VEAI 3.0 Has Major Issues Compared To VEAI 2.6.4 On Two High-end Macs

I processed a 2:14:20 Sony Hi-8 camcorder clip on both my Late 2019 Mac Pro 6.1 and my M1 Max MacBook Pro in both VEAI 3.0 and VEAI 2.6.4. My Mac Pro 6.1 I running the latest
MacOS 12.6, has 192 GB of ram, an AMD Radeon Pro Vega II graphics card with 32 GB of video ram. My M1 Max is the fastest available with the maximum specs. The video clip is of good quality and was enhanced to 1920 X 1080 using Dione DV 2X (since this is an interlaced clip). The result was output to H. 265 Main. The results are:

ON the Mac Pro VEAI required 57 min 4 sec to process this 2: 14 clip and processed it at 0.40 sec/frame. Exporting to ProResLT gave the same results.

Using VEAI 2.6.4 this exact same clip was upscaled to 1920 X 1080 using Dione Interlaced Robust and exported to H. 264 in 2 min 12 sec with a processing time of 0.09 sec/frame.

On my M1 Max this same clip processed in the same way as above on my Mac Pro required 4 hr 37 min and processed at 2.06 sec/frame. With VEAI 2.6.4 on my M1 Max this exact same clip processed using Dione Interlaced Robust to 1920X1080 in 2 min 45 sec and processed at 0.07 sec/frame.

The above results speak for themselves. VEAI 3.0 is DRAMATICALLY slower on both of these high end Macs than the previous VEAI 2.6.4. In fact, it is so slow that it is totally useless to me in its current state. I cannot recommend purchasing VEAI 3.0.

I am also sad to see that there is no way to contact Topazlabs support to report these results and get any help.


1 Like

Try to reinstall VEAI and try again, I have Macbook pro M1 Max with 32core GPU and 64 Gigs Ram and speed for me in both are similar one for few models 3.0 is little slower. As you told it quite a lot slower. Try fresh reinstall and share your results.

I uninstalled both VAI 3.0 and the latest beta 3.0. I rebooted my M1 Max. I downloaded VAI 3.0 from the Topaz web site and installed it. I set the memory usage in the app preferences to 100%, closed the app and rebooted the computer. I brought the exact same clip into VAI 3.0 and set the upscale to 1920 X 1080, interlaced, Dione 2X, and export to H.265. This time I could see that it was going to take about 23 min to export this 2 min 14 sec clip at a processing rate of 0.20 frames/sec. While better than previously, this is still almost 3X slower than VEAI 2.6.4 which was processing the exact same clip with the exact same settings at 0.07 frames/sec.

3.0 on my RTX 3080 12gb is 50% or more … slower on the same video doing chronos slow mo the same speed 30 FPS to 60 FPS Nvidia 264 codec … when 3.0 was in the middle on processing video i open the same video in 2.6.4 on the same machine and use Chronos slow mo V2 the same speed 30 to 60 FPS H264 and it show 50% less time of 3.0 for the Whole video !! For now i stay with 2.6.4.

1 Like

Curious what are the specs of your input file?

Is there a dramatic difference in quality output between 2.6.4 and 3.0?

In the middle of an upscale project, Mac M1 mini, nothing as beefy as your setup. Original video is 720x576, 25fps, 20 minutes. Video is processed over a fast network.

In VEAI 2.6.3, Dione Interlaced Dehalo V2, Top first, No noise. Output to HD, MP4, Constant Rate 18.
Estimated time: 2 hours 10 minutes, which is typical for the other videos in this batch.

In VAI 3.0, Out to HD (original pixel/letterbox/50 fps), Enhancement - Interlaced, Auto-detect, Dione TV 2x FPS, no noise, output to H.264 high, MP4, auto bitrate & audio. No other filters. Max 1 process. While not exactly the same configuration it seemed comparable.

Estimated time: 1 day 22 hours 9 minutes, after waiting for a few minutes of video to be processed (original estimate was 3 days). Like for you, 3.0 is so insanely slow it’s effectively unusable for me.

1 Like

Did you try the little purple chat icon on the main website, and flip the switch over to “Ask”? Ostensibly that email should go to support…

The video is 720x480 that I digitized some time ago. I’ll have to check it in QuickTime to get further info, but there is nothing special about the video file.

I have enabled Detailed Logging in VAI 3.0 and sent the results for my video project to Ida in Topaz Support. Hopefully Topaz can find some reason for the large decrease in performance between the last release and this one. Otherwise, VAI is so slow as to be useless to me.

I missed seeing that icon, but I was able to contact Ida in Topaz Support and send detailed logging results for my VAI 3.0 project. I am hopeful that these logs will help Topazlabs resolve the issue with the poor performance of VAI3.0.

I just ask, because 720x480 is interlaced video. Also, 720x480 uses non-square pixels (pixel aspect ratio .91 assuming it is a 4:3 image).

I always use video that is square pixels. So obviously, I have to de-interlace my 720x480i video into 640x480p video. I have not been very thrilled with the quality of VEAI deinterlacing. There are better methods out there at least for me.

I also want to make sure my file is not interpreted differently. Interlaced video gets messy.

I’ve always gotten faster and better results using progressive video than interlaced with VEAI.

I’m curious if the Mac version is actually utilizing the M1 chip?

1 Like

Currently VEAI is not utilizing gpu completely hope they will enhance speed and more optimize it. On M1 Mac it’s showing only maximum 28% utilization

1 Like

Well, looks like I’m waiting for 3.0.1
I wonder why it is so much slower when using the same models?

I have checked the video clip that I have been using in VideoAI 3.0 using Quicktime Info which reports the following: Resolution 720X480 (640X480); Data Rate: 37.99 Mbits/sec; Video Format: DV/DVCPRO - NTSC; Audio Format: Linear PCM, 24 bit little-endian signed integer, 48000 Hz; Encoded FPS: 29.97; data rate: 28.77 Mbit/s; Aspect Ratio: 4:3; Pixel Aspect: 10:11;; Color Primaries::SMPTE-C; Transfer Function:ITU-R BT.709;YCbCr Matrix:ITU​-R BT.601;Code Points:(6-1-6); Track ID:2

The video clip plays fine in VAI 3.0, but is processed many times slower in VAI 3.0 compared to the same clip processed with VEAI 2.6.4 as per my data presented earlier in this thread. .

Maybe convert the DV file to ProRes 422?