Lab quotas
Each lab starts with a budget (its size or its custom resources), and all of the Client's labs share the total capacity, especially GPU memory.
What the platform governs
- Offered sizes and default size. Saptiva AI defines them for your Client; members choose among them.
- Maximum per resource in custom mode (CPU, memory, VRAM, disk). It is shown in the form.
- Allowed disk classes. Only the default storage.
- Number of objects per lab (notebooks, volumes).
- GPU. A lab receives at most the VRAM of its size; jobs that request more are queued or rejected.
An example with numbers
A Client with 8 GPUs of 95 GB (760 GB of GPU memory) and sizes S (12 GB of VRAM), M (24 GB), and L (48 GB):
| Situation | Reserved | Available | What happens |
|---|---|---|---|
| Three models in service (200 GB) and nothing else | 200 GB | 560 GB | Any size is created right away. |
| Plus 10 M labs running | 440 GB | 320 GB | Six L labs still fit. |
| Plus 6 L labs | 728 GB | 32 GB | The next M lab is created; the next L is rejected with "No hay capacidad disponible" (No capacity available). |
| Same, but 8 of the M labs have their notebook stopped | 728 GB reserved, 250 GB in use | 32 GB | It still does not fit: the reservation counts even if it is not used. Ask those owners to delete the lab or move down a size. |
The key takeaway: stopping a notebook frees the GPU inside the lab, but the lab's size stays reserved for the Client. Only deleting the lab or changing it to a smaller size returns capacity to the dashboard.
What the member sees when they exceed it
The four messages are in Sizes and quota. If a member reports one, the cause is usually a running notebook consuming the GPU of the same lab.
What the platform does not protect
The platform sets the disk class and the resource caps, but lets the user choose the volume size, the mount path, and which volumes to delete. Two consequences worth telling new teams:
- A notebook with the volume mounted outside
/home/jovyanloses its work on the first restart (The notebook lost its files on restart). There is no way to recover it afterward. - Deleting a lab destroys its volumes immediately; there is no trash or retention.
- A lab created with custom resources and empty VRAM cannot use a GPU; one created at the exact maximum of CPU or memory does not start a notebook of that same size, because the container carries a fixed overhead that is charged to the same budget (case seen on 2026-08-24 with 64 GB: only 63 were usable). The full list is in Advanced configuration.
Before scheduled maintenance that restarts nodes, notify lab owners in advance so they save to the volume anything they have outside it.
How to free capacity
- In Panel (Dashboard), compare Reservada (Reserved) with actual usage.
- Identify Listo (Ready) labs with running notebooks and no activity (last activity in Accesos helps).
- Ask their owners to stop notebooks or delete test labs.
To expand the total capacity, the sizes, or the maximum per resource, contact Saptiva. The maximum per resource is adjusted per Client without reinstalling anything; ask for it to include the per-notebook overhead (about 0.5 CPU and 1.5 GB above the largest notebook you want to allow) so that the value the user sees in Máximo (Maximum) is actually usable.