Akash Provider Configurations

Community-maintained repository for Akash provider GPU configurations and hardware feature discovery.

Purpose

This repository serves as the central database for GPU hardware identification on the Akash Network. It enables:

• Automatic GPU Detection: Providers can automatically identify and advertise GPU models
• Accurate Deployment Matching: Tenants can discover and deploy to specific GPU models
• Network-Wide GPU Discovery: Real-time visibility into available GPU hardware across the network
• Proper Resource Pricing: Accurate GPU model identification ensures correct pricing

Repository Structure

provider-configs/
├── devices/
│   └── pcie/           # PCIe device configurations
├── httpServer/         # Configuration server
└── gpus.json          # GPU database (vendor/product IDs)

How to Contribute GPU Configurations

When to Submit

Submit GPU information if:

1. Your GPU model is not listed in gpus.json
2. You're adding new GPU models to your provider
3. Your GPUs aren't being detected correctly

Prerequisites

Before collecting GPU information:

• SSH access to each GPU-equipped node
• provider-services version 0.5.4 or higher (download here)
• jq installed for JSON processing (apt install -y jq)

Step 1: Verify GPU Models

Check if your GPU is already in the database:

1. Visit gpus.json
2. Search for your GPU vendor ID and product ID
3. If found, no submission needed
4. If not found, proceed to Step 2

Step 2: Collect GPU Details

Run this command on each GPU node:

provider-services tools psutil list gpu

Example Output:

{
  "cards": [
    {
      "address": "0000:00:04.0",
      "index": 0,
      "pci": {
        "driver": "nvidia",
        "address": "0000:00:04.0",
        "vendor": {
          "id": "10de",
          "name": "NVIDIA Corporation"
        },
        "product": {
          "id": "1eb8",
          "name": "TU104GL [Tesla T4]"
        },
        "revision": "0xa1"
      }
    }
  ]
}

Step 3: Extract Required Information

From the output, note:

• Vendor ID: "id": "10de" (NVIDIA in this example)
• Product ID: "id": "1eb8" (Tesla T4 in this example)
• GPU Model Name: "name": "TU104GL [Tesla T4]"

Step 4: Submit via Pull Request

1. Fork this repository
2. Edit gpus.json
3. Add your GPU entry in the following format:

{
  "vendor": "10de",
  "device": "1eb8",
  "name": "t4"
}

Field Guidelines:

• vendor: Vendor ID (lowercase hex, without "0x" prefix)
• device: Product ID (lowercase hex, without "0x" prefix)
• name: GPU model name (lowercase, alphanumeric, no spaces)
  • Examples: t4, a100, h100, rtx4090

4. Create a pull request with title: Add GPU: [Model Name]
5. Include the full provider-services tools psutil list gpu output in the PR description

Step 5: Update Provider Configuration

After your PR is merged, update your provider to use the new GPU attributes. See the Provider Attributes Documentation for details.

Naming Conventions

GPU model names should follow these guidelines:

• Lowercase only: a100, not A100
• No spaces: rtx4090, not rtx 4090
• No special characters: h100, not H100-SXM
• Consistent with market naming: Use common model designations

Examples:

• ✅ t4, a100, h100, v100, rtx4090, rtx3090ti
• ❌ T4, A-100, H100 SXM, RTX_4090

Validation

Before submitting, verify:

1. No duplicates: Check if the vendor/device ID combo already exists
2. Valid hex IDs: Vendor and device IDs are 4-digit lowercase hex (without "0x")
3. Lowercase name: Model name is lowercase alphanumeric
4. Valid JSON: Your edit doesn't break JSON formatting

Questions or Issues?

• Documentation: Akash Provider Attributes Guide
• Support: Akash Discord #providers channel
• Issues: Open an issue