I only have the 2 onboard NVMe drives in the Z6G4 chassis, all the mechanical drives are in an external case, connected through the HBA & expander card.
Actual cost (very roughly): 200W @28p/kwh for 19 hours & 6.5p/kwh for 5 hours a day = £1.13/day = a little over £400/year, but obviously not all this will be saved with any solution.
I was in a similar situation, I built a server with desktop components, HBA's, GPU and many HDDs/SSDs/NVMEs, but not matter what, 80W+ was the minimum idle..
With a little bit of rationalisation , I swapped to a dedicated NAS Appliance and am running
- 2 x VMs (Win 11 and Home Assistant)
- 35+ Docker Containers (including Plex, Jellyfin, Channels DVR, TVHeadend, Duplicacy (for backup), All the download/indexers including Gluetun VPN, etc, etc, etc)
- Unraid OS.
I initially rationalised all that on 5 x 14TB HDDs, 1 x 4TB SSD and 2 x NVME to allow me to move to a 6 bay (+ 2 x NVME) NAS Appliance.
With careful thought, the main HDD Array is largely idle, only spun up if someone watches a movie/tv or listens to music, and then its a single disk, everything else is using cache either RAM, SSD or NVME..
Until very recently all that was in a Terramaster F6-424Max (6 Bay + 2 x NVME, i5-1235U 10C12T / 64GB RAM) and idle is 30-35W with everything running, and goes to 45-50W when people are accessing media.
This uses the iGPU for HW Transcoding, ML (for immich) etc, so ditched the full GPU.
I've just upgraded to a WTR Max (AMD 8845hs 8C/16T 6 Bay + 5 NVME), primarily as I started compiling more on the NAS and it has more cores/grunt..
Its not as energy efficient, but idles around 38-45W and is around 50-55W when people are accessing media.
The iGPU is more powerful than the Intel and I've got it doing the same transcoding/ML although support is lagging Intel, it's just about there.
I also use a JetKVM (isolated on an IoT VLAN) to save me having a monitor plugged in so I can reboot, go in the bios etc remotely.
The other thing to do is grab a free for life VM on Oracle (need to give them CC Details, but you don't be billed), I maxed out the 4 x ARM Cores + 24GB RAM / 120GB HDD space and run inconsequential dockers on that (behind a cloudlfare tunnel) and have that on my Tailscale tailnet so it can access my server (also on the Tailnet).. great for containers that are quite CPU intensive but don't need to be running locally in my house.. might as well use their electricity for free!