GPU Pods — ZorgDeepCore
ZorgDeepCore is ZorgSocial's infrastructure layer that lets you rent powerful GPU servers for running intensive AI workloads — directly from within the platform.
What Are GPU Pods?
A GPU Pod is a virtual machine equipped with a dedicated GPU that you can spin up on demand. Use cases include:
- Running advanced AI image/video generation workflows (ComfyUI).
- Installing and running large generative models not available in the standard Media Studio.
- Any GPU-intensive computing task related to content creation or AI work.
Why Use ZorgDeepCore Instead of Media Studio?
| Media Studio | ZorgDeepCore GPU Pods |
|---|---|
| Simple, guided interface | Full control over the environment |
| Pre-built generation tools | Install any model you need |
| No technical knowledge needed | Requires technical familiarity with AI tools |
| Credits consumed per generation | Credits consumed per minute of uptime |
Use Media Studio for standard content creation. Use ZorgDeepCore when you need advanced, custom AI workflows.
Getting to GPU Pods
Click ZorgDeepCore (the chip icon) in the sidebar.
GPU Pools
ZorgSocial sources GPU capacity from an infrastructure pool called Pool 2. The page shows three summary cards at the top:
- Total Pods — all pods you own.
- Active — pods currently running.
- Stopped — pods that are stopped or terminated.
A Refresh button in the top-right reloads the list. The list also auto-refreshes every 30 seconds.
A Show All Stopped toggle controls whether stopped/terminated instances appear in the list.
Searching for a GPU
The Search GPU Offers panel lets you find available machines to launch:
- Use the GPU Picker to filter by GPU model name(s) (e.g., RTX 4090, A100, H100).
- Click Search to fetch live results.
- Results appear in a table with these columns:
| Column | Description |
|---|---|
| GPU | GPU model name |
| Count | Number of GPUs on the instance |
| VRAM | GPU memory in GB |
| Credits/hr | Cost in platform credits per hour |
| Disk | Included disk space in GB |
| Location | Geographic region |
| Reliability | Uptime reliability score (green ≥ 99%, yellow ≥ 95%, red < 95%) |
| Action | Launch button |
Launching a Pod
- Click Launch next to the offer you want.
- A dialog confirms the selected GPU, count, cost, and location.
- Fill in:
- Instance Label (optional) — a human-readable name.
- Disk Size (GB) — storage to attach (default 50 GB).
- Click Launch.
The pod appears in your list with Creating status. Once provisioned it moves to Running.
⚠️ If the offer expires before you confirm, the list automatically refreshes and you can pick a new one.
Pod Cards
Each pod in the list shows:
- Name and Pool badge.
- GPU model and credits/hr rate.
- Status badge — Creating · Running · Stopped · Terminated.
- Public IP (when running).
- Age — how long ago the pod was created.
- ComfyUI status — "Ready" (green ✓) or "Starting" (spinner).
- ComfyUI URL — a direct link when available.
Pod actions
| Button | When visible | Action |
|---|---|---|
| Details | Always | Opens the pod detail page |
| Open ComfyUI | Running + ComfyUI ready | Opens the ComfyUI web interface in a new tab |
| Reboot | Running | Reboots the instance |
| Destroy | Always | Permanently destroys the pod (confirmation required) |
Pod Detail Page
Click Details on any pod card to open its full detail page /gpu-pods/[instanceId]).
Header actions
- Refresh — reload instance data.
- Try Models — opens the [Model Playground](./02-model-playground.md) (only when ComfyUI is ready).
- Reboot — restart the instance.
- Destroy Instance — permanently destroy (confirmation required).
Info cards
| Card | Contents |
|---|---|
| Status | Status badge, ComfyUI readiness, direct ComfyUI link |
| GPU | GPU model, number of GPUs, VRAM, credits/hr rate |
| Network | Public IP, ComfyUI port, disk size |
Billing card
Shows live billing data while the pod is running:
- Elapsed time — live HH:MM:SS counter since billing started.
- Credits used — running total, updated every second.
- Rate — credits per second and credits per hour.
- Live badge (green) when billing is active and up-to-date.
- Stale badge (yellow) if the live data hasn't been updated recently.
- Balance Depleted badge (red) if the pod was stopped due to insufficient credits.
Installing Models
The Model Installations section lets you download AI models directly onto a running pod with one click. Models require the pod to be running and ComfyUI to be ready.
| Model | Description | Approx. Size |
|---|---|---|
| Z-Image Turbo | Qwen 3 4B text encoder, AE VAE, Z-Image Turbo BF16 diffusion model | ~8 GB |
| LTX-2 | 19B fp8 checkpoint, Gemma 3 text encoder, spatial upscaler, distilled LoRA, camera control LoRA | ~25 GB |
| LTX-2.3 | 22B checkpoint, Gemma 3 12B encoder, spatial upscaler v1.1, distilled & abliterated LoRAs | ~30 GB |
| Wan 2.2 | UMT5-XXL encoder, VAE, high & low noise 14B diffusion models, LightX2V 4-step LoRA | ~30 GB |
| InfiniteTalk | Wan 2.1 i2v 14B, LightX2V LoRA, InfiniteTalk weights, VAE, CLIP Vision, wav2vec2, UMT5-XXL | ~45 GB |
Each install button streams live progress output to the terminal panel below.
Terminal & Container Logs
The bottom of the detail page has a two-tab panel:
SSH Terminal tab
An in-browser terminal for running shell commands on the pod via SSH:
- Type a command in the input field and press Enter or click Run.
- Output streams back in real time.
- Clear button removes all log history.
- Previously queued commands are automatically restored when you reopen the detail page.
Container Logs tab
Displays the Docker container logs (last 1 000 lines):
- Auto-refreshes every 10 seconds while the pod is running.
- Pause / Resume button to pause auto-refresh.
- Refresh button to fetch logs immediately.
Billing for GPU Pods
GPU pods consume platform credits per minute of uptime based on the GPU type and count. Credits are deducted continuously while the pod is running. If your credit balance runs out, the pod is automatically stopped.
You can monitor credit usage on the pod detail page's Billing card or in your [Wallet & Credits](../account/03-wallet-and-credits.md).
⚠️ Destroy — not just stop — your pod to fully end billing. Destroyed pods and their data cannot be recovered.