Sizes and quota
A size is a predefined resource package on five axes: CPU, memory, GPU cards, GPU memory per card and disk. If you do not know the notation (500m, 16Gi, 24576) or how the GPU is divided, start with How a lab works under the hood. Your installation publishes the available sizes and their values; Studio shows the ones Frida offers for your Client and a default size. You can also request Recursos personalizados (Custom resources), within a maximum cap per resource that Studio shows in each field: "Máximo: ..." (Maximum: ...).
Sizes in the reference installation (yours may vary):
| Size | CPU | Memory | VRAM | Disk | Intended for |
|---|---|---|---|---|---|
| S | 2 | 6 GB | 2 GB | 50 GB | Exploring data, notebooks without a GPU, tests with a small model. |
| M | 5 | 18 GB | 12 GB | 50 GB | Training medium models, inference with quantized models of up to about 7B. |
| L | 9 | 34 GB | 24 GB | 100 GB | Fine-tuning, vision models, several notebooks at once. |
The lab itself uses part of that budget (file browser and per-notebook overhead); the recommended notebook for each size is in Advanced configuration.
The size is not final: the owner can change it later, raise or lower each axis, or switch to custom resources (Resize a lab). The only axis that cannot go down is disk.
The size is a budget, not a per-notebook reservation: everything you launch inside the lab (notebooks, pipeline jobs, training jobs) shares those caps. If you share the lab, editors also consume from the same budget.
An example
Size M with 5 CPU, 18 GB of memory, 12 GB of VRAM and 50 GB of disk:
| What you launch | CPU | Memory | VRAM | Fits |
|---|---|---|---|---|
| Notebook with 1 GPU (2 CPU, 8 GB) | 2 / 5 | 8 / 18 | 12 / 12 | Yes |
| In addition, a training job with 1 GPU | 4 / 5 | 16 / 18 | 24 / 12 | No: it waits in the queue until you stop the notebook. |
| Instead, a CPU-only pipeline (1 CPU, 2 GB) | 3 / 5 | 10 / 18 | 12 / 12 | Yes |
The chips on the lab card in Studio show exactly these sums (usage / cap).
When something does not fit
A notebook or job that requests more than what is left in the budget does not start, and Studio shows the reason on the lab card under "Tu laboratorio rechazó una petición" (Your lab rejected a request):
| Message | What it means | What to do |
|---|---|---|
| "El notebook pedía más de lo que permite tu talla." (The notebook requested more than your size allows.) | It exceeds the available CPU, memory or GPU. | Lower the notebook's resources or stop another one. |
| "Tu laboratorio ya tiene tantos objetos de este tipo como permite su presupuesto." (Your lab already has as many objects of this type as its budget allows.) | Limit on the number of notebooks, disks or other objects. | Delete the ones you do not use. |
| "El disco pedido es de una clase que no se permite en los laboratorios." (The requested disk is of a class that is not allowed in labs.) | An unauthorized storage class was requested. | Use the default storage. |
| "La GPU pedida no cabe en el presupuesto de tu talla o queda fuera del grupo asignado a laboratorios." (The requested GPU does not fit in your size's budget or is outside the group assigned to labs.) | More GPU than allowed, or an unavailable type, was requested. | Request a GPU within your assigned VRAM. |
Administrators see the Client's total capacity in the Infrastructure dashboard.