GitHub Actions GPU Runners | NVIDIA GPU Specifications
Machine provides a variety of GPU runners for GitHub Actions to match your specific CI/CD and machine learning workload requirements. Each GitHub Actions GPU runner type comes with pre-installed NVIDIA Device Drivers 555.58, CUDA 12.1.0 and cuDNN 9.2.1, allowing you to start using GPU acceleration immediately without any configuration.
You are always free to install additional drivers, CUDA, or cuDNN versions, or even build your own from source.
GitHub Actions GPU Runners - Available Types
Machine supports all GPU instances currently available on AWS, including NVIDIA GPUs and AWS Inferentia accelerators.
NVIDIA GPU Runners for GitHub Actions
| GPU Type | GPU Memory | CUDA Cores | Tensor Cores | Use Cases |
|---|---|---|---|---|
| T4G | 16GB | 2,560 | 320 | Entry-level ML training, inference |
| T4 | 16GB | 2,560 | 320 | General-purpose ML, computer vision |
| L4 | 24GB | 7,680 | 240 | Balanced training/inference, mid-range ML |
| A10G | 24GB | 9,216 | 288 | Advanced training, larger models |
| L40S | 48GB | 18,176 | 568 | Large model training, high-performance ML |
AWS AI Accelerators
| Accelerator Type | vCPU | RAM | Accelerator Memory | Use Cases |
|---|---|---|---|---|
| TRAINIUM | 8 | 32GB | 32GB | High-performance training |
| INFERENTIA2 | 4 | 16GB | 32GB | Next-gen inference optimization |
Pre-installed GPU Software for CI/CD
Each runner comes with the following software pre-installed:
- NVIDIA Device Drivers 555.58
- CUDA 12.1.0
- cuDNN 9.2.1
- NVIDIA Container Toolkit
- AWS Neuron SDK
GitHub Actions Runner Specifications
Besides the GPU, Machine runners offer configurable CPU and RAM options to match your specific workload requirements.
All GPU Runner Configurations and Pricing
Every GPU type supports multiple CPU/RAM configurations. Default configurations (used when cpu and ram labels are omitted) are marked with bold pricing.
NVIDIA T4G (ARM64, 16GB VRAM)
| vCPU | RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|
| 4 | 8 GB | $0.0037 | $0.0140 |
| 8 | 16 GB | $0.0027 | $0.0185 |
| 16 | 32 GB | $0.0031 | $0.0276 |
NVIDIA T4 (X64, 16GB VRAM)
| vCPU | RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|
| 4 | 16 GB | $0.0048 | $0.0175 |
| 8 | 32 GB | $0.0069 | $0.0251 |
| 16 | 64 GB | $0.0110 | $0.0401 |
NVIDIA L4 (X64, 24GB VRAM)
| vCPU | RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|
| 4 | 16 GB | $0.0057 | $0.0268 |
| 8 | 32 GB | $0.0036 | $0.0326 |
| 16 | 64 GB | $0.0046 | $0.0441 |
NVIDIA A10G (X64, 24GB VRAM)
| vCPU | RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|
| 4 | 16 GB | $0.0126 | $0.0335 |
| 8 | 32 GB | $0.0110 | $0.0404 |
| 16 | 64 GB | $0.0174 | $0.0541 |
NVIDIA L40S (X64, 48GB VRAM)
| vCPU | RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|
| 4 | 32 GB | $0.0161 | $0.0620 |
| 8 | 64 GB | $0.0179 | $0.0747 |
| 16 | 128 GB | $0.0153 | $0.1001 |
AWS AI Accelerators
| Accelerator | vCPU | RAM | Accelerator RAM | $/Min (Spot) | $/Min (On-Demand) |
|---|---|---|---|---|---|
| Inferentia2 | 4 | 16 GB | 32 GB | $0.0025 | $0.0253 |
| Inferentia2 | 32 | 128 GB | 32 GB | $0.0098 | $0.0656 |
| Trainium | 8 | 32 GB | 32 GB | $0.0052 | $0.0448 |
Prices shown are the best available rates across all regions and are subject to change. Additional EBS storage charges apply. See Pricing for full details.
All runners include a 100GB gp3 root volume by default, with configurable size, IOPS, and throughput via runner labels. See Configuration Options for storage label details.
Instance Metrics
Machine runners collect CloudWatch metrics by default, providing real-time visibility into resource utilization for every job. After a job completes, metrics are displayed as sparkline charts on the Machine dashboard.
Collected GPU metrics:
- GPU utilization percentage
- GPU memory utilization and usage
- GPU temperature
- GPU power draw
Collected system metrics:
- CPU utilization
- Memory usage
- Disk read/write operations
- Network bytes in/out
You can control metrics collection per job using runner labels:
runs-on: - machine - gpu=A10G - metrics=true # Enable metrics (default) - metrics_interval=10 # Collect every 10 seconds (default: 60)To disable metrics for a job, set metrics=false. See Configuration Options for full details.
Next Steps
- Learn how to Configure Your Workflows for optimal performance
- Explore Cost Optimization Strategies