dstack 0.19.8__tar.gz → 0.19.9__tar.gz

dstack-0.19.9/.github/workflows/docker-amd-smi.yml

@@ -0,0 +1,51 @@
+name: Build AMD SMI Docker image
+
+on:
+  workflow_dispatch:
+    inputs:
+      image_name:
+        description: "Docker image name"
+        required: true
+        default: "dstackai/amd-smi"
+      rocm_version:
+        description: "ROCm version"
+        required: true
+        default: "6.4"
+      dstack_revision:
+        description: "Docker image revision"
+        required: true
+        default: 0
+
+jobs:
+  build-amd-smi:
+    defaults:
+      run:
+        working-directory: docker/amd-smi
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Build and upload to DockerHub
+        run: |
+          IMAGE_NAME=${{ inputs.image_name }}
+          docker buildx build . \
+            --load \
+            --provenance=false \
+            --platform linux/amd64 \
+            --build-arg IMAGE_NAME=${IMAGE_NAME} \
+            --build-arg UBUNTU_VERSION=noble \
+            --build-arg ROCM_VERSION=${{ inputs.rocm_version }} \
+            --build-arg DSTACK_REVISION=${{ inputs.dstack_revision }} \
+            --build-arg BUILD_DATE=$(date --utc --iso-8601=seconds)Z \
+            --tag ${IMAGE_NAME}:latest
+          VERSION=$(docker inspect --format '{{ index .Config.Labels "org.opencontainers.image.version" }}' ${IMAGE_NAME})
+          docker tag ${IMAGE_NAME}:latest ${IMAGE_NAME}:${VERSION}
+          docker push ${IMAGE_NAME}:${VERSION}
+          docker push ${IMAGE_NAME}:latest

{dstack-0.19.8 → dstack-0.19.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dstack
-Version: 0.19.8
+Version: 0.19.9
 Summary: dstack is an open-source orchestration engine for running AI workloads on any cloud or on-premises.
 Project-URL: Homepage, https://dstack.ai
 Project-URL: Source, https://github.com/dstackai/dstack
@@ -54,7 +54,7 @@ Requires-Dist: azure-mgmt-network<28.0.0,>=23.0.0; extra == 'all'
 Requires-Dist: azure-mgmt-resource>=22.0.0; extra == 'all'
 Requires-Dist: azure-mgmt-subscription>=3.1.1; extra == 'all'
 Requires-Dist: backports-entry-points-selectable; extra == 'all'
-Requires-Dist: boto3; extra == 'all'
+Requires-Dist: boto3>=1.38.13; extra == 'all'
 Requires-Dist: botocore; extra == 'all'
 Requires-Dist: datacrunch; extra == 'all'
 Requires-Dist: docker>=6.0.0; extra == 'all'
@@ -90,7 +90,7 @@ Requires-Dist: alembic>=1.10.2; extra == 'aws'
 Requires-Dist: apscheduler<4; extra == 'aws'
 Requires-Dist: asyncpg; extra == 'aws'
 Requires-Dist: backports-entry-points-selectable; extra == 'aws'
-Requires-Dist: boto3; extra == 'aws'
+Requires-Dist: boto3>=1.38.13; extra == 'aws'
 Requires-Dist: botocore; extra == 'aws'
 Requires-Dist: docker>=6.0.0; extra == 'aws'
 Requires-Dist: fastapi; extra == 'aws'
@@ -231,7 +231,7 @@ Requires-Dist: alembic>=1.10.2; extra == 'lambda'
 Requires-Dist: apscheduler<4; extra == 'lambda'
 Requires-Dist: asyncpg; extra == 'lambda'
 Requires-Dist: backports-entry-points-selectable; extra == 'lambda'
-Requires-Dist: boto3; extra == 'lambda'
+Requires-Dist: boto3>=1.38.13; extra == 'lambda'
 Requires-Dist: botocore; extra == 'lambda'
 Requires-Dist: docker>=6.0.0; extra == 'lambda'
 Requires-Dist: fastapi; extra == 'lambda'
@@ -338,24 +338,27 @@ orchestration for ML teams across top clouds and on-prem clusters.
 
 #### Accelerators
 
-`dstack` supports `NVIDIA`, `AMD`, `Google TPU`, and `Intel Gaudi` accelerators out of the box.
+`dstack` supports `NVIDIA`, `AMD`, `Google TPU`, `Intel Gaudi`, and `Tenstorrent` accelerators out of the box.
 
-## Major news ✨
+## Latest news ✨
 
-- [2025/02] [dstack 0.18.41: GPU blocks, Proxy jump, inactivity duration, and more](https://github.com/dstackai/dstack/releases/tag/0.18.41)
-- [2025/01] [dstack 0.18.38: Intel Gaudi](https://github.com/dstackai/dstack/releases/tag/0.18.38)
-- [2025/01] [dstack 0.18.35: Vultr](https://github.com/dstackai/dstack/releases/tag/0.18.35)
-- [2024/12] [dstack 0.18.30: AWS Capacity Reservations and Capacity Blocks](https://github.com/dstackai/dstack/releases/tag/0.18.30)
-- [2024/10] [dstack 0.18.21: Instance volumes](https://github.com/dstackai/dstack/releases/tag/0.18.21)
-- [2024/10] [dstack 0.18.18: Hardware metrics monitoring](https://github.com/dstackai/dstack/releases/tag/0.18.18)
+- [2025/05] [dstack 0.19.8: Nebius clusters, GH200 on Lambda](https://github.com/dstackai/dstack/releases/tag/0.19.8)
+- [2025/04] [dstack 0.19.6: Tenstorrent, Plugins](https://github.com/dstackai/dstack/releases/tag/0.19.6)
+- [2025/04] [dstack 0.19.5: GCP A3 High clusters](https://github.com/dstackai/dstack/releases/tag/0.19.5)
+- [2025/04] [dstack 0.19.3: GCP A3 Mega clusters](https://github.com/dstackai/dstack/releases/tag/0.19.3)
+- [2025/03] [dstack 0.19.0: Prometheus](https://github.com/dstackai/dstack/releases/tag/0.19.0)
 
-## Installation
+## How does it work?
+
+<img src="https://dstack.ai/static-assets/static-assets/images/dstack-architecture-diagram-v8.svg" width="750" />
+
+### Installation
 
 > Before using `dstack` through CLI or API, set up a `dstack` server. If you already have a running `dstack` server, you only need to [set up the CLI](#set-up-the-cli).
 
-### Set up the server
+#### Set up the server
 
-#### (Optional) Configure backends
+##### (Optional) Configure backends
 
 To use `dstack` with cloud providers, configure backends
 via the `~/.dstack/server/config.yml` file.
@@ -365,21 +368,21 @@ For more details on how to configure backends, check [Backends](https://dstack.a
 > For using `dstack` with on-prem servers, create [SSH fleets](https://dstack.ai/docs/concepts/fleets#ssh) 
 > once the server is up.
 
-#### Start the server
+##### Start the server
 
 You can install the server on Linux, macOS, and Windows (via WSL 2). It requires Git and
 OpenSSH.
 
-##### pip
+##### uv
 
 ```shell
-$ pip install "dstack[all]" -U
+$ uv tool install "dstack[all]" -U
 ```
 
-##### uv
+##### pip
 
 ```shell
-$ uv tool install "dstack[all]" -U
+$ pip install "dstack[all]" -U
 ```
 
 Once it's installed, go ahead and start the server.
@@ -392,25 +395,28 @@ The admin token is "bbae0f28-d3dd-4820-bf61-8f4bb40815da"
 The server is running at http://127.0.0.1:3000/
 ```
 
-For more details on server configuration options, see the
+> For more details on server configuration options, see the
 [Server deployment](https://dstack.ai/docs/guides/server-deployment) guide.
 
-### Set up the CLI
+
+<details><summary>Set up the CLI</summary>
+
+#### Set up the CLI
 
 Once the server is up, you can access it via the `dstack` CLI. 
 
 The CLI can be installed on Linux, macOS, and Windows. It requires Git and OpenSSH.
 
-##### pip
+##### uv
 
 ```shell
-$ pip install dstack -U
+$ uv tool install dstack -U
 ```
 
-##### uv
+##### pip
 
 ```shell
-$ uv tool install dstack -U
+$ pip install dstack -U
 ```
 
 To point the CLI to the `dstack` server, configure it
@@ -425,9 +431,9 @@ $ dstack config \
 Configuration is updated at ~/.dstack/config.yml
 ```
 
-## How does it work?
+</details>
 
-### 1. Define configurations
+### Define configurations
 
 `dstack` supports the following configurations:
    
@@ -440,7 +446,7 @@ Configuration is updated at ~/.dstack/config.yml
 
 Configuration can be defined as YAML files within your repo.
 
-### 2. Apply configurations
+### Apply configurations
 
 Apply the configuration either via the `dstack apply` CLI command or through a programmatic API.
 
@@ -452,6 +458,7 @@ out-of-capacity errors, port-forwarding, and more &mdash; across clouds and on-p
 For additional information, see the following links:
 
 * [Docs](https://dstack.ai/docs)
+* [Examples](https://dstack.ai/examples)
 * [Discord](https://discord.gg/u8SmfwPpMd)
 
 ## Contributing

{dstack-0.19.8/src → dstack-0.19.9}/README.md

@@ -19,24 +19,30 @@ orchestration for ML teams across top clouds and on-prem clusters.
 
 #### Accelerators
 
-`dstack` supports `NVIDIA`, `AMD`, `Google TPU`, and `Intel Gaudi` accelerators out of the box.
+`dstack` supports `NVIDIA`, `AMD`, `Google TPU`, `Intel Gaudi`, and `Tenstorrent` accelerators out of the box.
 
-## Major news ✨
+## Latest news ✨
 
-- [2025/02] [dstack 0.18.41: GPU blocks, Proxy jump, inactivity duration, and more](https://github.com/dstackai/dstack/releases/tag/0.18.41)
-- [2025/01] [dstack 0.18.38: Intel Gaudi](https://github.com/dstackai/dstack/releases/tag/0.18.38)
-- [2025/01] [dstack 0.18.35: Vultr](https://github.com/dstackai/dstack/releases/tag/0.18.35)
-- [2024/12] [dstack 0.18.30: AWS Capacity Reservations and Capacity Blocks](https://github.com/dstackai/dstack/releases/tag/0.18.30)
-- [2024/10] [dstack 0.18.21: Instance volumes](https://github.com/dstackai/dstack/releases/tag/0.18.21)
-- [2024/10] [dstack 0.18.18: Hardware metrics monitoring](https://github.com/dstackai/dstack/releases/tag/0.18.18)
+- [2025/05] [dstack 0.19.8: Nebius clusters, GH200 on Lambda](https://github.com/dstackai/dstack/releases/tag/0.19.8)
+- [2025/04] [dstack 0.19.6: Tenstorrent, Plugins](https://github.com/dstackai/dstack/releases/tag/0.19.6)
+- [2025/04] [dstack 0.19.5: GCP A3 High clusters](https://github.com/dstackai/dstack/releases/tag/0.19.5)
+- [2025/04] [dstack 0.19.3: GCP A3 Mega clusters](https://github.com/dstackai/dstack/releases/tag/0.19.3)
+- [2025/03] [dstack 0.19.0: Prometheus](https://github.com/dstackai/dstack/releases/tag/0.19.0)
 
-## Installation
+## How does it work?
+
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="https://dstack.ai/static-assets/static-assets/images/dstack-architecture-diagram-v8-dark.svg"/>
+  <img src="https://dstack.ai/static-assets/static-assets/images/dstack-architecture-diagram-v8.svg" width="750" />
+</picture>
+
+### Installation
 
 > Before using `dstack` through CLI or API, set up a `dstack` server. If you already have a running `dstack` server, you only need to [set up the CLI](#set-up-the-cli).
 
-### Set up the server
+#### Set up the server
 
-#### (Optional) Configure backends
+##### (Optional) Configure backends
 
 To use `dstack` with cloud providers, configure backends
 via the `~/.dstack/server/config.yml` file.
@@ -46,21 +52,21 @@ For more details on how to configure backends, check [Backends](https://dstack.a
 > For using `dstack` with on-prem servers, create [SSH fleets](https://dstack.ai/docs/concepts/fleets#ssh) 
 > once the server is up.
 
-#### Start the server
+##### Start the server
 
 You can install the server on Linux, macOS, and Windows (via WSL 2). It requires Git and
 OpenSSH.
 
-##### pip
+##### uv
 
 ```shell
-$ pip install "dstack[all]" -U
+$ uv tool install "dstack[all]" -U
 ```
 
-##### uv
+##### pip
 
 ```shell
-$ uv tool install "dstack[all]" -U
+$ pip install "dstack[all]" -U
 ```
 
 Once it's installed, go ahead and start the server.
@@ -73,25 +79,28 @@ The admin token is "bbae0f28-d3dd-4820-bf61-8f4bb40815da"
 The server is running at http://127.0.0.1:3000/
 ```
 
-For more details on server configuration options, see the
+> For more details on server configuration options, see the
 [Server deployment](https://dstack.ai/docs/guides/server-deployment) guide.
 
-### Set up the CLI
+
+<details><summary>Set up the CLI</summary>
+
+#### Set up the CLI
 
 Once the server is up, you can access it via the `dstack` CLI. 
 
 The CLI can be installed on Linux, macOS, and Windows. It requires Git and OpenSSH.
 
-##### pip
+##### uv
 
 ```shell
-$ pip install dstack -U
+$ uv tool install dstack -U
 ```
 
-##### uv
+##### pip
 
 ```shell
-$ uv tool install dstack -U
+$ pip install dstack -U
 ```
 
 To point the CLI to the `dstack` server, configure it
@@ -106,9 +115,9 @@ $ dstack config \
 Configuration is updated at ~/.dstack/config.yml
 ```
 
-## How does it work?
+</details>
 
-### 1. Define configurations
+### Define configurations
 
 `dstack` supports the following configurations:
    
@@ -121,7 +130,7 @@ Configuration is updated at ~/.dstack/config.yml
 
 Configuration can be defined as YAML files within your repo.
 
-### 2. Apply configurations
+### Apply configurations
 
 Apply the configuration either via the `dstack apply` CLI command or through a programmatic API.
 
@@ -133,6 +142,7 @@ out-of-capacity errors, port-forwarding, and more &mdash; across clouds and on-p
 For additional information, see the following links:
 
 * [Docs](https://dstack.ai/docs)
+* [Examples](https://dstack.ai/examples)
 * [Discord](https://discord.gg/u8SmfwPpMd)
 
 ## Contributing

dstack-0.19.9/docker/amd-smi/Dockerfile

@@ -0,0 +1,30 @@
+ARG UBUNTU_VERSION
+
+FROM ubuntu:${UBUNTU_VERSION}
+
+ARG IMAGE_NAME
+ARG UBUNTU_VERSION
+ARG ROCM_VERSION
+ARG DSTACK_REVISION
+ARG BUILD_DATE
+
+ENV PATH="/opt/rocm/bin:${PATH}"
+
+RUN \
+    export DEBIAN_FRONTEND=noninteractive && \
+    apt-get update && \
+    apt-get install -y --no-install-recommends wget ca-certificates && \
+    base_url="https://repo.radeon.com/amdgpu-install/${ROCM_VERSION}/ubuntu/${UBUNTU_VERSION}/" && \
+    deb_name=$(wget -qO- "${base_url}" | grep -Po '(?<=href=")amdgpu-install_[^"]+') && \
+    wget -O amdgpu-install.deb "${base_url}${deb_name}" && \
+    apt-get install -y --no-install-recommends ./amdgpu-install.deb && \
+    rm ./amdgpu-install.deb && \
+    apt-get update && \
+    apt-get install -y amd-smi-lib
+
+ENTRYPOINT ["/opt/rocm/bin/amd-smi"]
+CMD ["--help"]
+
+LABEL org.opencontainers.image.title="${IMAGE_NAME}"
+LABEL org.opencontainers.image.version="${ROCM_VERSION}-${DSTACK_REVISION}"
+LABEL org.opencontainers.image.created="${BUILD_DATE}"

dstack-0.19.9/docker/amd-smi/README.md

@@ -0,0 +1,9 @@
+# dstack AMD SMI
+
+An Ubuntu-based image with [AMD SMI](https://rocm.docs.amd.com/projects/amdsmi/en/latest/) preinstalled. Suitable for AMD GPU detection.
+
+## Usage
+
+```shell
+docker run --rm --device /dev/kfd --device /dev/dri dstackai/amd-smi static
+```

{dstack-0.19.8 → dstack-0.19.9}/docs/assets/stylesheets/extra.css

@@ -546,7 +546,7 @@ h4.doc-heading {
         background-color: transparent;
         border-radius: 3px;
         border: 0.5px solid rgba(0, 0, 0, 0.87);
-        height: 1.95rem;
+        height: 1.84rem;
     }
 
     [data-md-color-primary=white] .md-search__form:hover {
@@ -813,13 +813,13 @@ body {
         z-index: 101;
     }
 
-    .md-search__form::before {
+    /* .md-search__form::before {
         content: "⌘";
         color: white;
         font-weight: 600;
         position: absolute;
         padding: 7px;
-        margin: 4.5px;
+        margin: 3.5px;
         font-size: .65rem;
         background-color: rgba(0,0,0,.87);
         border-radius: 5px;
@@ -835,16 +835,15 @@ body {
         font-weight: 600;
         position: absolute;
         padding: 7px;
-        margin: 4.5px;
+        margin: 3.5px;
         font-size: .65rem;
         background-color: rgba(0,0,0,.87);
         border-radius: 5px;
         right: 1px;
         top: 0;
-        /*z-index: 1;*/
         width: 30px;
         text-align: center;
-    }
+    } */
 
     .md-nav--lifted > .md-nav__list > .md-nav__item > [for] {
         display: none;
@@ -1440,6 +1439,10 @@ html .md-footer-meta.md-typeset a:is(:focus,:hover) {
 }
 
 @media screen and (min-width: 76.25em) {
+    .md-search {
+        margin-top: 2px;
+    }
+
     .md-search .md-search__inner {
         /*padding-top: 0.58rem;*/
         margin-right: 8px;

{dstack-0.19.8 → dstack-0.19.9}/docs/assets/stylesheets/landing.css

@@ -232,7 +232,7 @@
 }
 
 .md-header__buttons .md-button {
-    font-size: 18.7px;
+    font-size: 17.5px;
     /*letter-spacing: -0.5px;*/
 }
 
@@ -293,7 +293,7 @@
 .md-header__buttons .md-button-secondary {
     font-weight: 400 !important;
     white-space: nowrap;
-    padding: 0.4rem 0.8rem;
+    padding: 0.44rem 0.8rem;
 }
 
 .md-header__buttons .md-button-secondary:hover, .md-typeset .md-button-secondary:hover {
@@ -355,6 +355,7 @@
 
 .md-header__buttons {
     white-space: nowrap;
+    padding-top: 2px;
 }
 
 .tx-landing__highlights {
@@ -517,7 +518,7 @@
             border-left: 0.5px dotted rgba(0, 0, 0, 0.75);
         }
 
-        &:nth-child(n+8) {
+        &:nth-child(n+7) {
             border-bottom: 0.5px dotted rgba(0, 0, 0, 0.75);
         }
 
@@ -945,7 +946,7 @@
     border-bottom: none;
 }
 
-@media screen and (min-width: 76.1875em) {
+@media screen and (min-width: 44.984375em) {
     .supported_clouds_item {
         &:nth-child(1) {
             border-top-left-radius: 3px;
@@ -963,7 +964,19 @@
             border-bottom-right-radius: 3px;
         }
 
-        &:nth-child(4), &:nth-child(8), &:nth-child(12) {
+        &:nth-child(4), &:nth-child(8), &:nth-child(11) {
+            border-right: 0.5px solid black;
+        }
+
+        &:nth-child(n+8) {
+            border-bottom: 0.5px solid black;
+        }
+    }
+}
+
+@media screen and (max-width: 44.984375em) {
+    .supported_clouds_item {
+        &:nth-child(3), &:nth-child(6), &:nth-child(9), &:nth-child(11) {
             border-right: 0.5px solid black;
         }
 

dstack-0.19.9/docs/blog/posts/gh200-on-lambda.md

@@ -0,0 +1,89 @@
+---
+title: "Supporting ARM and NVIDIA GH200 on Lambda"
+date: 2025-05-12
+description: "TBA"
+slug: gh200-on-lambda
+image: https://dstack.ai/static-assets/static-assets/images/dstack-arm--gh200-lambda-min.png
+categories:
+  - ARM
+  - Cloud fleets
+  - SSH fleets
+---
+
+# Supporting ARM and NVIDIA GH200 on Lambda
+
+The latest update to `dstack` introduces support for NVIDIA GH200 instances on [Lambda](../../docs/concepts/backends.md#lambda)
+and enables ARM-powered hosts, including GH200 and GB200, with [SSH fleets](../../docs/concepts/fleets.md#ssh).
+
+<img src="https://dstack.ai/static-assets/static-assets/images/dstack-arm--gh200-lambda-min.png" width="630"/>
+
+<!-- more -->
+
+## ARM support
+
+Previously, `dstack` only supported x86 architecture with both cloud providers as well as on-prem clusters. With the latest update, it’s now possible to use both cloud and SSH fleets with ARM-based CPUs too. To request ARM CPUs in a run or fleet configuration, specify the arm architecture in the `resources`.`cpu` property:
+
+```yaml
+resources:
+  cpu: arm:4..  # 4 or more ARM cores
+```
+
+If the hosts in an SSH fleet have ARM CPUs, `dstack` will automatically detect both ARM-based CPUs as well as ARM-based GPU Superchips such as GH200 and enable their use.
+
+To see available offers with ARM CPUs, pass `--cpu arm` to the `dstack offer` command.
+
+## About GH200
+
+NVIDIA Grace is the first NVIDIA data center CPU, built on top of ARM specifically for AI workloads. The NVIDIA GH200 Superchip brings together a 72-core NVIDIA Grace CPU with an NVIDIA H100 GPU, connected with a high-bandwidth, memory-coherent NVIDIA NVLink-C2C interconnect.
+
+| CPU           | GPU  | CPU Memory               | GPU Memory         | NVLink-C2C |
+| ------------- | ---- | ------------------------ | ------------------ | ---------- |
+| Grace 72-core | H100 | 480GB LPDDR5X at 512GB/s | 96GB HBM3 at 4TB/s | 900GB/s    |
+
+The GH200 Superchip’s 450 GB/s bidirectional bandwidth enables KV cache offloading to CPU memory. While prefill can leverage CPU memory for optimizations like prefix caching, generation benefits from the GH200’s higher memory bandwidth.
+
+## GH200 on Lambda
+
+[Lambda :material-arrow-top-right-thin:{ .external }](https://cloud.lambda.ai/sign-up?_gl=1*1qovk06*_gcl_au*MTg2MDc3OTAyOS4xNzQyOTA3Nzc0LjE3NDkwNTYzNTYuMTc0NTQxOTE2MS4xNzQ1NDE5MTYw*_ga*MTE2NDM5MzI0My4xNzQyOTA3Nzc0*_ga_43EZT1FM6Q*czE3NDY3MTczOTYkbzM0JGcxJHQxNzQ2NzE4MDU2JGo1NyRsMCRoMTU0Mzg1NTU1OQ..){:target="_blank"} provides secure, user-friendly, reliable, and affordable cloud GPUs. Since end of last year, Lambda started to offer on-demand GH200 instances through their public cloud. Furthermore, they offer these instances at the promotional price of $1.49 per hour until June 30th 2025.
+
+With the latest `dstack` update, it’s now possible to use these instances with your Lambda account whether you’re running a dev environment, task, or service:
+
+<div editor-title=".dstack.yml"> 
+
+```yaml
+type: dev-environment
+name: my-env
+image: nvidia/cuda:12.8.1-base-ubuntu20.04
+ide: vscode
+
+resources:
+  gpu: GH200:1
+```
+
+</div>
+
+> Note, you have to use an ARM-based Docker image.
+
+To determine whether Lambda has GH200 on-demand instances available, run `dstack apply`:
+
+<div class="termy">
+
+```shell
+$ dstack apply -f .dstack.yml
+
+ #   BACKEND             RESOURCES                          INSTANCE TYPE  PRICE
+ 1   lambda (us-east-3)  cpu=arm:64 mem=464GB GH200:96GB:1  gpu_1x_gh200   $1.49
+```
+
+</div>
+
+!!! info "Retry policy"
+    Note, if GH200s are not  available at the moment, you can specify the [retry policy](../../docs/concepts/dev-environments.md#retry-policy) in your run configuration so that `dstack` can run the configuration once the GPU becomes available.
+
+> If you have GH200 or GB200-powered hosts already provisioned via Lambda, another cloud provider, or on-prem, you can now use them with [SSH fleets](../../docs/concepts/fleets.md#ssh).
+
+!!! info "What's next?"
+    1. Sign up with [Lambda :material-arrow-top-right-thin:{ .external }](https://cloud.lambda.ai/sign-up?_gl=1*1qovk06*_gcl_au*MTg2MDc3OTAyOS4xNzQyOTA3Nzc0LjE3NDkwNTYzNTYuMTc0NTQxOTE2MS4xNzQ1NDE5MTYw*_ga*MTE2NDM5MzI0My4xNzQyOTA3Nzc0*_ga_43EZT1FM6Q*czE3NDY3MTczOTYkbzM0JGcxJHQxNzQ2NzE4MDU2JGo1NyRsMCRoMTU0Mzg1NTU1OQ..){:target="_blank"}
+    2. Set up the [Lambda](../../docs/concepts/backends.md#lambda) backend
+    3. Follow [Quickstart](../../docs/quickstart.md)
+    4. Check [dev environments](../../docs/concepts/dev-environments.md), [tasks](../../docs/concepts/tasks.md), [services](../../docs/concepts/services.md), and [fleets](../../docs/concepts/fleets.md)

{dstack-0.19.8 → dstack-0.19.9}/docs/blog/posts/mpi.md

@@ -31,7 +31,7 @@ cluster usage—we’ve added support for it.
 
 Below is an example of a task that runs AllReduce test on 2 nodes, each with 4 GPUs (8 processes in total).
 
-<div editor-title="examples/misc/nccl-tests/.dstack.yml">
+<div editor-title="examples/distributed-training/nccl-tests/.dstack.yml">
 
 ```yaml
 type: task
@@ -101,5 +101,5 @@ as well as use MPI for other tasks.
 
 !!! info "What's next?"
     1. Learn more about [dev environments](../../docs/concepts/dev-environments.md), [tasks](../../docs/concepts/tasks.md), [services](../../docs/concepts/services.md), and [fleets](../../docs/concepts/fleets.md)
-    2. Check the [NCCL tests](../../examples/misc/nccl-tests/index.md) example
+    2. Check the [NCCL tests](../../examples/distributed-training/nccl-tests/index.md) example
     2. Join [Discord :material-arrow-top-right-thin:{ .external }](https://discord.gg/u8SmfwPpMd){:target="_blank"}

{dstack-0.19.8 → dstack-0.19.9}/docs/blog/posts/nvidia-and-amd-on-vultr.md

@@ -2,7 +2,7 @@
 title: Supporting NVIDIA and AMD accelerators on Vultr
 date: 2025-02-17
 description: "Introducing integration with Vultr: The new integration allows Vultr customers to train and deploy models on both AMD and NVIDIA GPUs."  
-slug: nvidia-and-amd-on-vultr.md
+slug: nvidia-and-amd-on-vultr
 image: https://dstack.ai/static-assets/static-assets/images/dstack-vultr.png
 categories:
   - Cloud fleets

{dstack-0.19.8 → dstack-0.19.9}/docs/docs/concepts/backends.md

@@ -774,7 +774,7 @@ projects:
 
 Also, the `vastai` backend supports on-demand instances only. Spot instance support coming soon.
 
-### TensorDock
+<!-- ### TensorDock
 
 Log into your [TensorDock :material-arrow-top-right-thin:{ .external }](https://dashboard.tensordock.com/) account, click Developers in the sidebar, and use the `Create an Authorization` section to create a new authorization key.
 
@@ -795,7 +795,7 @@ projects:
 
 </div>
 
-The `tensordock` backend supports on-demand instances only. Spot instance support coming soon.
+The `tensordock` backend supports on-demand instances only. Spot instance support coming soon. -->
 
 ### CUDO
 

{dstack-0.19.8 → dstack-0.19.9}/docs/docs/concepts/fleets.md

@@ -71,7 +71,16 @@ This ensures all instances are provisioned in the same backend and region with o
     If the `aws` backend config has `public_ips: false` set, `dstack` enables the maximum number of interfaces supported by the instance.
     Otherwise, if instances have public IPs, only one EFA interface is enabled per instance due to AWS limitations.
 
-> The `cluster` placement is supported only for `aws`, `azure`, `gcp`, `oci`, and `vultr`
+??? info "Nebius"
+    `dstack` automatically creates an [InfiniBand cluster](https://docs.nebius.com/compute/clusters/gpu)
+    if all instances in the fleet support it.
+    Otherwise, instances are only connected by the default VPC subnet.
+
+    An InfiniBand fabric for the cluster is selected automatically.
+    If you prefer to use some specific fabrics, configure them in the
+    [backend settings](../reference/server/config.yml.md#nebius).
+
+> The `cluster` placement is supported only for `aws`, `azure`, `gcp`, `nebius`, `oci`, and `vultr`
 > backends.
 
 #### Resources