Hardware Unit Shapes by Region
This page lists the hardware shapes available for dedicated AI clusters in OCI Generative AI by region.
Use hardware shapes when you create a dedicated AI cluster to host a model. The hardware
shape identifies the underlying hardware for the dedicated AI cluster, such as
A100_80G or H100.
The region summary tables list only the hardware shape. Provider prefixes and unit counts aren't shown in the region summary tables.
The full unit shape name in the Console or API depends on how the model is made available:
- Models offered by OCI Generative AI for dedicated hosting: The unit shape
name begins with a provider or model-family prefix. Examples include
Cohere_A100_80G_X4,Meta_H100_X8, andOAI_A10_X2. - Imported models: The unit shape name doesn't include a provider prefix. For
example, use
A100_80G_X4orH100_X8. When you create the hosting dedicated AI cluster, you select the imported model separately from its hardware unit shape.
In both naming patterns, values such as A100_80G and
H100 identify the hardware shape, and the value after
X identifies the number of hardware units.
The AI unit count is based on the hardware shape and the number of hardware units, not
the provider prefix. For example, H100_X2 and
Cohere_H100_X2 both use two H100 hardware units and have the same AI unit
count. See Hardware Unit Shapes, Service
Limits, and AI Unit Counts for the AI unit count of each hardware unit shape.
North America
The following table lists hardware shapes for the North America regions.
| US East (Ashburn) (OC1) |
US Midwest (Chicago) (OC1) |
US West (Phoenix) (OC1) |
|---|---|---|
|
|
|
|
South America
The following table lists hardware shapes for the South America regions.
| Brazil East (Sao Paulo) (OC1) |
|---|
|
|
Europe
The following table lists hardware shapes for the Europe regions.
| Germany Central (Frankfurt) (OC1) |
EU Sovereign Central (Frankfurt) (OC19) |
UK South (London) (OC1) |
UK Gov South (London) (OC4) |
|---|---|---|---|
|
|
|
|
|
Middle East
The following table lists hardware shapes for the Middle East regions.
| Saudi Arabia Central (Riyadh) (OC1) |
UAE Central (Abu Dhabi) (OC1) |
UAE East (Dubai) (OC1) |
|---|---|---|
|
|
|
|
Asia Pacific
The following table lists hardware shapes for the Asia Pacific regions.
| India South (Hyderabad) (OC1) |
Japan Central (Osaka) (OC1) |
|---|---|
|
|
|
Hardware Unit Shapes, Service Limits, and AI Unit Counts
The following table lists the service limit, required hardware units, and AI unit
count for each hardware unit shape. The AI unit count is determined by the hardware
shape and the number after X. A provider or model-family prefix
doesn't change the AI unit count. For example, look up H100_X2 in
this table for either H100_X2 or
Cohere_H100_X2.
The AI Unit Count is the pricing multiplier for one replica of a hardware unit shape. To estimate the hourly price for one replica, multiply the applicable AI Unit Per Hour price on the Pricing Page by the AI unit count in the table. If the cluster has more than one replica, also multiply the result by the number of replicas.
| Hardware Unit Shape | Service Limit Name | Required Hardware Units | AI Unit Count |
|---|---|---|---|
A10_X1 |
dedicated-unit-a10-count |
1 | 1.77 |
A10_X2 |
dedicated-unit-a10-count |
2 | 3.54 |
A10_X4 |
dedicated-unit-a10-count |
4 | 7.08 |
A100_40G_X1 |
dedicated-unit-a100-40g-count |
1 | 2.70 |
A100_40G_X2 |
dedicated-unit-a100-40g-count |
2 | 5.40 |
A100_40G_X4 |
dedicated-unit-a100-40g-count |
4 | 10.80 |
A100_40G_X8 |
dedicated-unit-a100-40g-count |
8 | 21.60 |
A100_80G_X1 |
dedicated-unit-a100-80g-count |
1 | 3.24 |
A100_80G_X2 |
dedicated-unit-a100-80g-count |
2 | 6.48 |
A100_80G_X4 |
dedicated-unit-a100-80g-count |
4 | 12.96 |
A100_80G_X8 |
dedicated-unit-a100-80g-count |
8 | 25.92 |
B200_X1 |
dedicated-unit-b200-count |
1 | 8.02 |
B200_X2 |
dedicated-unit-b200-count |
2 | 16.04 |
B200_X4 |
dedicated-unit-b200-count |
4 | 32.08 |
B200_X8 |
dedicated-unit-b200-count |
8 | 64.16 |
H100_X1 |
dedicated-unit-h100-count |
1 | 6.01 |
H100_X2 |
dedicated-unit-h100-count |
2 | 12.02 |
H100_X4 |
dedicated-unit-h100-count |
4 | 24.04 |
H100_X8 |
dedicated-unit-h100-count |
8 | 48.08 |
H100_X16 |
dedicated-unit-h100-count |
16 | 96.16 |
H200_X1 |
dedicated-unit-h200-count |
1 | 6.22 |
H200_X2 |
dedicated-unit-h200-count |
2 | 12.44 |
H200_X4 |
dedicated-unit-h200-count |
4 | 24.88 |
H200_X8 |
dedicated-unit-h200-count |
8 | 49.76 |
Use the Oracle Cloud Cost Estimator to estimate the cost of your dedicated AI cluster configuration. If the tenancy doesn't have enough limits for the selected hardware shape and unit count, Creating a Limit Increase Request.
Hardware Unit Shapes for Imported Models
For an imported model, select a compatible hardware unit shape that's available in
the target region. Imported-model unit shape names don't include a provider prefix
because you select the imported model separately when you create the hosting dedicated
AI cluster. For example, H100_X2 requires two H100 hardware units.
See Hardware Unit Shapes, Service Limits,
and AI Unit Counts for its service limit and AI unit count.
Use the region summary tables on this page to identify the hardware shapes available in each region. For the recommended dedicated AI cluster unit shapes for each imported model, see Compatible Models. The recommended shape can be used as the baseline, but you can use more hardware resources when the model supports them and the shape is available in the region.
To calculate the hourly price for an imported-model cluster, multiply the AI Unit Per Hour price for Oracle Cloud Infrastructure Generative AI - Model Import on the Pricing Page by the AI unit count for the selected shape. If the cluster has more than one replica, also multiply the result by the number of replicas.
Imported models don't require the minimum hosting commitment of 744 unit-hours that applies when you host pretrained models available in OCI Generative AI on dedicated AI clusters.
For the imported-model workflow, see Managing Imported Models.
Baseline Hardware Unit Shapes for OCI Models
The following regional tables show the complete provider-specific unit shape names for models offered by OCI Generative AI for dedicated hosting. To find the AI unit count, remove the provider or model-family prefix and find the remaining hardware unit shape in Hardware Unit Shapes, Service Limits, and AI Unit Counts. For example, use the H100_X2 row to find the AI unit count for Cohere_H100_X2.
North America Baseline Shapes
The following table lists baseline hardware unit shape names for models in North America. You can specify more replicas of a baseline shape when you create a dedicated AI cluster, or later when you update the cluster.
| Model Name | US East (Ashburn) (OC1) |
US Midwest (Chicago) (OC1) |
US West (Phoenix) (OC1) |
|---|---|---|---|
| Cohere Command A |
|
|
|
| Cohere Command A Reasoning |
|
|
|
| Cohere Command A Vision |
|
|
|
| Cohere Embed 4 |
|
|
- |
| Cohere Rerank 4 Pro |
|
|
|
| Cohere Rerank 4 Fast |
|
|
|
| Meta Llama 4 Maverick |
|
|
- |
| Meta Llama 4 Scout |
|
|
- |
| Meta Llama 3.3 70B (Standard) | - |
|
|
| Meta Llama 3.3 70B (Dynamic FP8) | - |
|
|
| OpenAI gpt-oss-20b |
|
|
|
| OpenAI gpt-oss-120b |
|
|
|
South America Baseline Shapes
The following table lists baseline hardware unit shape names for models in South America. You can specify more replicas of a baseline shape when you create a dedicated AI cluster, or later when you update the cluster.
| Model Name | Brazil East (Sao Paulo) (OC1) |
|---|---|
| Cohere Command A |
|
| Cohere Command A Reasoning |
|
| Cohere Command A Vision |
|
| Cohere Embed 4 |
|
| Cohere Rerank 4 Pro |
|
| Cohere Rerank 4 Fast |
|
| Meta Llama 4 Maverick |
|
| Meta Llama 4 Scout |
|
| Meta Llama 3.3 70B (Standard) |
|
| Meta Llama 3.3 70B (Dynamic FP8) |
|
| OpenAI gpt-oss-20b |
|
| OpenAI gpt-oss-120b |
|
Europe Baseline Shapes
The following table lists baseline hardware unit shape names for models in Europe. You can specify more replicas of a baseline shape when you create a dedicated AI cluster, or later when you update the cluster.
| Model Name | Germany Central (Frankfurt) (OC1) |
EU Sovereign Central (Frankfurt) (OC19) |
UK South (London) (OC1) |
UK Gov South (London) (OC4) |
|---|---|---|---|---|
| Cohere Command A |
|
|
| - |
| Cohere Command A Reasoning |
| - |
| - |
| Cohere Command A Vision |
| - |
| - |
| Cohere Embed 4 |
| - |
| - |
| Cohere Rerank 4 Pro |
| - |
| - |
| Cohere Rerank 4 Fast |
| - |
| - |
| Meta Llama 4 Maverick | - | - |
| - |
| Meta Llama 4 Scout | - | - |
| - |
| Meta Llama 3.3 70B (Standard) |
|
|
|
|
| Meta Llama 3.3 70B (Dynamic FP8) |
|
|
|
|
| OpenAI gpt-oss-20b |
| - |
| - |
| OpenAI gpt-oss-120b |
| - |
| - |
Middle East Baseline Shapes
The following table lists baseline hardware unit shape names for models in the Middle East. You can specify more replicas of a baseline shape when you create a dedicated AI cluster, or later when you update the cluster.
| Model Name | Saudi Arabia Central (Riyadh) (OC1) |
UAE Central (Abu Dhabi) (OC1) |
UAE East (Dubai) (OC1) |
|---|---|---|---|
| Cohere Command A |
| - |
|
| Cohere Command A Reasoning |
| - |
|
| Cohere Command A Vision |
|
|
|
| Cohere Embed 4 |
|
|
|
| Cohere Rerank 4 Pro |
| - | - |
| Cohere Rerank 4 Fast |
| - | - |
| Meta Llama 4 Maverick |
|
| - |
| Meta Llama 4 Scout |
|
| - |
| Meta Llama 3.3 70B (Standard) |
| - | - |
| Meta Llama 3.3 70B (Dynamic FP8) | - |
|
|
| OpenAI gpt-oss-20b |
| - |
|
| OpenAI gpt-oss-120b |
| - |
|
Asia Pacific Baseline Shapes
The following table lists baseline hardware unit shape names for models in Asia Pacific. You can specify more replicas of a baseline shape when you create a dedicated AI cluster, or later when you update the cluster.
| Model Name | India South (Hyderabad) (OC1) |
Japan Central (Osaka) (OC1) |
|---|---|---|
| Cohere Command A |
|
|
| Cohere Command A Reasoning |
|
|
| Cohere Command A Vision |
|
|
| Cohere Embed 4 |
|
|
| Cohere Rerank 4 Pro |
|
|
| Cohere Rerank 4 Fast |
|
|
| Meta Llama 4 Maverick |
|
|
| Meta Llama 4 Scout |
|
|
| Meta Llama 3.3 70B (Standard) |
|
|
| Meta Llama 3.3 70B (Dynamic FP8) |
|
|
| OpenAI gpt-oss-20b |
|
|
| OpenAI gpt-oss-120b |
|
|