'Tis the season. Spending some family time on the weekend decorating the tree

Here's a breakdown of the lab power consumption based on the data I currently have.

Unidentified UPS loads (total UPS power output minus load attributed to individual managed PDU outputs) is 2.25 kW or 43% of total. This is mostly stuff on the main T&M bench which is not metered yet, but also includes my desk and workstation in the office which is also UPS backed. This category will shrink over time as I gradually roll out metering to more loads.

Cooling (non-UPS) is the single biggest identified line item, 1.2 kW or 23% of the total. I hope to reduce the sum of lab and house cooling through energy harvesting over the next year or two (heating the house in the winter with waste lab heat)

Next largest is 570W of "networking" loads. This includes ~30W of monitors that I will probably break out into their own category, the 100G core router/firewall, the PoE injector, the 40/100G switching, and the legacy 1G switching. I expect to cut quite a bit off this in the coming months deploying more modern 1G switches to replace the legacy ones to save energy before my own FPGA-based switch is done.

Next is 555W of virtualization loads. This includes the main xcp-ng server on the IT rack, the workstation-turned-into-second-VM-server on the microscope bench, the NUC used for testing ngscopeclient with Intel integrated graphics, and the Mac Mini also used for ngscopeclient CI. I want to try to find a way to optimize for power here, but it's tricky because I can't consolidate the servers - both have PCIe passthrough devices on them so I can't migrate all the VMs to one box and shut down the other, and both have physical cable constraints to other lab equipment preventing them from moving.

Next is 320W for the Ceph storage cluster. This honestly isn't bad for three 2U servers with Xeon CPUs, 96GB of RAM, 100G NICs, etc. So I'm probably not going to do much on this front.

Next is 229W for my friend's backup server. This is a single box pulling almost as much juice as my whole storage cluster so I'd like to try to find a way to reduce its demand at some point.

Finally, there's 53W of various FPGA/MCU devkits installed in long-term test setups and 23W of T&M equipment (ThunderScope and some power supplies) accounted for so far. I expect the T&M category to grow massively once I start metering the main electronics bench. But the stuff I have instrumented so far isn't worth putting any effort into scaling back on.

This is about $2320/yr in power at today's prices just to keep the lab from melting.

Hopefully once I revamp the HVAC in the house I can set up a VRF heat pump to heat the house in the winter with waste heat from the lab and save a bunch on this.

Also interesting: the total power draw at the UPS output is a touch under 4 kW, adding in cooling gives roughly 5.2 kW attributable to the lab. The house as a whole is averaging something like 5.6 kW, meaning I have only ~400W (this time of year at least) in non-lab consumption - not bad at all.

The Fluke current clamp meter probe I ordered came in. Preliminary single-point measurement of the lab mini-split is showing roughly 5.1A at 240V, or 1.22 kW, draw for cooling.

This is of course a one-time measurement and I don't know how much higher or lower it gets over time with changes in weather etc. But that's a fair bit of juice 24/7.

To all the folks talking about how horrible being rude and abusive to a slopbot is:

Clippy wants to know where you were 25 years ago when he needed you.

Finished the layout for the 4x 2:1 relay board. Still have to do cutouts in the ground planes under the relays then add some artwork and revision ID text etc but that should be relatively quick.

This is intended to drive one ThunderScope from eight different signal sources on the probe demo/test board.

Member of technical staph
Staph infection
Senior staph infection
Principal staph infection

I have too many projects I need like 3x as many jtag dongles as i have lol

I wish I had a friend named Claude who was an engineer or software dev.

We could have so much fun trolling people.

My immediate use is rack level automated unit testing of ThunderScope from a variety of signal generators so I only *need* 500 MHz or so of reasonably flat performance. But I'll happily take more and if it's faster I can use it with future hardware as well.

BOM for the changeover version is looking like $30 of SMAs, $88 of relays, plus the PCB (just JLC S1000-2M, nothing fancy at these frequencies), some kind of enclosure TBD, a little STM32, and some support passives.

If I add two more relays per channel to support termination, I would expect the terminated version to cost roughly an additional $175.

Working on building a low cost, open hardware RF relay system.

Four 2:1 channels using Axicom HF3 series latching RF relays. USB-C for power plus SCPI control.

The initial version will be a simple changeover design (i.e. the unselected port will float, rather than being terminated). I may look at making a version that terminates in the future, but that will double or triple the cost since it'll need more relays and the stuff I'm working with right now is fine driving an open-circuit load without damage.

The relays I'm using are specified from DC to 3 GHz (no blocking capacitors) and I'll be characterizing end-to-end performance of the whole system on my PicoVNA to 8.5 GHz.

Does this sound interesting to people? Would a version with switchable termination be worth designing in the future?

Also just ordered another managed PDU, planning to continue the rollout. Over the next year or two I am planning to try and get energy consumption data across basically the entire lab so i can better target optimization efforts.

Yep, summer is over. Just got a $889.30 power bill for 4.077 MWh for an effective price of $0.218/kWh.

Solar production for last month was only 0.77 MWh and it's gonna keep falling for the rest of the year.

This comes out to an average continuous consumption of about 5.66 kW. Of this, about 1.5 are accounted for by the managed PDUs on the lab IT racks, which now feed all of the servers, switches, and devkits. (Breaking down further: 334W for storage cluster, 355W for VM servers, 559W for routers/switches, 210W for my friend's backup server, 47W for devkits).

An additional ~2.5 kW is used by the remainder of the lab, mostly T&M equipment, but is not accounted for in detail since I don't yet have managed PDUs on the other work benches.

The 559W of networking gear is quite a bit higher than I would like (costing just over $1K/yr in power). The cheap FS switch I replaced a failed legacy Cisco with is pulling 14W compared to 79-95W for the Ciscos depending on how many ports are up.

I'm very tempted to buy a couple more of the FS switches and retire the Ciscos later this year, rather than waiting for LATENTRED to be finished (replacing them with LATENTRED had always been the plan, but is taking its sweet time as I work on higher priority stuff).

I still have three of the old Ciscos in service. Preliminary calculations say that each one I replace will save roughly 575 kWh and $125 in power costs every year. The FS switches cost $425 after taxes and shipping, so that's about a 3.4 year payoff period assuming I depreciate the switches completely after that period. If I resell the switches after I replace them with my own designs, the effective cost of each will drop.

Does anyone sell specifically Bessel LPF filters (exact order doesn't matter but maybe 4th would be decent) in SMA connectorized form factor?

Chebyshev, Butterworth, etc are in common use. But I need linear phase response

There's an actual decapped PIC12F683 on display under a magnifying glass, along with a bare die and fully packaged sample.

And of course nice high res optical micrographs since it's still tiny even with the magnifier

If anybody is in Singapore in the next few months, check out the ArtScience museum. They have an exhibit on open source silicon that I helped out a bit with. Tiny Tapeout and a bunch of other folks participated too.

I haven't seen it in person but the museum sent me a few photos of the setup.

The existence of ICANN implies the existence of ICANT

If you hear "GRC" without context you think...

not only is it indefinitely backordered there's also a ~50% tariff