Update Jupyter Notebooks

Author: AndiH

10-H_CUDA_Graphs_and_Device-initiated_Communication_with_NVSHMEM/.master/Device-initiated_Communication_with_NVSHMEM/Instructions.ipynb

@@ -52,15 +52,15 @@
  "\n",
  "#### Note\n",
  "\n",
- "The Slurm installation on JUWELS-Booster sets `CUDA_VISIBLE_DEVICES`\n",
- "automatically so that each spawned process only sees the GPU it should\n",
- "use (see [GPU\n",
- "Devices](https://apps.fz-juelich.de/jsc/hps/juwels/booster-overview.html#gpu-devices)\n",
- "in the JUWELS Booster Overview documentation). This is not supported for\n",
- "NVSHMEM. The automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled\n",
- "by setting `CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes\n",
- "srun. With `CUDA_VISIBLE_DEVICES` set all spawned processes can see all\n",
- "GPUs listed. This is automatically done for the `sanitize`, `run` and\n",
+ "The Slurm installation on JEDI sets `CUDA_VISIBLE_DEVICES` automatically\n",
+ "so that each spawned process only sees the GPU it should use (see\n",
+ "[Hardware\n",
+ "Configuration](https://apps.fz-juelich.de/jsc/hps/jedi/configuration.html#hardware-configuration)\n",
+ "in the JEDI user documentation). This is not supported for NVSHMEM. The\n",
+ "automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled by setting\n",
+ "`CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes srun. With\n",
+ "`CUDA_VISIBLE_DEVICES` set all spawned processes can see all GPUs\n",
+ "listed. This is automatically done for the `sanitize`, `run` and\n",
  "`profile` make targets.\n",
  "\n",
  "### Advanced Task: Use `nvshmemx_float_put_nbi_block`\n",
@@ -85,7 +85,7 @@
  " repository](https://github.com/NVIDIA/multi-gpu-programming-models)\n",
  " implements the same strategy."
  ],
- "id": "ff43766e-b9d0-47c1-bf8c-242e07bb5239"
+ "id": "17eb86cd-3a72-4214-82c1-84818a0d1eca"
  }
  ],
  "nbformat": 4,

10-H_CUDA_Graphs_and_Device-initiated_Communication_with_NVSHMEM/solutions/Device-initiated_Communication_with_NVSHMEM/Instructions.ipynb

@@ -52,15 +52,15 @@
  "\n",
  "#### Note\n",
  "\n",
- "The Slurm installation on JUWELS-Booster sets `CUDA_VISIBLE_DEVICES`\n",
- "automatically so that each spawned process only sees the GPU it should\n",
- "use (see [GPU\n",
- "Devices](https://apps.fz-juelich.de/jsc/hps/juwels/booster-overview.html#gpu-devices)\n",
- "in the JUWELS Booster Overview documentation). This is not supported for\n",
- "NVSHMEM. The automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled\n",
- "by setting `CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes\n",
- "srun. With `CUDA_VISIBLE_DEVICES` set all spawned processes can see all\n",
- "GPUs listed. This is automatically done for the `sanitize`, `run` and\n",
+ "The Slurm installation on JEDI sets `CUDA_VISIBLE_DEVICES` automatically\n",
+ "so that each spawned process only sees the GPU it should use (see\n",
+ "[Hardware\n",
+ "Configuration](https://apps.fz-juelich.de/jsc/hps/jedi/configuration.html#hardware-configuration)\n",
+ "in the JEDI user documentation). This is not supported for NVSHMEM. The\n",
+ "automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled by setting\n",
+ "`CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes srun. With\n",
+ "`CUDA_VISIBLE_DEVICES` set all spawned processes can see all GPUs\n",
+ "listed. This is automatically done for the `sanitize`, `run` and\n",
  "`profile` make targets.\n",
  "\n",
  "### Advanced Task: Use `nvshmemx_float_put_nbi_block`\n",
@@ -85,7 +85,7 @@
  " repository](https://github.com/NVIDIA/multi-gpu-programming-models)\n",
  " implements the same strategy."
  ],
- "id": "ff43766e-b9d0-47c1-bf8c-242e07bb5239"
+ "id": "17eb86cd-3a72-4214-82c1-84818a0d1eca"
  }
  ],
  "nbformat": 4,

10-H_CUDA_Graphs_and_Device-initiated_Communication_with_NVSHMEM/tasks/Device-initiated_Communication_with_NVSHMEM/Instructions.ipynb

@@ -52,15 +52,15 @@
  "\n",
  "#### Note\n",
  "\n",
- "The Slurm installation on JUWELS-Booster sets `CUDA_VISIBLE_DEVICES`\n",
- "automatically so that each spawned process only sees the GPU it should\n",
- "use (see [GPU\n",
- "Devices](https://apps.fz-juelich.de/jsc/hps/juwels/booster-overview.html#gpu-devices)\n",
- "in the JUWELS Booster Overview documentation). This is not supported for\n",
- "NVSHMEM. The automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled\n",
- "by setting `CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes\n",
- "srun. With `CUDA_VISIBLE_DEVICES` set all spawned processes can see all\n",
- "GPUs listed. This is automatically done for the `sanitize`, `run` and\n",
+ "The Slurm installation on JEDI sets `CUDA_VISIBLE_DEVICES` automatically\n",
+ "so that each spawned process only sees the GPU it should use (see\n",
+ "[Hardware\n",
+ "Configuration](https://apps.fz-juelich.de/jsc/hps/jedi/configuration.html#hardware-configuration)\n",
+ "in the JEDI user documentation). This is not supported for NVSHMEM. The\n",
+ "automatic setting of `CUDA_VISIBLE_DEVICES` can be disabled by setting\n",
+ "`CUDA_VISIBLE_DEVICES=0,1,2,3` in the shell that executes srun. With\n",
+ "`CUDA_VISIBLE_DEVICES` set all spawned processes can see all GPUs\n",
+ "listed. This is automatically done for the `sanitize`, `run` and\n",
  "`profile` make targets.\n",
  "\n",
  "### Advanced Task: Use `nvshmemx_float_put_nbi_block`\n",
@@ -85,7 +85,7 @@
  " repository](https://github.com/NVIDIA/multi-gpu-programming-models)\n",
  " implements the same strategy."
  ],
- "id": "ff43766e-b9d0-47c1-bf8c-242e07bb5239"
+ "id": "17eb86cd-3a72-4214-82c1-84818a0d1eca"
  }
  ],
  "nbformat": 4,