Several GPUs on a server are responsible for the issue.
Several GPUs on a server are responsible for the issue.
You're dealing with a tricky setup. Since you can't test multiple PSUs easily, here’s how you can isolate the issue:
- Connect only one GPU at a time to the motherboard and power it on.
- Use a logic analyzer or multimeter to check for voltage spikes or irregularities when the GPU powers up.
- If the second GPU works but the first doesn’t, the problem is likely with the first PSU.
- Try swapping the two GPUs one by one while monitoring the PSU connections and any error codes.
- If possible, use a known-good PSU to test if the issue is with the original unit or the power supply itself.
Here are the specifications:
- PSU: 650W, 80+ Gold, 2x MOSFETs
- Case: Aorus AX2000
- Motherboard: ASUS TUF Gaming B550M-HDMI
- CPU: Intel i7-12700K
- RAM: 16GB DDR5-600
- Storage: M.2 NVMe SSD, 512GB
- PSU Type: 80+ Bronze certified
Regarding PSU reliability: A decent PSU should last long without failing. Look for models with high-quality components and good thermal management. Popular brands include MSI, ASUS, and Cooler Master.
PCIE slot power connectors are available for testing. Using a continuity meter in the right mode, check for shorts at 12V and 3.3V. Connect a probe to a ground point—like an HDMI video output—and link the other pins to the external brobe and the 12V supply from your power source. A zero resistance reading with a long beep indicates a short circuit. This guide references the PCIE X16 Pinout.pdf document.
Unclear why you'd need three RTX 4090s on the same power source, no matter how powerful it is. This configuration calls for a motherboard with an aux VGA 6,8 pin to support the PCIe bus, because each card can draw up to 75W from that connection. Only boards I know that fit are EVGA models and possibly HEDT ones.
The power source is a Hela 2050 Platinum, the processor is an i7 13700K with 128GB RAM. It's running Ubuntu 22.04 without a graphical interface. I'm ordering one tomorrow and will follow the instructions below to test for continuity. Hope this helps with my search for solutions. We rely heavily on GPUs, so each server has either three 4090s or four 3090 TIs. The motherboard is an ASUS Z690 Prime inside a server chassis.