HP ML350 G5 Power Measurements

Front_267x400

The Thrifty & Thirsty NAS

I recently commissioned a new personal NAS using a frankensteined (this seems to be a trend with me…) HP ML350 G5. This generation of the ML350 is the HP version of the Dell PowerEdge 2900, or the HP DL360 G5 / Dell PowerEdge 2950, and all these servers have a reputation for destroying your power bill. The G5 I acquired was a base model that originally came equipped with a single dual core Xeon E5205 – but I had a pair of quad core Xeons so I decided to see how they affected energy consumption. Armed with my trusty kill-a-watt (seriously, it’s worth the $20), I arrived at the following measurements:

 

 

  • Base: 1x Intel Xeon E5205 2GHz dual core, 32 GB (8x 4GB) RAM, one PSU, no drives: 223 watts when sitting in the BIOS.
  • Powered off: 34 watts (iLO).
  • Upgrade to single quad core: 1x E5430 2.66 GHz quad core Xeon, 32 GB RAM, one PSU, no drives: 223 watts sitting in the BIOS.
  • Upgrade to second power regulator, but still single CPU. Everything else identical. 223 watts.
  • Upgrade to dual quad core Xeons: 2x 5430s, everything else identical: 230 watts in the BIOS.
  • Upgraded BIOS to recognize new CPUs: 232 watts in the BIOS. I consider this within the margin of error.
  • Removed “onboard” RAID card: 232 watts.

Lessons Learned

First of all, the ML350 is a capable space heater. These suckers draw a lot of power even in base configurations. Secondly, the Core 2 based Xeons are clearly capable of power saving when idle. I was shocked the quad core drew exactly the same amount of power as the dual core Xeon. Never in a million years would I expect this result… but I confirmed this to be the case and assure you that the results are true. I was again shocked that the second quad core Xeon only drew an additional seven (7) watts of power! For seven watts, I was able to more than quadruple the computational ability of this system! Clearly, this would not be the case under load, but I see zero reason to run lower spec processors to save power; add the extra cores and CPUs, then consume the power only when needed. Another surprise was how much power drives consume. I don’t have exact numbers, but with two Seagate enterprise drives installed and a third consumer drive installed, power consumption climbed to 280 watts! This was enough to overload my poor APC Back-UPS Pro 420, so I ordered a CyberPower CP1500PFCLCD 1500VA / 900W “pure sinewave” UPS to support this machine. Each drive appears to add at least 10 watts at the wall – extra consumption is likely due to the addition of an IBM Serveraid M1115 SAS/SATA HBA (flashed to IT mode with LSI firmware) and the HP SAS backplane.

Furthermore, I learned some other interesting facts regarding the HP ML350 G5s:

  • Attempting to boot dual CPUs without the extra power regulator is a no-no. Not only does it not work, reverting back to a single CPU doesn’t appear to work either. Only when I re-installed the power regulator was the system able to boot again – on a single CPU. Once the system successfully POSTed with the single CPU, it was able to POST again without the second power regulator.
  • I saw a note on HP’s website indicating a BIOS version was fixed to allow the ML350 G5 to boot with processors that had different steppings. While it is common to order and install identical / matched processors, it is clearly not an absolute requirement. What I can say is that installing a dual core Xeon and a quad core Xeon doesn’t work. Not that I expected it to!
  • CPUs not explicitly recognized by the BIOS will POST, but not boot into an OS. The BIOS will only allow booting into HP’s ROMPaq utility which, to be honest, is a PITA. Since I was using a USB flash drive, I actually reinstalled the default CPU, upgraded the BIOS firmware, then reinstalled the E5430s.
  • The ML350 G5 is picky about which PCI-E slot HBAs (or maybe just RAID controllers) are installed in. Once again, it would POST, but not boot into an OS when the IBM RAID controller was installed in any slot except the bottommost PCI-E slot.
  • The RAM gets HOT! While I didn’t test it, I think it’s clear that the RAM is a significant contributor to the power draw. Be sure to install the airflow baffles too!

In the end, I deployed this system as a FreeNAS server with both QC Xeons installed. The system is snappy and while I am not currently using the extra CPU power, it’s nice to have it available – especially since FreeNAS supports using jails & virtualization.