for anyone curious about gtx 1650

2x 4k transcodes work
+1x 4k transcode fall back to cpu decode + gpu encode
note gpu memory 4 gig not quite enough for 3 gpu decodes
Unless they can do a ton more optimization of memory usage of the linux nvdec, @~1.4 gig video ram per 4k transcode
2g p400 = 1x 4k xcode
4g 1650 = 2x 4k xcode
5g p2000 = 3x 4k xcode
8g 1080 or whatever = 5x 4k xcode
for whatever reason, VIDEO RAM is the bottleneck for linux transcoding.
If you need more transcodes per ram, then windows is apparently way more efficient and your best bet, instead of linux.