|
| 1 | +name: Test - Multi-GPU Runners |
| 2 | + |
| 3 | +on: |
| 4 | + workflow_dispatch: |
| 5 | + push: |
| 6 | + paths: |
| 7 | + - ".github/workflows/test-g6-runners.yaml" |
| 8 | + |
| 9 | +jobs: |
| 10 | + test-1gpu-runner: |
| 11 | + runs-on: gpu-1gpu-runner |
| 12 | + steps: |
| 13 | + - name: Checkout |
| 14 | + uses: actions/checkout@v4 |
| 15 | + |
| 16 | + - name: Job Info |
| 17 | + run: | |
| 18 | + echo "=== 1-GPU Runner Test ===" |
| 19 | + echo "Runner: $(hostname)" |
| 20 | + echo "Timestamp: $(date -u)" |
| 21 | +
|
| 22 | + - name: Check GPU Info |
| 23 | + run: | |
| 24 | + echo "=== GPU Information ===" |
| 25 | + docker run --rm --gpus=all \ |
| 26 | + nvidia/cuda:12.2.0-base-ubuntu22.04 \ |
| 27 | + nvidia-smi --query-gpu=index,name,memory.total,memory.used --format=csv |
| 28 | +
|
| 29 | + - name: Check Node Resources |
| 30 | + run: | |
| 31 | + echo "=== Node Information ===" |
| 32 | + echo "Hostname: $(hostname)" |
| 33 | + echo "CPU cores: $(nproc)" |
| 34 | + echo "Memory: $(free -h | grep Mem | awk '{print $2}')" |
| 35 | +
|
| 36 | + - name: Run GPU Test |
| 37 | + run: | |
| 38 | + docker run --rm --gpus=all \ |
| 39 | + -e CUDA_VISIBLE_DEVICES=0 \ |
| 40 | + nvidia/cuda:12.2.0-base-ubuntu22.04 \ |
| 41 | + bash -c " |
| 42 | + echo '=== 1-GPU Test ===' |
| 43 | + nvidia-smi |
| 44 | + echo '' |
| 45 | + echo '✅ 1-GPU test completed' |
| 46 | + " |
| 47 | +
|
| 48 | + test-2gpu-runner: |
| 49 | + runs-on: gpu-2gpu-runner |
| 50 | + steps: |
| 51 | + - name: Checkout |
| 52 | + uses: actions/checkout@v4 |
| 53 | + |
| 54 | + - name: Job Info |
| 55 | + run: | |
| 56 | + echo "=== 2-GPU Runner Test ===" |
| 57 | + echo "Runner: $(hostname)" |
| 58 | + echo "Timestamp: $(date -u)" |
| 59 | +
|
| 60 | + - name: Check GPU Info |
| 61 | + run: | |
| 62 | + echo "=== GPU Information ===" |
| 63 | + docker run --rm --gpus=all \ |
| 64 | + nvidia/cuda:12.2.0-base-ubuntu22.04 \ |
| 65 | + nvidia-smi --query-gpu=index,name,memory.total,memory.used --format=csv |
| 66 | +
|
| 67 | + - name: Check Node Resources |
| 68 | + run: | |
| 69 | + echo "=== Node Information ===" |
| 70 | + echo "Hostname: $(hostname)" |
| 71 | + echo "CPU cores: $(nproc)" |
| 72 | + echo "Memory: $(free -h | grep Mem | awk '{print $2}')" |
| 73 | +
|
| 74 | + - name: Run GPU Test |
| 75 | + run: | |
| 76 | + docker run --rm --gpus=all \ |
| 77 | + -e CUDA_VISIBLE_DEVICES=0,1 \ |
| 78 | + nvidia/cuda:12.2.0-base-ubuntu22.04 \ |
| 79 | + bash -c " |
| 80 | + echo '=== 2-GPU Test ===' |
| 81 | + nvidia-smi |
| 82 | + echo '' |
| 83 | + echo 'GPU 0 and GPU 1 should be visible' |
| 84 | + echo '✅ 2-GPU test completed' |
| 85 | + " |
| 86 | +
|
| 87 | + summary: |
| 88 | + needs: [test-1gpu-runner, test-2gpu-runner] |
| 89 | + runs-on: ubuntu-latest |
| 90 | + steps: |
| 91 | + - name: Test Summary |
| 92 | + run: | |
| 93 | + echo "=== Multi-GPU Runner Test Complete ===" |
| 94 | + echo "✅ 1-GPU runner test: Success" |
| 95 | + echo "✅ 2-GPU runner test: Success" |
| 96 | + echo "Timestamp: $(date -u)" |
0 commit comments