Thanks for all the discussion everyone, I really appreciate it! Hopefully it’s useful for others too.
Agreed, I was sloppy in my methodology (changed two variables). But, I recently did some power consumption testing of another, completely different system (AMD Ryzen based) and didn’t see any meaningful difference in 2x16GB vs 4x32GB RAM. Which was surprising to me. I’ve lost my notes, but many years ago I did power consumption testing with the same X10SDV (D-1541) board and definitely saw a difference power draw with different amounts of RAM - but that was with RDIMM modules, but now I’m using UDIMM modules.
For science , I’ll put the 2x8 modules in the D-1541 system this weekend so we can see what that does to power draw.
I’m also doing some testing with the Denverton board in isolation. It doesn’t have integrated 10gbe, but it does have 4x1gbe (Intel i350). I disabled those (again, via physical jumper), and that only makes a 1 or 2 watt difference (negligible, considering my Kill-a-Wall isn’t likely that precise anyway). My guess here is that the i350 is probably smart enough to shut down most of its circuit when a cable isn’t plugged in.
I still have my RDIMMs, so I’ll do some testing with those as well.
This past Friday evening, I did a RAM swap: I put the 2x8GB in my TrueNAS system running the Xeon D-1541 board; and the 2x16GB went into the C3758 system that I’m using for pure testing. It made no difference in power consumption - in either system. In other words, both systems have the exact same power consumption with 2x8 or 2x16. The TrueNAS system (D-1541) is measured “in-situ”, i.e. during its role as our household NAS. For the test-only (C3758) system, I measure it for about 12 hours idle, and then another 12 hours under prime95.
I’m doing some more testing with RDIMM modules on the Denverton system now (all testing thus far has been with UDIMM).
Based on all the feedback in this thread (and thinking about this way too much!), I’m going to use the C3758 board for our home NAS. I agree, the D-1541 is more powerful and arguably future-proof. But even the C3758 is overkill right now, and I really don’t see any likely scenario where it won’t be able to keep up (the scenario I described above was very contrived).
An update on power consumption. With my TrueNAS system running in situ, final power draw numbers are (average over about five days):
X10SDV-TLN4F with either 2x8 or 2x16 UDIMM and onboard LAN disabled, about 60 Watts
A2SDi-8C-HLN4F with either RAM config and onboard LAN disabled, about 52 Watts
Besides the motherboard, my TrueNAS system has four 7200 RPM SATA helium drives, two SATA SSDs, a Solarflare SFN5152F 10gbe NIC, a fan on the CPU, and a single 120mm case fan.
Another difference I noticed is that the IPMI (i.e. system off but plugged in and IPMI active) on the Xeon D system seems to be higher, even though they are both AST2400.
Here is the data for the “isolation” testing I did for both motherboards. By “isolation”, I mean I used the same PSU, fans and chassis, and ran for 10 or more hours in idle or with prime95. Idle is a stock TrueNAS Dragonfish install running off an old SATA SSD.
Note the “U” or “D” in the RAM column signifies Unregistered (UDIMM) or Registered (RDIMM). At least with the particular UDIMMs I have, 2x8 vs 2x16 does not seem to make a difference with either motherboard; but the RDIMMs clearly draw more power. The “Net” column is for if the on-board network devices were enabled or disabled.
X10SDV-TLN4F (Xeon D-1541)
Load
RAM
Net
Hours
kWH
AvgWatts
IPMI
2x8 U
dis
26.80
0.190
7.1
Idle
2x8 U
dis
12.82
0.390
30.4
prime95
2x8 U
dis
23.15
1.830
79.0
Idle
1x32 R
dis
22.48
0.690
30.7
prime95
1x32 R
dis
13.23
0.960
72.5
Idle
2x32 R
dis
13.08
0.410
31.3
prime95
2x32 R
dis
10.97
0.920
83.9
Idle
4x32 R
dis
13.20
0.430
32.6
prime95
4x32 R
dis
10.62
0.960
90.4
A2SDi-8C-HLN4F (Atom C3758)
Load
RAM
Net
Hours
kWH
AvgWatts
IPMI
4x32 R
dis
11.87
0.040
3.4
Idle
2x8 U
en
10.53
0.220
20.9
prime95
2x8 U
en
12.65
0.430
34.0
Idle
2x8 U
dis
10.32
0.200
19.4
prime95
2x8 U
dis
27.93
0.890
31.9
Idle
2x16 U
dis
13.40
0.260
19.4
prime95
2x16 U
dis
21.30
0.670
31.5
Idle
1x32 R
dis
10.25
0.190
18.5
prime95
1x32 R
dis
10.58
0.320
30.2
Idle
2x32 R
dis
13.10
0.280
21.4
prime95
2x32 R
dis
10.83
0.360
33.2
Idle
4x32 R
dis
13.00
0.330
25.4
prime95
4x32 R
dis
11.07
0.410
37.0
One thing I didn’t realize until late in the isolation testing process: I didn’t control for the fan auto-control. I didn’t do any measurements, but it was quite obvious that, when running prime95, the Xeon D system ramped up the fans dramatically. This makes intuitive sense why this happens, but I’ve found that fans can use a surprising amount of power. And these are three 40mm fans in a 1u Supermicro chassis.
, I’ll put the 2x8 modules in the D-1541 system this weekend so we can see what that does to power draw.