Windows vs MacOS vs Linux AMD Results

So I decided to run some tests between Windows 10 Pro 20H2/ MacOS Catalina / Manjaro 5.9.8-AMD - Using opengl workaround AMD 20.40
- AMD 3900X, 32GB 3200Mhz, Couple identical NVMe's, AMD 5700XT
- Davinci Resolve 16.7 - FREE
- All OS's are fully update to date as at 2020-11-18. Windows 10 Pro and Mac are both clean installs with latest drivers.
- 2 x "identical" projects were tested. a) Source H264/AAC b) Source in dnxhr_hq.
- 3 x Outputs of 1 x WAV, 1 x Cineform YUV 10bit, 1 x Cineform RGB 16 bit
- Filesystems used: Windows - NTFS, MacOS APFS, Linux - XFS (OS) + F2FS (Scratch)
- Windows and Mac both used physically different Source and Destination drives. Linux same to same partition.
- I did notice Linux CPU throttling near the end of the 3rd test. CPU hit 82 - Stock cooler. Mac and Windows no throttling.

- Final renders I haven't done yet but I will as I use FFMPEG and who knows maybe there's some differences.
- Some of the source files are around 12GiB in size - with DnxHR

Project is a short mix of fusion, chromakey, colouring and transitions.

Here are the encode times

Windows 10 Pro - H264 Source - 00:36, 02:57, 05:43
Windows 10 Pro - DnxHR Source - 00:33, 02:48, 05:33
RAM Peak 14GiB
VRAM Peak 5.8GiB


MacOS - H264 Source - 00:28, 01:45, 01:49
MacOS - DnxHR Source - 00:27, 01:48, 01:58
RAM Peak 12GiB


Linux - DnxHR Source - 00:32, 01:29, 01:58
RAM Peak 8GiB

Final output file is around 24GiB.

I was pleasantly surprised with the Linux result as last I tested it sucked with fusion. I can't wait to test with 17. When I figure out how to add VRAM for the others I will update this post.

Just an update for those interested. Now you'll understand why it could be worthwhile to run Linux.

I have also observed the difference in favour of Mac OS on the same AMD hardware. Both GPU and CPU show substantially better performance, the CPU also in Cinebench so not only in raw speed test. It seems opencl on Windows is in a sorry state.

Sent from my Mi 9T using Tapatalk