Search
Contact salesLog in to AI StudioLog in to AI Cloud

Nebius AI Cloud is now integrated with SkyPilot

April 2, 2025
6 mins to read

We’re excited to announce our integration with SkyPilot, an open-source framework that simplifies running AI and batch workloads across cloud platforms. This collaboration enables direct access to Nebius AI Cloud resources via SkyPilot. If you’re already using SkyPilot or considering it as a solution for your workloads, you can now conveniently use our AI Cloud through the same interface. It’s yet another way to access Nebius — alongside our API, CLI, Terraform recipes and others.

What is SkyPilot?

SkyPilot is a framework designed to optimize GPU availability and reduce costs when running AI workloads. It provides unified access to cloud resources through simple configuration files.

How to use SkyPilot with Nebius AI Cloud

Setting up SkyPilot with Nebius AI Cloud is straightforward:

Configure Nebius access

wget https://raw.githubusercontent.com/nebius/nebius-solution-library/refs/heads/main/skypilot/nebius-setup.sh
  • Run the following commands:
chmod +x nebius-setup.sh 
./nebius-setup.sh
  • You’ll be prompted to choose a Nebius tenant and project id from a list.

Install SkyPilot

Install SkyPilot with Nebius support:

pip install "skypilot-nightly[nebius]"

Run your first job

Create a simple YAML configuration file named sky.yaml:

resources:
  cloud: nebius
  accelerators: H100:1
  region: eu-north1

file_mounts:
  /my_data:
  source: nebius://my-nebius-bucket # must be unique; replace with your own bucket name
  
setup: |
  echo "Setup will be executed on every `sky launch` command on all nodes"
  
run: |
  echo "Run will be executed on every `sky exec` command on all nodes"
  echo "Do we have GPUs?"
  nvidia-smi 
  echo "Do we have data?"
  ls -l /my_data

Launch your job:

$ sky launch -c sky-test sky.yaml
YAML to run: sky.yaml
Considered resources (1 node):
--------------------------------------------------------------------------------------------------------------
CLOUD       INSTANCE                           vCPUs  Mem(GB)  ACCELERATORS   REGION/ZONE   COST ($)   CHOSEN
--------------------------------------------------------------------------------------------------------------
Nebius      gpu-h100-sxm_8gpu-128vcpu-1600gb   128    1600     H100:8         eu-north1     23.60      ✔
--------------------------------------------------------------------------------------------------------------

Launching a new cluster 'sky-test'. Proceed? [Y/n]: Y
⚙︎ Launching on Nebius eu-north1.
└── Instance is up.
✓ Cluster launched: sky-test. View logs: sky api logs -l sky-2025-02-27-11-36-07-257438/provision.log

⚙︎ Syncing files.
✓ Setup detached.

⚙︎ Job submitted, ID: 1
├── Waiting for task resources on 1 node.
└── Job started. Streaming logs... (Ctrl-C to exit log streaming; job will not be killed)
...
(setup pid=3800) Setup will be executed on every command on all nodes
(task, pid=3800) Command 'sky' not found, but can be installed with:
(task, pid=3800) sudo apt install beneath-a-steel-sky
(task, pid=3800) Run will be executed on every command on all nodes
(task, pid=3800) Do we have GPUs?
(task, pid=3800) Mon Mar 24 12:35:14 2025
(task, pid=3800) +-----------------------------------------------------------------------------------------+
(task, pid=3800) | NVIDIA-SMI 550.127.08 Driver Version: 550.127.08 CUDA Version: 12.4                    |
(task, pid=3800) |-----------------------------------------+------------------------+----------------------|
(task, pid=3800) | GPU Name        Persistence-M           | Bus-Id     Disp.A      | Volatile Uncorr. ECC |
(task, pid=3800) | Fan  Temp  Perf  Pwr:Usage/Cap          | Memory-Usage          | GPU-Util  Compute M. |
(task, pid=3800) |                                         |                      | MIG M.               |
(task, pid=3800) |=========================================+========================+======================|
(task, pid=3800) | 0  NVIDIA H100 80GB HBM3 On             | 00000000:8A:00.0  Off | 0                    |
(task, pid=3800) | N/A 27C   P0    67W / 700W              | 1MiB / 81559MiB       | 0%       Default     |
(task, pid=3800) |                                         |                      | Disabled             |
(task, pid=3800) +-----------------------------------------+------------------------+----------------------+

(task, pid=3800) +-----------------------------------------------------------------------------------------+
(task, pid=3800) | Processes:                                                                             |
(task, pid=3800) | GPU   GI   CI   PID   Type   Process name                          GPU Memory Usage    |
(task, pid=3800) |=========================================================================================|
(task, pid=3800) | No running processes found                                                             |
(task, pid=3800) +-----------------------------------------------------------------------------------------+

(task, pid=3800) Do we have data?
(task, pid=3800) total 377487364
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:21 file_1
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:25 file_10
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:26 file_11
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:26 file_12
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:21 file_2
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:22 file_3
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:22 file_4
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:23 file_5
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:23 file_6
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:24 file_7
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:24 file_8
(task, pid=3800) -rw-r--r-- 1 ubuntu ubuntu 32212254720 Mar 10 14:25 file_9
(task, pid=3800) drwxr-xr-x 2 ubuntu ubuntu 4096 Mar 24 12:35 joblogs

✓ Job finished (status: SUCCEEDED).

Here’s a brief overview of what SkyPilot just did:

  1. Provisioned a Nebius AI Cloud instance.

  2. Mounted the my-nebius-bucket cloud bucket to /my_data on the VM.

  3. Executed the setup and run commands to verify GPU availability and access to the mounted data.

For more details, check out our docs.

Multi-node example

High-performance NVIDIA Quantum InfiniBand connections make Nebius AI Cloud ideal for distributed training. Here’s a sample configuration for testing network connectivity:

resources:
  cloud: nebius
  accelerators: H100:8
  region: eu-north1
  
num_nodes: 2
  
setup: |
  sudo apt install perftest -y
  
run: |
  MASTER_ADDR=(echo"(echo "SKYPILOT_NODE_IPS" | head -n1)
  if [ "${SKYPILOT_NODE_RANK}" == "0" ]; then
    ib_send_bw --report_gbits -n 1000 -F > /dev/null
  elif [ "${SKYPILOT_NODE_RANK}" == "1" ]; then
    echo "Testing connection to: $MASTER_ADDR"
    sleep 2 # Wait for master to start
    ib_send_bw $MASTER_ADDR --report_gbits -n 1000 -F
  fi

$ sky launch -c test-sky sky.yaml
YAML to run: sky.yaml
Running on cluster: test-sky
⚙︎ Launching on Nebius eu-north1.
...

(worker1, rank=1, pid=33870, ip=192.168.0.15) ----------------------------------------------------------------
(worker1, rank=1, pid=33870, ip=192.168.0.15) Send BW Test
(worker1, rank=1, pid=33870, ip=192.168.0.15) Dual-port : OFF Device : mlx5_0
(worker1, rank=1, pid=33870, ip=192.168.0.15) Number of qps : 1 Transport type : IB
(worker1, rank=1, pid=33870, ip=192.168.0.15) Connection type : RC Using SRQ : OFF
(worker1, rank=1, pid=33870, ip=192.168.0.15) PCIe relax order: ON
(worker1, rank=1, pid=33870, ip=192.168.0.15) ibv_wr* API : ON
(worker1, rank=1, pid=33870, ip=192.168.0.15) TX depth : 128
(worker1, rank=1, pid=33870, ip=192.168.0.15) CQ Moderation : 1
(worker1, rank=1, pid=33870, ip=192.168.0.15) Mtu : 4096[B]
(worker1, rank=1, pid=33870, ip=192.168.0.15) Link type : IB
(worker1, rank=1, pid=33870, ip=192.168.0.15) Max inline data : 0[B]
(worker1, rank=1, pid=33870, ip=192.168.0.15) rdma_cm QPs : OFF
(worker1, rank=1, pid=33870, ip=192.168.0.15) Data ex. method : Ethernet
(worker1, rank=1, pid=33870, ip=192.168.0.15) ----------------------------------------------------------------
(worker1, rank=1, pid=33870, ip=192.168.0.15) local address: LID 0x1334 QPN 0x0131 PSN 0xcdddde
(worker1, rank=1, pid=33870, ip=192.168.0.15) remote address: LID 0x132f QPN 0x0131 PSN 0x90f79b
(worker1, rank=1, pid=33870, ip=192.168.0.15) ----------------------------------------------------------------
(worker1, rank=1, pid=33870, ip=192.168.0.15) #bytes #iterations BW peak[Gb/sec] BW average[Gb/sec] MsgRate[Mpps]
(worker1, rank=1, pid=33870, ip=192.168.0.15) 65536 1000 361.82 361.67 0.689839
(worker1, rank=1, pid=33870, ip=192.168.0.15) ----------------------------------------------------------------

✓ Job finished (status: SUCCEEDED).

Next steps

We’re committed to further enhancing this integration and welcome your feedback. To learn more about SkyPilot’s features and capabilities, visit the SkyPilot documentation and a dedicated section on our docs.

Explore Nebius AI Cloud

Compute Cloud​
Managed Kubernetes​
Object Storage​
Managed PostgreSQL​

Explore Nebius AI Studio

Models​
Playground​
Fine-tuning​
Rate limits​
Docs​
author
Alexander Kim
Cloud Solutions Architect at Nebius
author
Alex Salikov
Head of External Solutions Department at Nebius

See also

Nebius opens pre-orders for NVIDIA Blackwell GPU-powered clusters

We are now accepting pre-orders for NVIDIA GB200 NVL72 and NVIDIA HGX B200 clusters to be deployed in our data centers in the United States and Finland from early 2025. Based on NVIDIA Blackwell, the architecture to power a new industrial revolution of generative AI, these new clusters deliver a massive leap forward over existing solutions.

December 4, 2024
3 mins to read
Sign in to save this post