Post B6apImXVPcuaJTtZRo by azonenberg@ioc.exchange
(DIR) More posts by azonenberg@ioc.exchange
(DIR) Post #B6aiIrof5uLL020tcG by azonenberg@ioc.exchange
0 likes, 0 repeats
Some interesting data on lab power consumption / energy efficiency: shut down the microscope bench workstation yesterday evening while troubleshooting some stuff and left it down overnight.This is a Xeon 8362, 128GB of RAM, and a (small) nvidia card, for reference.Looks like it was idling around ~100W, or ~2.5% of my total consumption. Not nothing, but also not huge compared to e.g. the T&M equipment or some of the bigger iron.But good to have some actual measured data on base consumption of some of these workstation platforms.
(DIR) Post #B6aj2w7tOwi9giya6S by azonenberg@ioc.exchange
0 likes, 0 repeats
(these numbers are pretty rough because the magnitude of the dip is close to the amount of noise in the data, so I'd say +/- 50% or so unless i did a lot of averaging on the raw measurements)Extrapolating slightly... 100W/day = 2.4 kWh/day = 52 cents a day in direct power costs.I don't know my actual PUE for the lab, although I can set an upper bound: my whole house had an average power consumption of around 150 kWh/day or 6.25 kW and the lab's consumption at the UPS output is a pretty steady 4 kW, so my PUE cannot be worse than 1.56.But this is an upper bound since it includes the "hotel load" from home HVAC, water heater, lighting, cooking, etc. The only things I really should be counting towards PUE that aren't on the UPS are lab lighting, lab cooling, and UPS conversion losses (plus the soldering ovens and other non-UPS-backed tools which are considered "productive work" but not metered on the UPS, but my usage of them this past month was pretty low so I'll ignore those).Using a somewhat guesstimated 1.2 PUE, the workstation has a net cost of ~120W or 2.9 kWh/day, or 62 cents a day at today's power prices ignoring solar generation. This comes out to $18.60 a month or $226 a year in operating costs - small compared to the overall lab budget, but not nothing and probably worth trying to eliminate if I can.
(DIR) Post #B6akG0EtWLdaAA2t9c by azonenberg@ioc.exchange
0 likes, 0 repeats
Also, this means my non-lab power consumption is bounded at 2.25 kW (this is the sum of hotel load and support power for the lab). But that data is from earlier this month when I had very little heating energy consumption.Going back to March when it was colder and I spent more power heating the house, on higher consumption (presumably colder) days I averaged... around 170.5 kWh/day which suggests my heating consumption is on the order of 20 kWh/day. This is far from exact because the data is noisy, plus lab cooling likely uses less power in cold weather (it's easier to reject heat to a colder sink).At some point I want to put some current clamps on the lab mini-split as well as the heat pump to the house, so I can quantify my average/peak energy usage for HVAC better.
(DIR) Post #B6akL0xcGZbhqdLyts by azonenberg@ioc.exchange
0 likes, 0 repeats
Ultimately what I would really like to do (before spending a ton of money on the project) is determine how much power I am spending cooling the lab, and how much heating the house, and how much I might save with a VRF heat recovery system.
(DIR) Post #B6aonGuIyE7HELXBhY by FritzAdalis@infosec.exchange
0 likes, 0 repeats
@azonenberg Do you already use power save features to cut power use when idle? For machines with ipmi you could probably script power on/off during builds but that adds some complexity.
(DIR) Post #B6apImXVPcuaJTtZRo by azonenberg@ioc.exchange
0 likes, 0 repeats
@FritzAdalis I use normal CPU frequency governors etc but nothing ever turns fully off.I usually have a lot of always-on stuff like jtag servers on various machines that make it annoying to turn them fully off (plus these supermicro mainboards take like 5+ minutes to POST).What I am trying to do instead is consolidate always-on stuff to as few machines as possible so I can shut down some of the workstations when not in use, or at least start using S3 suspend on them
(DIR) Post #B6apM97rtPaXEUVY0G by azonenberg@ioc.exchange
0 likes, 0 repeats
@FritzAdalis CI jobs run on the VM server which is never going down because it hosts my production DNS, web, etc. servers
(DIR) Post #B6apYqozGAhWE7jayW by FritzAdalis@infosec.exchange
0 likes, 0 repeats
@azonenberg Reasonable, thanks.
(DIR) Post #B6apik3jFxj89MLWYi by azonenberg@ioc.exchange
0 likes, 0 repeats
@FritzAdalis So far i've managed to shut down one machine whose purpose was entirely hosting jtag/uart servers by replacing it with a virtualized equivalent on the VM server using usb passthrough.The microscope bench box is planned to get the same treatment, then have the guts repurposed as extra big-iron compute that I can bring up on demand and shut down fully when not needed.But only if the microscope/camera software is happy running in a VM. I won't know until I test.
(DIR) Post #B6apvHrvNYYgU2o5Am by CliffsEsport@mastodon.social
0 likes, 0 repeats
@azonenberg just curious as homelaber, does the workstation need to be idling 24/7 with your workload? Or could you actually power it off for 6 to 12 hours a day? If your workflow allows for powering it off on regular cycle or something, that is easy savings.
(DIR) Post #B6aq0qRyEXn5BSJj7I by azonenberg@ioc.exchange
0 likes, 0 repeats
@CliffsEsport One of the ongoing projects I am trying to do is migrate always-on stuff to VMs on the Xen server and/or as few workstations as possible, so the remainder can be suspended or shut down when not in use.Figuring out which stuff can easily move or be virtualized takes time.
(DIR) Post #B6aq5yrUO39Yighb2e by azonenberg@ioc.exchange
0 likes, 0 repeats
@CliffsEsport But since the T&M gear never shuts down and is the dominant power user in the lab I suspect the marginal savings from suspending idle workstations will be small compared to e.g. if I were able to shut down the 16 GHz scope when not in use
(DIR) Post #B6aqIKlykPtC9Q10ng by azonenberg@ioc.exchange
0 likes, 0 repeats
@CliffsEsport like the VM server is my biggest big-iron box with half a TB of RAM and three GPUs and it eats ~400W under base load (and will never shut down since it runs 24/7 prod workloads).The microscope bench box is ~100 and is one I am hoping to be able to suspend.The 16 GHz scope is >600W and can never shut off.