All Cytognosis compute is in us-central1, zone us-central1-b (required for ARM t2a instances).
| Name | Type | Zone | vCPU | RAM | External IP | Cost | Purpose |
|---|---|---|---|---|---|---|---|
| cytohost | t2a-standard-2 (ARM64) | us-central1-b | 2 | 8 GB | 136.111.39.188 | ~$45/mo | Core services stack + GitHub Actions runner |
unless-stopped):
| Service | URL | Image |
|---|---|---|
| Caddy (reverse proxy + TLS) | — | caddy:2-alpine (ARM64) |
| Cal.com | cal.cytognosis.org | calcom/cal.com:latest (x86, QEMU) |
| Cal.com PostgreSQL | — | postgres:15 (ARM64) |
| Excalidraw | whiteboard.cytognosis.org | excalidraw/excalidraw:latest (ARM64) |
| Excalidraw Room | — | excalidraw/excalidraw-room:latest (x86, QEMU) |
| Mermaid Live | mermaid.cytognosis.org | ghcr.io/mermaid-js/mermaid-live-editor:latest (ARM64) |
| Logseq | notes.cytognosis.org | ghcr.io/logseq/logseq-webapp:latest (ARM64) |
| MLflow | mlflow.cytognosis.org | ghcr.io/mlflow/mlflow:latest (ARM64) |
x86 via QEMU: Cal.com and Excalidraw Room lack ARM64 images; they run under QEMU
emulation (platform: linux/amd64 in their service YAML). Performance is adequate for
scheduling tools — not recommended for CPU-intensive workloads.
``bashVia IAP (no public SSH port needed)
gcloud compute ssh cytohost \
--zone=us-central1-b \
--project=cytognosis-infrastructure \
--tunnel-through-iap
On-Demand Research Compute Tiers
Spin up as spot instances when needed; terminate immediately after. All in
us-central1-b.
| Profile | Type | vCPU | RAM | VRAM | Est. spot cost | Use case |
|---|---|---|---|---|---|---|
| Neo4j always-on |
e2-standard-2 | 2 | 8 GB | — | ~$49/mo | Knowledge graph + OLS4 |
| CPU-heavy ETL | n2-standard-8 spot | 8 | 32 GB | — | ~$0.10/hr | BioCypher, TileDB ingest |
| Memory-heavy | n2-highmem-8 spot | 8 | 64 GB | — | ~$0.13/hr | Large AnnData, SOMA |
| GPU small | n1-standard-4 + T4 spot | 4 | 15 GB | 16 GB | ~$0.22/hr | Embedding, small inference |
| GPU medium | a2-highgpu-1g spot | 12 | 85 GB | 40 GB | ~$0.90/hr | A100 — fine-tuning |
| GPU large | a3-highgpu-1g spot | 26 | 170 GB | 80 GB | ~$1.80/hr | H100 — foundation models |
Naming Convention
Persistent nodes follow the pattern
cytonode-NN:
cytohost — current core services node (ARM64, us-central1-b)
cytonode-02 — future second node if needed (use same t2a-standard-2 type)
Ephemeral research nodes are named
research- (auto-deleted when done).
All new persistent compute goes in us-central1-b` (t2a ARM64 availability zone).