Davinci Resolve 16 + TB2 eGPU + RX580 benchmarks
 
Notifications

Davinci Resolve 16 + TB2 eGPU + RX580 benchmarks  

  RSS

(@koncept_zero)
Eminent Member
Joined: 3 months ago
 

My 2015 iMac build is here: https://egpu.io/forums/builds/2015-imac-171-6th-4c-i5-skylake-amd-rx-580-20-gbps-tb2-razer-core-x-macos-catalina-10-15-1-for-davinci-resolve-tested-5700-xt/

I put this setup together to get better performance in Davinci Resolve. The gains are modest but improved nonetheless. With the internal GPU, even with proxy files and render cache, a 4k timeline playback was very choppy. Now, it's smooth with the RX580 after generating render cache. For benchmarking exports times, I tested all the various GPU options within DR16 preferences. The goal was to test exporting a 1 minute 4k timeline to the standard 4k Youtube h.264 preset. No optimized media was used and no render cache was generated. Just straight h.264 media to h.264 output. The Youtube preset was not modified in anyway.

Note: I'll be updating this original post as I perform more tests. I still have several more to try. I will say some of the tests are confusing. Where one GPU preference works better with one codec, it works worse with another and vice versa. Not one GPU preference setting works best in all situations. That said, just leaving GPU preferences on Auto will produce solid results if you just want to set it and forget it.

The following apply's to all renders and export testing below: Footage was 8bit 4k from a GH5 at 150mbps and 100mbps Phantom 4 pro 8bit 4k footage. Both were h.264. The timeline contained music, text graphics with motion, color grading and speed ramps. No fusion effects were generated.

---------------------------------------

Internal GPU only with OpenCL took 5:18.

Internal GPU only with Metal processing took 4:44.

The worst GPU performance was using only the RX580 and unchecking the internal gpu on OpenCL, took 3:55.

Using Manual GPU settings, Metal and only selecting the RX580 resulted in an export time of 2:58.

Leaving the GPU preferences entirely on Auto exported in 2:51.

The best GPU preference settings for this example were to use Metal, Manual GPU selection, and choosing both GPU's. It took 2:32.

---------------------------------------

Changing nothing, except exporting to 422LT (vs to H.264), Auto, internal gpu only (RX580 unplugged), yielded a time of 4:44.

Changing nothing, except exporting to 422LT (vs to H.264), Metal, both gpu's, yielded a time of 2:16.

Changing nothing, except exporting to 422LT (vs to H.264), Auto settings with both gpu's, yielded a time of 1:54. 

---------------------------------------

Same project. Now I've generated 422 LT proxy files and 422 LT render cache. Exporting to a 4k 422 LT prores file, the Auto dual gpu setting yielded a time of 0:34.

Same project. Now I've generated 422 LT proxy files and 422 LT render cache. Exporting to a 4k 422 LT prores file, the internal gpu yielded a time of 0:36! WOW!

---------------------------------------

Still to come: timeline render cache times, stabilization times, and noise reduction times.

 

 

 

 

 

 

This topic was modified 3 months ago

2018 Mac Mini (8th,6C,B) + Radeon VII @ 32Gbps-TB3 (Razer Core X) + macOS 10.15.2


itsage liked
ReplyQuote