Supported Compute Shapes
Data Science supports specific Compute shapes for various resources in the service.
VM Compute shapes and GPU Compute shapes describe these shapes in detail.
OCPUs represent physical CPU cores. Most CPU architectures, including x86, run two threads per physical core, so one OCPU is the equal of two vCPUs for x86-based compute. The cost per vCPU is half the cost of an OCPU, see Data Science pricing.
AMD CPU
| Shape | OCPUs | GB Memory |
|---|---|---|
| VM.Standard.E3.Flex | 1–114 | 1–1776 (dependent on OCPU count) |
| VM.Standard.E4.Flex | 1–114 | 1–1760 (dependent on OCPU count) |
| VM.Standard.E5.Flex | 1–94 | 1–1049 (dependent on OCPU count) |
For notebook sessions, the maximum memory for a standard flexible shape is 64 GB per OCPU
up to the maximum allowed memory. For example, for two OCPUs, the maximum allowed memory
is 2 * 64 = 128 GB. For 20 OCPUs, 20 * 64 = 1280 GB > 1024
GB. Therefore, for 20 OCPUs, the maximum allowed memory for both
VM.Standard.E3. Flex and VM.Standard.E4.Flex is 1,024 GB.
Example:
| OCPU | GB Memory |
|---|---|
| 1 | 1-64 |
| 2 | 1-128 |
| 3 | 1-192 |
| 4 | 1–256 |
When you use standard flexible shapes for model deployments:
- The minimum memory is 6 GB per OCPU used for the model server and load.
- The maximum memory is 64 GB per OCPU up to the allowed memory for each shape. For example, maximum allowed memory for the VM.Standard.E3. Flex shape is
1,024 GB.
ARM CPU
| Shape | OCPUs | GB Memory |
|---|---|---|
| VM.Standard.A1.Flex | 1–80 | 1–512 (dependent on OCPU count) |
| VM.Standard.A2.Flex | 1–78 | 1–946 (dependent on OCPU count) |
For notebook sessions, the maximum memory for a standard flexible shape is 64 GB per OCPU
up to the maximum allowed memory (1024 GB). For example, for two OCPUs, the maximum
allowed memory is 2 * 64 = 128 GB. For 20 OCPUs, 20 * 64 = 1280
GB > 1024 GB.
| OCPU | GB Memory |
|---|---|
| 1 | 1-64 |
| 2 | 1-128 |
| 3 | 1-192 |
| 4 | 1–256 |
Intel CPU (X9)
| Shape | OCPUs | GB Memory |
|---|---|---|
| VM.Standard3.Flex | 1–32 | 1–512 (dependent on OCPU count) |
| VM.Optimized3.Flex | 1–18 | 1–256 (dependent on OCPU count) |
Intel CPU (X7)
| Shape | OCPUs | GB Memory |
|---|---|---|
| VM.Standard2.1 | 1 | 15 |
| VM.Standard2.2 | 2 | 30 |
| VM.Standard2.4 | 4 | 60 |
| VM.Standard2.8 | 8 | 120 |
| VM.Standard2.16 | 16 | 240 |
| VM.Standard2.24 | 24 | 320 |
NVIDIA GPU
If you have made a reservation for GPU with Compute, you can transfer the GPU to be used in the Data Science service. Create a support request ticket and specify the shape, count, region, tenancy, and reservation ID of the GPU reservation to be transferred to theData Science service. Note: After the reserved GPU is transferred to the Data Science service, Data Science begins metering for the GPU shape.
| Shape | GPU Type | GPU Count |
GPU Memory (GB) |
OCPU Count | CPU Memory (GB) |
|---|---|---|---|---|---|
|
VM.GPU2.1 (older generation) |
P100 | 1 | 16 | 12 | 72 |
| VM.GPU3.1 | V100 | 1 | 16 | 6 | 90 |
| VM.GPU3.2 | V100 | 2 | 32 | 12 | 180 |
| VM.GPU3.4 | V100 | 4 | 64 | 24 | 360 |
|
BM.GPU2.2 |
P100 | 2 | 32 | 28 | 192 |
| BM.GPU3.8 | V100 | 8 | 128 | 52 | 768 |
| BM.GPU4.8 | A100 | 8 | 320 | 64 | 2048 |
| BM.GPUA10.4 | A10 | 4 | 96 | 64 | 1024 |
| BM.GPU.H100.8 | H100 | 8 | 640 | 112 | 2048 |
| VM.GPU.A10.1 | A10 | 1 | 24 | 15 | 240 |
| VM.GPU.A10.2 | A10 | 2 | 48 | 30 | 480 |
| BM.GPU.L40S-NC.4 | L40S | 4 | 192 | 112 | 1024 |