1 # Running HAT with Docker on NVIDIA GPUs
  2 
  3 ----
  4 * [Contents](hat-00.md)
  5 * Build Babylon and HAT
  6     * [Quick Install](hat-01-quick-install.md)
  7     * [Building Babylon with jtreg](hat-01-02-building-babylon.md)
  8     * [Building HAT with jtreg](hat-01-03-building-hat.md)
  9         * [Enabling the NVIDIA CUDA Backend](hat-01-05-building-hat-for-cuda.md)
 10 * [Testing Framework](hat-02-testing-framework.md)
 11 * [Running Examples](hat-03-examples.md)
 12 * [HAT Programming Model](hat-03-programming-model.md)
 13 * Interface Mapping
 14     * [Interface Mapping Overview](hat-04-01-interface-mapping.md)
 15     * [Cascade Interface Mapping](hat-04-02-cascade-interface-mapping.md)
 16 * Development
 17     * [Project Layout](hat-01-01-project-layout.md)
 18 * Implementation Details
 19     * [Walkthrough Of Accelerator.compute()](hat-accelerator-compute.md)
 20     * [How we minimize buffer transfers](hat-minimizing-buffer-transfers.md)
 21 * [Running HAT with Docker on NVIDIA GPUs](hat-07-docker-build-nvidia.md)
 22 ---
 23 
 24 ## Setting Up Docker Containers for NVIDIA GPUs
 25 
 26 To run Docker on NVIDIA GPUs, you need to install the NVIDIA Container Toolkit first.
 27 Follow the instructions [here](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html).
 28 
 29 Once the NVIDIA Container Toolkit has been installed, you can check access to the GPU by running the following image to run the `nvidia-smi` tool:
 30 
 31 ```bash
 32 sudo docker run --rm --runtime=nvidia --gpus all ubuntu nvidia-smi
 33 ```
 34 
 35 You will get an output similar to this:
 36 
 37 ```bash
 38 +-----------------------------------------------------------------------------------------+
 39 | NVIDIA-SMI 590.48.01              Driver Version: 590.48.01      CUDA Version: 13.1     |
 40 +-----------------------------------------+------------------------+----------------------+
 41 | GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
 42 | Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
 43 |                                         |                        |               MIG M. |
 44 |=========================================+========================+======================|
 45 |   0  NVIDIA GeForce RTX 5060        Off |   00000000:01:00.0  On |                  N/A |
 46 |  0%   46C    P8             12W /  145W |     578MiB /   8151MiB |      0%      Default |
 47 |                                         |                        |                  N/A |
 48 +-----------------------------------------+------------------------+----------------------+
 49 
 50 +-----------------------------------------------------------------------------------------+
 51 | Processes:                                                                              |
 52 |  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
 53 |        ID   ID                                                               Usage      |
 54 |=========================================================================================|
 55 |  No running processes found                                                             |
 56 +-----------------------------------------------------------------------------------------+
 57 ```
 58 
 59 ## DockerFile for HAT using NVIDIA SDK
 60 
 61 You can use the following `Dockerfile` for building a new docker container:
 62 
 63 ```dockerfile
 64 FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
 65 
 66 RUN apt-get update -q && apt install -qy \
 67         build-essential git cmake vim maven curl bash unzip zip wget
 68 
 69 WORKDIR /opt/babylon/
 70 RUN wget https://download.java.net/java/early_access/jdk26/22/GPL/openjdk-26-ea+22_linux-x64_bin.tar.gz
 71 RUN tar xvzf openjdk-26-ea+22_linux-x64_bin.tar.gz
 72 ENV JAVA_HOME=/opt/babylon/jdk-26/
 73 ENV PATH=$JAVA_HOME/bin:$PATH
 74 RUN java --version
 75 
 76 ## Configure Babylon/HAT from source
 77 RUN git clone https://github.com/openjdk/babylon.git
 78 WORKDIR /opt/babylon/babylon
 79 
 80 RUN apt-get update -y
 81 RUN apt-get install -y autoconf libfreetype6-dev
 82 RUN apt-get install -y file
 83 RUN apt-get install -y libasound2-dev
 84 RUN apt-get install -y libcups2-dev
 85 RUN apt-get install -y libfontconfig1-dev
 86 RUN apt-get install -y libx11-dev libxext-dev libxrender-dev libxrandr-dev libxtst-dev libxt-dev
 87 
 88 RUN bash configure --with-boot-jdk=${JAVA_HOME}
 89 RUN make clean
 90 RUN make images
 91 
 92 # Configure HAT
 93 WORKDIR /opt/babylon/babylon/hat
 94 RUN wget https://download.java.net/java/early_access/jextract/22/6/openjdk-22-jextract+6-47_linux-x64_bin.tar.gz
 95 RUN tar xvzf openjdk-22-jextract+6-47_linux-x64_bin.tar.gz > /dev/null
 96 ENV PATH=/opt/babylon/babylon/hat/jextract-22/bin:$PATH
 97 ENV PATH=/opt/babylon/babylon/build/linux-x86_64-server-release/jdk/bin/:$PATH
 98 ENV JAVA_HOME=/opt/babylon/babylon/build/linux-x86_64-server-release/jdk
 99 RUN /bin/bash -c "source env.bash"
100 
101 RUN java @hat/clean
102 RUN java @hat/bld
103 
104 ## Expose a volume to pass files in the local directory
105 WORKDIR /opt/babylon/babylon/hat/
106 VOLUME ["/data"]
107 ```
108 
109 ## Build Image
110 
111 Run the following command in the same directory of the `Dockerfile` with the previous configuration:
112 
113 ```bash
114 docker build . -t babylon
115 ```
116 
117 ## Running Examples on the NVIDIA GPU
118 
119 Check `nvidia-smi` tool from NVIDIA with the new image, so we have connection to the GPU:
120 
121 ```bash
122 docker run -it --rm --runtime=nvidia --gpus all babylon nvidia-smi
123 ```
124 
125 All setup! Now you can run HAT on NVIDIA GPUs.
126 
127 Run matrix-multiply example:
128 
129 ```bash
130 docker run -it --rm --runtime=nvidia --gpus all babylon java -cp hat/job.jar hat.java run ffi-cuda matmul --size=1024 --kernel=2DREGISTERTILING_FP16
131 ```
132 
133 ## Enable debug info
134 
135 ```bash
136 docker run -it --rm --runtime=nvidia --gpus all babylon java -cp hat/job.jar hat.java run ffi-cuda -DHAT=INFO matmul --size=1024 --kernel=2DREGISTERTILING_FP16
137 ```
138 
139 Expected output:
140 
141 ```bash
142 [INFO] Input Size     : 1024x1024
143 [INFO] Check Result:  : false
144 [INFO] Num Iterations : 100
145 [INFO] NDRangeConfiguration: 2DREGISTER_TILING_FP16
146 
147 [INFO] Using NVIDIA GPU: NVIDIA GeForce RTX 5060
148 [INFO] Dispatching the CUDA kernel
149         \_ BlocksPerGrid   = [16,16,1]
150         \_ ThreadsPerBlock = [16,16,1]
151 ```