Skip to content

docs(gpu): add mig profile for B300 instances #6174

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
bene2k1 merged 1 commit into from
Feb 10, 2026
Merged

docs(gpu): add mig profile for B300 instances #6174

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions pages/gpu/how-to/use-mig-with-kubernetes.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@ title: How to use NVIDIA MIG technology with Kubernetes
description: This section provides information about MIG with Kubernetes
tags: NVIDIA H100 MIG multi-instance gpu
dates:
validation: 2025-07-21
validation: 2026-02-10
posted: 2023-09-19
---
import Requirements from '@macros/iam/requirements.mdx'
Expand All @@ -27,7 +27,8 @@ In this guide, we will explore the capabilities of NVIDIA MIG within a Kubernete
- A [Kubernetes cluster](/kubernetes/quickstart/#how-to-create-a-kubernetes-cluster) with a [GPU Instance](/gpu/reference-content/choosing-gpu-instance-type/) as node

<Message type="tip">
MIG is fully supported on [Scaleway managed Kubernetes](/kubernetes/quickstart/) clusters (Kapsule and Kosmos).
- MIG is fully supported on [Scaleway managed Kubernetes](/kubernetes/quickstart/) clusters (Kapsule and Kosmos).
- MIG is available on Scaleway [H100, H100-SXM](/gpu/how-to/use-nvidia-mig-technology/#how-to-list-mig-profiles), and [B3OO](/gpu/how-to/use-nvidia-mig-technology/#mig-profiles-for-b300-gpu-instances) GPU Instances.
</Message>

## Configure MIG partitions inside a Kubernetes cluster
Expand Down
89 changes: 52 additions & 37 deletions pages/gpu/how-to/use-nvidia-mig-technology.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@ title: How to use the NVIDIA MIG technology on GPU Instances
description: This section provides information about NVIDIA's MIG technology
tags: NVIDIA H100 MIG multi-instance gpu
dates:
validation: 2025-07-21
validation: 2026-02-10
posted: 2023-08-31
---
import Requirements from '@macros/iam/requirements.mdx'
Expand Down Expand Up @@ -48,7 +48,7 @@ By default, the MIG feature of NVIDIA GPUs is disabled. To use it with your GPU

1. [Connect to your GPU Instance](/gpu/how-to/create-manage-gpu-instance/#how-to-connect-to-a-gpu-instance) as root using SSH.
2. Check the status of the MIG mode of your Instance running `nvidia-smi`. It shows that MIG mode is disabled:
```s
```bash
root@my-h100-instance:~# nvidia-smi -i 0
Tue Aug 22 11:58:39 2023
+-----------------------------------------------------------------------------+
Expand All @@ -64,13 +64,13 @@ By default, the MIG feature of NVIDIA GPUs is disabled. To use it with your GPU
+-------------------------------+----------------------+----------------------+
```
2. Run the following command to enable MIG mode on the GPU:
```s
```bash
root@my-h100-instance:~# sudo nvidia-smi -i 0 -mig 1
Enabled MIG Mode for GPU 00000000:01:00.0
All done.
```
3. Run the following command to verify that MIG mode is enabled on the GPU:
```s
```bash
root@my-h100-instance:~# nvidia-smi -i 0 --query-gpu=pci.bus_id,mig.mode.current --format=csv
pci.bus_id, mig.mode.current
00000000:01:00.0, Enabled
Expand All @@ -85,8 +85,8 @@ These profiles determine the sizes and functionalities available of the MIG part

Refer to the official documentation for more information about the supported [MIG profiles on H100 GPU Instances](https://docs.nvidia.com/datacenter/tesla/mig-user-guide/#h100-profiles).

1. Run the command `nvidia-smi mig -lgip` to retrieve a list of the available MIG profiles for the Instance. An output similar to the following displays:
```s
1. Run the command `nvidia-smi mig -lgip` to retrieve a list of the available MIG profiles for the Instance. An output similar to the following displays for H100 GPU Instances:
```bash
root@my-h100-instance:~# nvidia-smi mig -lgip
+-----------------------------------------------------------------------------+
| GPU instance profiles: |
Expand Down Expand Up @@ -115,8 +115,11 @@ Refer to the official documentation for more information about the supported [MI
| 8 7 1 |
+-----------------------------------------------------------------------------+
```



2. Run the following command to list the possible placements available. The syntax of the placement is `{<index>}:<GPU Slice Count>` and shows the placement of the instances on the GPU.
```s
```bash
root@my-h100-instance:~# nvidia-smi mig -lgipp
GPU 0 Profile ID 19 Placements: {0,1,2,3,4,5,6}:1
GPU 0 Profile ID 20 Placements: {0,1,2,3,4,5,6}:1
Expand All @@ -127,31 +130,43 @@ Refer to the official documentation for more information about the supported [MI
GPU 0 Profile ID 0 Placement : {0}:8
```

### MIG profiles for B300 GPU Instances

<Message type="important">
The MIG profiles for the NVIDIA B300 GPU are currently **not officially published** by NVIDIA. The information below is based on preliminary data confirmed via direct communication with NVIDIA and is **subject to change**. Be advised to proceed with caution in production environments.
</Message>

While NVIDIA has not yet released the official MIG profile specifications for the **B300**, NVIDIA indicates that the MIG profile configurations will [mirror those of the B200](https://docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-mig-profiles.html#b200-mig-profiles), but with increased memory allocation due to the B300's larger onboard memory.

The expected MIG profile memory capacities for the B300 are estimated as follows:

| B200 Profile (Memory) | Estimated B300 Equivalent (Memory) |
|-----------------------|------------------------------------|
| 1g.10gb (180 GB) | → 288 GB (estimated) |
| 1g.5gb (90 GB) | → 144 GB (estimated) |
| 1g.3gb (45 GB) | → 72 GB (estimated) |
| 1g.2gb (23 GB) | → 36 GB (estimated) |

These values are **proportional scaling** based on the B200-to-B300 memory ratio and internal confirmation, but are **not final**. Always verify available profiles using `nvidia-smi -L` or `nvidia-smi mig -ldev` on your Instance before deployment.

Once NVIDIA officially releases the B300 MIG specifications, we will update the documentation accordingly.
### MIG profiles for B300 GPU Instances

On NVIDIA B300 Instances, the following MIG profiles are available:
```bash

+-----------------------------------------------------------------------------+
| GPU instance profiles: |
| GPU Name ID Instances Memory P2P SM DEC ENC |
| Free/Total GiB CE JPEG OFA |
|=============================================================================|
| 0 MIG 1g.34gb 19 7/7 30.50 No 18 1 0 |
| 1 1 0 |
+-----------------------------------------------------------------------------+
| 0 MIG 1g.34gb+me 20 1/1 30.50 No 18 1 0 |
| 2 1 1 |
+-----------------------------------------------------------------------------+
| 0 MIG 1g.67gb 15 4/4 66.50 No 30 1 0 |
| 2 1 0 |
+-----------------------------------------------------------------------------+
| 0 MIG 2g.67gb 14 3/3 66.50 No 70 2 0 |
| 4 2 0 |
+-----------------------------------------------------------------------------+
| 0 MIG 3g.135gb 9 2/2 133.50 No 70 3 0 |
| 6 3 0 |
+-----------------------------------------------------------------------------+
| 0 MIG 4g.135gb 5 1/1 133.50 No 62 4 0 |
| 8 4 0 |
+-----------------------------------------------------------------------------+
| 0 MIG 7g.269gb 0 1/1 268.00 No 148 7 0 |
| 16 7 1 |
+-----------------------------------------------------------------------------+
```

## How to partition the GPU into several MIG partitions

1. Run the following command to divide the H100 GPU Instance into four slices (MIG partitions):
```s
```bash
root@my-h100-instance:~# sudo nvidia-smi mig -cgi 9,19,19,19 -C
Successfully created GPU instance ID 2 on GPU 0 using profile MIG 3g.40gb (ID 9)
Successfully created compute instance ID 0 on GPU 0 GPU instance ID 2 using profile MIG 3g.40gb (ID 2)
Expand Down Expand Up @@ -184,7 +199,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
</Message>

2. Run the following command to verify the MIG configuration of the GPU:
```s
```bash
root@my-h100-instance:~# sudo nvidia-smi mig -lgi
+-------------------------------------------------------+
| GPU instances: |
Expand All @@ -201,7 +216,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
+-------------------------------------------------------+
```
3. Display the UUID of each of your MIG partitions:
```s
```bash
root@my-h100-instance:~# nvidia-smi -L
GPU 0: NVIDIA H100 PCIe (UUID: GPU-7cd6d4d6-9fa8-13be-3c42-09a1b2280a02)
MIG 3g.40gb Device 0: (UUID: MIG-da06d78f-7534-56a0-a062-62fef012be91)
Expand All @@ -212,7 +227,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the

4. Run `nvidia-smi` on three 1g.10gb MIG partitions to display their characteristics:
* MIG 1g.10gb 1
```s
```bash
root@my-h100-instance:~# sudo docker run --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=MIG-8aa1fc52-9818-58ec-bc64-8f0cae121bb4 nvidia/cuda:11.0.3-base-ubuntu20.04 nvidia-smi
Tue Aug 22 12:53:40 2023
+-----------------------------------------------------------------------------+
Expand Down Expand Up @@ -247,7 +262,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
+-----------------------------------------------------------------------------+
```
* MIG 1g.10gb 2
```s
```bash
root@my-h100-instance:~# sudo docker run --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=MIG-42fa9c93-1430-5ccc-b623-c02fb93b7f5a nvidia/cuda:11.0.3-base-ubuntu20.04 nvidia-smi
Tue Aug 22 12:54:11 2023
+-----------------------------------------------------------------------------+
Expand Down Expand Up @@ -282,7 +297,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
+-----------------------------------------------------------------------------+
```
* MIG 1g.10gb 3
```s
```bash
root@my-h100-instance:~# sudo docker run --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=MIG-6d96b431-44ba-5360-80b0-9359561c927d nvidia/cuda:11.0.3-base-ubuntu20.04 nvidia-smi
Tue Aug 22 12:54:54 2023
+-----------------------------------------------------------------------------+
Expand Down Expand Up @@ -317,7 +332,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
+-----------------------------------------------------------------------------+
```
5. Launch a Jupyter Notebook within a Docker container on the MIG 3g.40gb MIG partition. Once initiated, you can reach it by connecting to port 8888 using the public IP of your GPU Instance.
```s
```bash
root@my-h100-instance:~# sudo docker run --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=MIG-da06d78f-7534-56a0-a062-62fef012be9 -p 8888:8888 jupyter/minimal-notebook
```

Expand All @@ -328,7 +343,7 @@ Once NVIDIA officially releases the B300 MIG specifications, we will update the
Once you have finished your jobs, you can delete the MIG partitions.

1. Run the following command to remove all MIG partitions and their corresponding compute instances:
```s
```bash
root@my-h100-instance:~# sudo nvidia-smi mig -dci && sudo nvidia-smi mig -dgi
Successfully destroyed compute instance ID 0 from GPU 0 GPU instance ID 7
Successfully destroyed compute instance ID 0 from GPU 0 GPU instance ID 8
Expand All @@ -340,7 +355,7 @@ Once you have finished your jobs, you can delete the MIG partitions.
Successfully destroyed GPU instance ID 2 from GPU 0
```
2. Verify that the MIG partitions have been removed by running the `nvidia-smi` command:
```s
```bash
+-----------------------------------------------------------------------------+
| MIG devices: |
+------------------+----------------------+-----------+-----------------------+
Expand All @@ -355,7 +370,7 @@ Once you have finished your jobs, you can delete the MIG partitions.
## How to disable MIG on a GPU Instance

To disable MIG on your H100 GPU Instance, run the following command:
```s
```bash
root@my-h100-instance:~# nvidia-smi -mig 0
Disabled MIG Mode for GPU 00000000:01:00.0
All done.
Expand Down
Loading