaced DV Model V. 3. Total Processing Time = 6 min 21 sec at 0.09 sec/frame.
  3. Using the Dione Interlaced TV Model. Total Processing Time = 6 min 16 sec at 0.08 sec/frame.

During processing Activity Monitor showed a cpu usage of ca. 1,009% and a gpu usage of ca. 44%.
ETA’s were accurate.
All processed video had no audio.

Test 3: M1 Max and VEAI 2.6.0

  1. Using the Dione Interlaced Robust Model. Total Processing Time = 2 min 53 sec at 0.07 sec/frame.
  2. Using the Dione Interlaced DV Model. V. 3. Total Processing Time = 6 min 48 sec.
  3. Using the Dione TV Model. Total Processing Time = 5 min 54 sec at 0.08 sec/frame.

During processing cpu usage was ca. 400% and gpu usage was ca. 3-4% as shown by Activity Monitor.
ETA’s were accurate.
All processed video had no audio.

Conclusions and Observations

My Late 2019 Mac Pro has no neural network and so VEAI is forced to use the cpu and gpu for its processing in both Windows 10 and Mac OS Monterey. The M1 Max has a neural processing network which VEAI is using, but VEAI makes low use of the cpu and gpu in its processing. Further improvements in the native Apple Silicon version of VEAI are needed to utilize the full power of this computer.

The Dione Robust Model V. 3 runs significantly faster on all three platforms. It also appears to run somewhat faster on the M1 Max than on either the Mac Pro in Monterey or in Windows 10. The Dione Interlaced DV Model runs faster on Windows 10 than on the Mac Pro under Monterey or the M1 Max. The Dione Interlaced TV Model runs faster under Windows 10 and on the M1 Max than on the Mac Pro under Monterey.

For purpose of improving the quality of interlaced DV tapes, the Dione Robust Model V. 3 running on the M1 Max provides the fastest processing time and the highest quality result when upscaling to HD (1920X1080).

Tom

6 Likes

Thanks for that.
Was the High Compression setting used on these?
I ended up picking the Robust Model for speed.
Watching from the couch I really can’t see a difference. But on the computer screen I see that (for me) TV model is a bit better quality.

Yes, the high compression setting was recommended and that is what I have used. I have not experimented much with other models on these DV tapes. The Interlaced Robust model is giving me very good results and so I haves stayed with that as processing is very fast (0.07-0.08 sec/frame on my M1 Max 16in. MacBook Pro. I am very pleased with this latest Beta of VEAI for enhancing my archived video. I have noticed only one issue. At the end of a processing session, if I try to play back one of the VEAI processed video by double clicking on it QuickTime does not open and nothing happens. If I reboot my M1 Max, the processed video is played back in QT as expected.

Tom

1 Like

Got the same MBP :grin:
I seem to have found a bug with batches using Dione Interlaced DV. The first video 's frame rate is doubled (as expected) the rest are not. :man_shrugging:
This is on v2.6.1 though. I unfortunately am not a beta tester.

I just checked my latest batch of DV videos process with Dione Robust Interlaced DV and all the videos were processed to the expected 30 fps. This is using the latest beta version and scaling each video to 1080p.

1 Like