So I've had some more time to run some thermal testing this week and results are a bit of a mixed bag but make sense the more I think about it.
Under the same GPU only load at 180W, the liquid temp does not change from the previous set up with the idling CPU in the loop. The fans run slightly slower but not by much. I guess this makes sense since an idling 5600X is really only pushing like 25W. Here's a quick summary of GPU only temps
| GPU TDP | Liquid Temp | Fan Speed |
| GPU + CPU in Loop | 180 W | 48°C | 82% |
| GPU only + CPU on AXP-90 | 180 W | 48°C | 76% |
For a combined CPU and GPU load things are probably the most disappointing. Comparisons are not directly comparable because I'm now using the new PBO2 with the new curve optimizer but, as mentioned earlier, there's definitely not the same overclocking headroom. I'm definitely right up against the thermal throttling limits on the CPU with the stock 76W PPT. Previously I was able to push that up to 105W at similar CPU temps.
| GPU TDP | CPU PPT | CPU Temp | Liquid Temp | Fan Speed |
| GPU + CPU in Loop | 195 W | 105 W | 95°C | 54°C | 100% |
| GPU only + CPU on AXP-90 | 195 W | 76 W | 95°C | 47°C | 100% |
Fan control is also more of a hassle. I can't set the fan speed based solely on liquid temp because of the shared CPU/Radiator fan. There could be scenarios where the CPU is under load but the GPU is not. The GPU loop temperature also has an effect on the CPU temperature because the hotter the loop temp, the hotter the exhaust is that's cooling the CPU heat sink. I've settled with using some virtual temperature sensors on the Aquacomputer Quadro that are enveloped for the worst case of relative loop liquid temps and CPU temps. It's a little annoying but seems to work well enough.
I'm also not sure if the set-up has the thermal overhead for the RTX 3080 I was planning to upgrade to. Ideally I would run stock settings because I have very little appetite for tweaking undervolt settings but that would probably be required. And at that point I think I may as well add the CPU back into the loop.
Not sure if I would recommend this set-up - but probably need to do a little more testing on more real world use cases instead of artificial loads. Will probably try some more things this weekend
