Find more information about the provisioning limitations, which are primarily determined by the physical hardware constraints.
These constraints are relevant to GPU instances that match the specific GPU model. It's important to keep in mind that for each GPU, at least one vCPU should be allocated. For example, if you request provisioning for 8 GPUs, you must have a minimum of 8 vCPUs allocated for the request to succeed.
|
GPU
|
Max GPUs per Instance
|
Max vCPU/GPU
|
Max vCPUs per Instance
|
Max RAM/vCPU
|
Max RAM per Instance
|
Network
|
|---|---|---|---|---|---|---|
| H100 80GB SXM | 8 | 48 | 128 | 48GB | 1024GB | 100Gbps |
| A100 80GB SXM | 8 | 30 | 94 | 48GB | 695GB | 100Gbps |
| A100 40GB SXM | 8 | 18 | 94 | 24GB | 492GB | 100Gbps |
| A100 80GB PCIE | 8 | 30 | 94 | 48GB | 695GB | 40Gbps |
| A100 40GB PCIE | 8 | 18 | 94 | 24GB | 492GB | 40Gbps |
| A40 | 8 | 40 | 94 | 48GB | 702GB | 40Gbps |
| RTX A6000 | 8 | 40 | 126 | 124GB | 712GB | 40Gbps |
| RTX A5000 | 8 | 30 | 122 | 48GB | 706GB | 40Gbps |
| RTX A4000 | 7 | 32 | 46 | 8GB | 366GB | 40Gbps |
| Tesla V100 NVLink | 8 | 8 | 60 | 120GB | 936GB | 10Gbps |
| Tesla V100 PCIE | 7 | 8 | 44 | 60GB | 240GB | 10Gbps |
| Quadro RTX 5000 | 4 | 24 | 30 | 120GB | 170GB | 10Gbps |
| Quadro RTX 4000 | 7 | 22 | 30 | 120GB | 240GB | 10Gbps |
These limitations are relevant to CPU-based instances and help us ensure availability for all of our customers.
|
CPU
|
Max vCPUs per Instance
|
RAM/vCPU Ratio
|
Max RAM per Instance
|
Network
|
|---|---|---|---|---|
| AMD EPYC Milan | 46 | 4GB | 176GB | 10Gbps |
| AMD EPYC Rome | 46 | 4GB | 176GB | 10Gbps |
| Intel Xeon Scalable | 94 | 4GB | 238GB | 10Gbps |
| Intel Xeon v4 | 60 | 4GB | 238GB | 10Gbps |
| Intel Xeon v3 | 70 | 4GB | 170GB | 10Gbps |