photo pending
edge / router
Edge / router- cpu
- Intel Core i5-6500 · 4C/4T
- ram
- 16 GB
- net
- 10G
- os
- OPNsense
runs
- OPNsense
- TLS termination
- routing for every service on this page
The Homelab
Every service here runs on hardware in my house.
No cloud instances, no managed databases, no API keys to someone else's models. News ingestion, search indexes, embeddings, reranking and LLM inference all run on these machines.
how it fits together →globirie api · on snow
checking… · feed + stories
globe frontend · on snow
checking… · cesium embed
jina-search · on really
checking… · embeddings + rerank
checked live from your browser just now: a lit dot is a request that actually succeeded
photo pending
storage
Storage node- cpu
- Intel Core i5-9400 · 6C/6T
- ram
- 32 GB
- disk
- 2.5 TB SSD: Samsung 990 PRO 2 TB + 860 500 GB
- net
- 10G
- os
- Ubuntu 24.04 LTS
runs
- bulk storage
- backups
This one has stayed up for more than ten weeks and holds the rest of the lab's data. It mostly just works.
photo pending
really
Compute / retrieval node- cpu
- AMD Ryzen 7 3700X · 8C/16T
- ram
- 64 GB
- gpu
- NVIDIA RTX 2080 Ti · 11 GB
- disk
- 2 TB NVMe: Samsung 990 PRO
- net
- 10G
- os
- Ubuntu 24.04 LTS
runs
- jina-search: embeddings + listwise reranking
- project services
- background workloads
photo pending
snow
Controller- cpu
- Intel Core i3-13100 · 4C/8T
- ram
- 32 GB
- gpu
- Intel Arc A380
- disk
- 1.5 TB SSD: 970 EVO Plus + Bestoss 512 GB + 860 480 GB
- net
- 2.5G
- os
- Ubuntu 24.04 LTS
runs
- Globirie: ingestion, API, globe frontend
- this website: Caddy + static build
- service orchestration
- transcode / light inference
Globirie runs here, then sends summarisation, embedding and reranking work to the GPU boxes over the LAN.
photo pending
furnace
GPU / AI node- cpu
- AMD Ryzen 7 3800X · 8C/16T
- ram
- 64 GB
- gpu
- NVIDIA RTX 4070 Ti SUPER 16 GB + RTX 4070 SUPER 12 GB
- disk
- 1.6 TB: Samsung 9100 PRO 1 TB + Crucial P1 500 GB
- net
- 1G
- os
- Ubuntu 26.04 LTS
runs
- local model inference
- AI workloads
Two Ada cards, 28 GB of VRAM between them, where the heavier local models actually run.
photo pending
xkdesktop
Gaming, AI when idle- cpu
- AMD Ryzen 5 7600 · 6C/12T
- ram
- 32 GB
- gpu
- NVIDIA RTX 3080 · 10 GB
- disk
- 3.2 TB: Samsung 9100 PRO 2 TB + Toshiba 1 TB HDD + 512 GB NVMe
- net
- 1G
- os
- Windows 11 Pro
runs
- games, honestly
- spare inference capacity when not gaming