You can request NVIDIA® GPUs optionally for your Workload.
Model | Kubernetes resource name | VRAM | NUMBER OF CUDA CORES |
NVIDIA® A100 | nvidia.com/gpu | 40GB | 13824 |
NVIDIA® A100 Half | nvidia.com/mig-3g.20gb | 20GB | 5925 |
NVIDIA® A100 Quarter | nvidia.com/mig-2g.10gb | 10GB | 3950 |
Puzl is the first cloud platform offering a fractional Nvidia GPU rent. Fractional devices are based on Nvidia MIG technology. MIG-based devices are fully isolated from the other users and processes, which means that you can use them like a regular GPU.
Fractional GPUs let you save the costs for the tasks when you don't need an excessive computing capacity. They are especially good at horizontal scaling of AI\ML inference.
Usage is billed on per-second basis. The amount of GPU-hours spent by the Customer is calculated to the nearest second. Standard Service Level Agreement applies .
Refer to a particular service to see price.