Alright, I have words and results to share. I got my 5600G and 5700G to work. I updated the motherboard BIOS and had to fiddle with my docker pool (2x500gb SSD) in order to add the newer GPU to the Plex App. It would NOT work with my previous pool created with the 4600G. Don't ask me why, I just know that's what I had to do before I could add it with a 5000 series processor. This is a new lab build, so no harm no foul. No changes were necessary swapping between the 5700G and 5600G.
THESE TESTS ARE NOT SCIENTIFIC, do NOT treat them as hard evidence of ANYTHING. Take them with a grain of salt.
However: these results are okay enough for my book given the time and resources allotted. In short, I'm just a guy that had time and some processors on hand. I didn't intend to do this testing, but here I am writing this at 1:30 am.
YOUR MILEAGE WILL VARY.
The screenshots bounced around +/- 2% on Plex and +/- 8% ish on CPU usage. I grabbed them when they appeared consistent, so neither best nor worst transcoding speed or CPU usage.
System Specs:
- AMD 5600G with Noctua NH-L9x65 SE-AM4 (SEALED CASE)
- AMD 5700G with Noctua NH-L9a AM4 (OPEN CASE)
- 64GB RAM
- 5x14TB RAIDz2 + Mirrored Intel Optane Metadata vDev (Data / Transcoding Pool)
- 2x500GB Mirrored SSD Pool (Docker+ Config Pool)
Results:
- First place is 5700G with hardware disabled
- Second place was the 5600G with hardware disabled
- Third was 5700G, hardware = Cezanne
- Fourth was 5600G, hardware = Cezanne
- Fifth was the 4600G, not shown, not tested. GPU DNF.
I tested by optimizing files for TV (
output files are 1080p, 8Mbps)
1080p SDR source file (MKV | 1920x1080 | H.264 | 23,000 kbps | 18ish GB)
4K HDR source file (MKV | 3840x2160 | HEVC | 57,000 kbps | 79ish GB)
It appears that there is a performance cap of about 10-11x speed on transcoding in the software. The CPU was never pegged at 100% usage in any test, but had the thermal and CPU headroom to do more
It also appears that the CPU is put in charge of tone-mapping metadata. Whenever Hardware Transcoding is enabled with tone-mapping the CPU was put to work but never as hard as CPU transcoding.
The Cezanne GPU within these processors is also slower to transcode than using the CPU, in fact the 5600G CPU was 1.75x faster than the GPU, and 5700G was 1.6x faster than its GPU. That said, based on temperatures, hardware is likely more energy efficient given the CPU core heat (again, totally unscientific).
Different situations require different configurations. If you're in need of CPU cycles for other apps, then having the hardware is nice. It's doubtful however that in a home scenario you'll find yourself pegging these powerful CPUs very high.
