neurenix hardware

Manage hardware settings including device selection and benchmarking

The hardware command provides tools for managing hardware configurations, including listing available devices, auto-selecting optimal hardware, and benchmarking performance.

Usage

neurenix hardware [action] [options]

Actions

Action	Description
list	List all available hardware devices (default)
info	Show current device information
auto	Auto-select optimal hardware
select	Manually select a specific device
benchmark	Benchmark available devices

Options

Option	Type	Default	Description
--device	string	None	Device to select (required for select action)
--precision	string	float32	Precision (float32, float16, mixed, int8)
--memory-limit	float	None	Memory limit in GB

Actions in Detail

list - List Available Devices

Displays all available hardware devices with their specifications.

neurenix hardware list

Available hardware devices:
1. NVIDIA GeForce RTX 3090 (cuda)
   - Memory: 24.0 GB
   - Compute capability: 8.6
   - Supported precisions: float32, float16, mixed, int8

2. Intel Core i9-12900K (cpu)
   - Memory: 64.0 GB
   - Compute capability: N/A
   - Supported precisions: float32, float16

info - Current Device Information

Shows detailed information about the currently selected device.

neurenix hardware info

Current device information:
Name: NVIDIA GeForce RTX 3090
Type: cuda
Memory: 24.0 GB
Compute capability: 8.6
Current precision: float32
Memory usage: 2.3 / 24.0 GB

auto - Auto-Select Optimal Device

Automatically selects the best available hardware based on capabilities.

neurenix hardware auto

Auto-selecting optimal hardware...
Selected device: NVIDIA GeForce RTX 3090 (cuda)
Memory: 24.0 GB
Precision: float32
Updated configuration in config.json

select - Manually Select Device

Manually choose a specific device.

neurenix hardware select --device cuda:0

Selecting hardware: cuda:0
Selected device: NVIDIA GeForce RTX 3090 (cuda)
Memory: 24.0 GB
Precision: float32
Updated configuration in config.json

benchmark - Benchmark Devices

Run performance benchmarks on all available devices.

neurenix hardware benchmark

Benchmarking available hardware...

Benchmark Results:

NVIDIA GeForce RTX 3090:
  - Inference time: 2.3456 ms
  - Training time: 12.5432 ms
  - Memory usage: 3.45 GB
  - Throughput: 856.32 samples/sec

Intel Core i9-12900K:
  - Inference time: 45.6789 ms
  - Training time: 234.5678 ms
  - Memory usage: 2.12 GB
  - Throughput: 42.67 samples/sec

Recommended device: NVIDIA GeForce RTX 3090

Examples

List available hardware

neurenix hardware list

Auto-select with mixed precision

neurenix hardware auto --precision mixed

Auto-selecting optimal hardware...
Selected device: NVIDIA GeForce RTX 3090 (cuda)
Memory: 24.0 GB
Precision: mixed
Updated configuration in config.json

Select specific GPU

neurenix hardware select --device cuda:1 --precision float16

Selecting hardware: cuda:1
Selected device: NVIDIA GeForce RTX 3080 (cuda)
Memory: 10.0 GB
Precision: float16
Updated configuration in config.json

Select CPU with memory limit

neurenix hardware select --device cpu --memory-limit 16.0

Selecting hardware: cpu
Selected device: Intel Core i9-12900K (cpu)
Memory: 64.0 GB
Precision: float32
Updated configuration in config.json

Benchmark with specific precision

neurenix hardware benchmark --precision float16

Benchmarking available hardware...

Benchmark Results:

NVIDIA GeForce RTX 3090:
  - Inference time: 1.2345 ms
  - Training time: 6.7890 ms
  - Memory usage: 1.89 GB
  - Throughput: 1624.56 samples/sec

Recommended device: NVIDIA GeForce RTX 3090

Precision Modes

float32 (Default)

Standard 32-bit floating point precision:

• Highest accuracy
• More memory usage
• Slower computation

neurenix hardware select --device cuda:0 --precision float32

float16

Half-precision floating point:

• Good accuracy
• Reduced memory usage
• Faster computation on modern GPUs

neurenix hardware select --device cuda:0 --precision float16

mixed

Mixed precision training:

• Combines float32 and float16
• Best balance of speed and accuracy
• Recommended for modern GPUs

neurenix hardware auto --precision mixed

int8

8-bit integer quantization:

• Lower accuracy
• Minimal memory usage
• Fastest inference
• Best for deployment

neurenix hardware select --device cuda:0 --precision int8

Configuration Updates

When using auto or select actions, the hardware settings are automatically saved to config.json:

{
  "hardware": {
    "device": "cuda:0",
    "precision": "mixed"
  }
}

Use Cases

1. Initial setup

Auto-select optimal hardware when starting a new project:

neurenix init --name my-project
cd my-project
neurenix hardware auto

2. Multi-GPU selection

Select a specific GPU in multi-GPU systems:

neurenix hardware list
neurenix hardware select --device cuda:1

3. Performance optimization

Benchmark devices to find the best option:

neurenix hardware benchmark --precision mixed
neurenix hardware select --device cuda:0 --precision mixed

4. Resource-constrained training

Limit memory usage for shared systems:

neurenix hardware select --device cuda:0 --memory-limit 8.0 --precision float16

5. CPU fallback

Switch to CPU when GPU is unavailable:

neurenix hardware select --device cpu

Error Handling

No GPU available

neurenix hardware select --device cuda:0

Error managing hardware: No CUDA devices available

Invalid device

neurenix hardware select --device invalid

Error managing hardware: Device 'invalid' not found

Missing required option

neurenix hardware select

Error: --device is required for 'select' action.

Best Practices

1. Use auto-selection for new projects

neurenix hardware auto --precision mixed

2. Benchmark before production

Test different configurations to find the optimal setup:

neurenix hardware benchmark --precision float32
neurenix hardware benchmark --precision float16
neurenix hardware benchmark --precision mixed

3. Monitor device information

Regularly check device utilization:

neurenix hardware info

4. Use mixed precision on modern GPUs

For GPUs with Tensor Cores (NVIDIA Volta, Turing, Ampere):

neurenix hardware select --device cuda:0 --precision mixed

5. Document hardware configurations

Keep track of hardware settings in your experiments:

neurenix hardware info > experiment_logs/hardware_config.txt

See Also

• Run command - Train with selected hardware
• Optimize command - Optimize for specific hardware
• Monitor command - Monitor hardware usage during training