gpuetc

Connect multiple GPUs to a single system for parallel tasks.

user1272047's photo
user1272047
·

3 min read

To connect multiple GPUs to a single system for parallel tasks, such as machine learning, rendering, or cryptocurrency mining, you need a motherboard with sufficient PCIe slots and power delivery capabilities. Here's how you can set up your GPUs without needing four separate motherboards:

Look for motherboards with multiple PCIe slots and support for multi-GPU setups:

High-End Motherboard Options:

  1. ASUS ROG STRIX Z790-E GAMING

    • PCIe Slots: 4 x PCIe x16 (for GPUs, supports different lane configurations).

    • Compatibility: Supports NVIDIA 1050, 1060, 4060, and 4090.

    • Price: ~$350–$400.

    • Key Features:

      • Support for mixed GPU configurations.

      • High power delivery for demanding GPUs like the 4090.

  2. MSI MPG Z790 CARBON WIFI

    • PCIe Slots: 3–4 PCIe x16 slots.

    • Compatibility: Works well with mixed GPUs.

    • Price: ~$350.

    • Key Features:

      • Ample spacing for large GPUs like the 4090.

      • Excellent cooling support.

  3. ASUS Pro WS WRX80E-SAGE SE WIFI (Workstation-Grade)

    • PCIe Slots: 7 x PCIe 4.0 x16 slots.

    • Price: ~$1,000.

    • Key Features:

      • Enterprise-level motherboard for maximum GPU expandability.

      • Ideal for connecting all 4 GPUs in your list.

2. Use a PCIe Riser or Mining Rig Frame

If your motherboard doesn't have enough PCIe slots, you can use PCIe risers or a mining rig frame:

PCIe Riser Solution:

  • What It Does: Allows you to connect GPUs to the motherboard using USB 3.0 or flexible riser cables.

  • How to Use:

    • Install the risers into the PCIe slots on the motherboard.

    • Connect the GPUs to the risers and provide power using a dedicated PSU.

  • Example Product: VER 009S PCIe Riser Kit (~$50 for a set of 6 risers).

Mining Rig Frame:

  • What It Does: Provides a physical structure to house multiple GPUs with risers.

  • How to Use:

    • Combine a mining rig frame with risers to hold and connect all GPUs.

    • Ensure adequate cooling and airflow.

3. Power Supply Requirements

Your setup will need a high-wattage PSU or multiple PSUs to power the GPUs:

Single PSU (High-Wattage):

  • Example: EVGA SuperNOVA 1600W T2 (~$400).

  • What It Does: Powers all GPUs and the motherboard.

Dual PSU Adapter:

  • If you use multiple PSUs, an add2PSU adapter can synchronize them.

4. Software and Configuration

Once the hardware is set up, configure the GPUs for parallel use:

Operating System:

  • Use a multi-GPU-compatible OS like Windows 10/11 or Ubuntu 22.04.

Software Frameworks:

  • CUDA (NVIDIA) for parallel computing.

  • PyTorch or TensorFlow for machine learning.

  • Blender for rendering.

GPU Assignment:

  • Assign specific tasks to each GPU using the software's GPU management settings.

Example Configuration:

  1. Hardware:

    • Motherboard: ASUS ROG STRIX Z790-E.

    • PCIe Risers: VER 009S PCIe Riser Kit.

    • GPUs: NVIDIA 1050, 1060, 4060, 4090.

    • PSU: EVGA SuperNOVA 1600W T2.

  2. Software:

    • CUDA for compute tasks.

    • Set up rendering software (e.g., Blender) to use all GPUs in parallel.

This setup allows you to connect and use all four GPUs on a single motherboard effectively.

GPU