Just wanted to mention, I’m also working on an exterior GPS antenna installation currently… I have all the parts here and have just not had the time (or the helping hand) to get everything mounted up!
Thanks for relaying all this info, I hope I’ll also have some bits that are helpful when I finally get around to documenting my install!
drive a rod into the ground and run a wire like they do
A lot of parts have started to come. Notably the Raspberry Pi CM5 Board as well as the low loss cables.
Unfortunately the star of the show, the GPS dev board and GPS antenna are still being cleared by the customs people and the courier I have chosen only delivers every Friday 
It seems I have the weekend and Tuesday to read all the pertinent manuals. Gonna read materials like I have major exams.
I cannot believe that of all people, my mother in law would recommend me for this (she is not very nerdy):
I should grab my family and head to the beach this evening and do sight seeing and try to identify them:
Of course Venus is only visible near sunset. Good luck with Uranus and Neptune without a real telescope (We only have a 8x42 binoculars and no tripod mount).
I still have that Stellarium App in my phone. Thanks lola for reminding me.
So we just got home from the sea side.
The immediate light pollution is a bit annoying, which is a shame because Im certain we have a reasonably low Bortle score. We need a place nearby that is a bit darker.
So the spouse used the binoculars and immediately complained that she couldnt see… Long story short, I got clearance to get better equiptment for viewing celestial phenomenon.
Nice 
We did see Venus, Saturn, Mars and Jupiter. I didn’t really try to find Neptune and Uranus because I assumed they’re significantly dim enough.
Its here now but I am having trouble looking for the time to actually mess around with it.
Currently have just a few more items on the way: an antenna grounding plane and a correct mounting solution. Its currently loose but its not really going anywhere for now.
The blinking light seem to indicate that I am receiving the GNSS signals. Right now I still cannot figure out how to pass the information from the Qwiic I2C to the raspberry Pi.
The smaller red board is a dedicated RTC clock that is supposed to be more accurate than the built in (and apparently not a real?) one and I also need to figure it out.
Some few updates:
Busy AF. Dunno why. Time Famine is getting real.
Ok so my Celestron Nexstar 6E Telescope has recently arrived and assembled but its cloudy rainy. Also I have to RTFM.
I still haven’t advanced progress in the GNSS time sync NTP server. I was too busy messing around with LLMs and now I am a bit obsessed how to increase the token generation but the unsloth DeepSeek-R1-Distill-Qwen-32B seems to be at the limit of usability at about 5 tokens/sec
Its on my gaming PC with Bazzite on it. No more games, but more LLMs. I wish to share this with the wife.
The highest I can run, though at a token per second is the DeepSeek-R1-Distill-LLaMa-70B. I am currently investigating on how to use this on a Ubuntu as what Wendell recommended on the post:
So where was I?
Yeah messing around with LLMs to help me build the GNSS time sync.
Now I dont have a gaming PC anymore…
I’ll probably need to game on my daily driver work laptop. At least it still has a laptop grade 3060 on it. Or maybe just game on the Steam Deck as Gaben intended.
So I reformatted the gaming PC to hopefully provide AI services network-wide (and beyond, thanks to Tailscale)
So I studied Wendell’s last year’s guide and I eventually got there…
I wasted so much bandwidth and so much disk writes here is where I think I got it right:
hg.co/unsloth repo and it is downloading finerm, rmi and purge commands./etc/systemd/system/ollama.service to addEnvironment=“OLLAMA_HOST=0.0.0.0:11434”
i think this was the crucial step. I was able to move forward after incorporating this step
For the long time I was stuck with an error saying that the TCP port I was using to run open-webui was already bound and I incorrectly assumed it was because I was doing something wrong with docker resulting to open-webui not being able to grab the models from the locally installed Ollama.
I am tempted to include some sort of image generation and some sort of voice assistant function but it sort of mish mash right now and includes a lot of messing around in the terminal and that feels like its a bit brittle right now with a lot of required python libraries to install.
Speaking of libraries to install, I saw in the guide that Wendell asked for PyTorch to be installed but it seems that ollama took care of some of that. I expected some sort of improved performance with text generation but in my last few installs, adding PyTorch does not seem to improve AI chat/text generation.
I watched and subscribed to NetworkChuck, an annoying guy for sure and he has that annoying Chad version of Linus vibe, but he does have some interesting content and I do learn from him.
Couldn’t portray it better.
I just want him to tone it down several notches.
I got a uConsole from Clockwork Pi. I need a device I can comfortably type and do SSH from the comfort of my bed if I can be bothered to stay awake
Also I reinstalled Proxmox to split the access to the nice GPU for gaming and AI/LLMs mainly because the kid was sad we cannot play anymore. Ollama stayed in the gaming PC and I split the OpenWebUI to the TrueNAS. It works and now I can talk to my AI from anywhere for as long as I have my Tailnet with me.
Speaking of Tailnets, I need to figure out how to install Tailscale direct to Proxmox so that I can turn on the VM with the AI remotely if I needed to. But IIRC this is not a supported use case and things might break for Proxmox but this isnt really an enterprise install and I can eff around to find out if I can.
Now I just need to add Wake on LAN to my pfSense to wake up the server.
I also checked Huggingface for the new AI hotness and it is Perplexity AI releasing a Tuned/Trained DeepSeek model with no Chinese censorship. 
I just watched Jeff Geerling put NUT on his Raspberry PI (His words, not mine )
This is perfect!
WoL to turn things on and NUT to turn things off!
More time famine. I brought the car to the dealership because I needed to replace the entire keyway because I lost my car key. Oopsie. Better spend a few dollars for safety and replace the whole damn keyways (Ignition, 2 Doors and Trunk) rather than save money and lose the whole car.
Also car dealership in the provinces is bad. Quite the polar opposite of the capital. Back there I got random upsells for things I dont need and here is the opposite, they will delay all the management they can even if it is right schedule for replacing for some reason. I dont get how they dont scale out here in the nearby urban hub.
Monster Hunter Wilds is coming but I wont be buying yet because I tried their benchmark demo and it crashed the Nobara install… its a pass until it gets a few patches…
So I bought Space Marines 2 instead…
Ah the struggle of choosing homelabbing, gaming or sleep.
I’m joking, enjoy the game!
Bro the little kid wont let me play at night.
This is the reality of it but its worth every sleepless night if I can make the little one giggle every now and then. The younger one looks at me with extra sas unlike the older one.
Ah bro the promo Crucial T705s are running out on Amazon now.
I think I should get one especially I couldn’t buy that stupid new Framework not-a-desktop with AI.
Do you need it? I’m still on skylake+++++++++ with PCIe 3.0 and am fine for home applications.
Sorry, I meant „you must consoom” 
I really feel the FOMO on that stupid machine. I really shouldn’t because I’ve seen the link where they listed the T705 on the forum, while I can’t really figure out most of the information it wants to convey, it seems like it cannot take full advantage of the 14,000 MB/s sec so this is a very suboptimal setup…
The root cause are social media. And tech channels.
And this forum