Configuring Dual Corsair HX1600i PSUs with an Add2PSU Relay for Synchronized ATX Startup

Your workstation blacks out mid-inference. You are halfway through a 128k context window when the entire rig hard-reboots. This isn't a Linux kernel panic or a driver issue. It is your primary PSU hitting its Over-Current Protection (OCP) limit. When an RTX 5090 spikes during heavy tensor operations, that transient draw can blow past the headroom of a single 1600W unit. If you are building a multi-GPU node for local LLM workloads, trying to run it off one power supply is a massive architectural mistake. It guarantees instability.

I have wasted too many hours troubleshooting rigs that fail exactly when they become useful: during high-load reasoning loops. I have seen $5,000 GPU clusters bricked because an architect ignored the math of transient spikes and failed to synchronize ATX startup.

Why do single PSU setups fail under LLM workloads?

Single PSUs cannot handle the massive transient power spikes generated by multiple RTX 5090s during peak inference. The instantaneous draw often exceeds the unit's OCP threshold, triggering an immediate shutdown. A dual-PSU configuration provides the necessary amperage headroom to absorb these spikes without tripping the protection circuits.

Single PSUs struggle because LLM inference is not a steady-state load. As the GPU memory controller and tensor cores cycle through massive context windows, they create transient loads that can briefly double the card's rated TDP. If you're trying to run dual RTX 5090s off a single Corsair HX1600i, the math fails. An RTX 5090 might sit at a 450W-600W base TDP, but those millisecond spikes hit much harder. Pair two of those with a high-core-count Threadripper or EPYC CPU and you're slamming the single unit's OCP threshold.

Peak transient loads hit 2000W. Your 1600W unit cannot handle that. You need to distribute the load across two independent power rails. Use an Add2PSU delay relay to manage the 12V-2x6 ATX 3.1 startup sequence. Do not use a cheap SATA-to-Molex adapter for the trigger voltage. Improper power sequencing will wreck your hardware stability.

The Verdict: Provides massive headroom and digital monitoring to ensure power delivery remains stable under heavy AI workloads.
Best For: Professional MLOps engineers building dual-GPU workstations.

How do you synchronize two PSUs using an Add2PSU relay?

You connect the Add2PSU relay to your primary PSU's SATA or Molex lead and the secondary PSU's trigger cable. When the motherboard sends the ATX 'power on' signal to the first unit, the relay closes the circuit for the second. This ensures both units spin up at the exact same millisecond, preventing voltage deltas.

Don't ignore the hardware details. If you mess this up, you'll fry your silicon.

The Verdict: If you are running more than one 5090, stop trying to save money on power. Buy two HX1600i units, get a proper relay, and sync your startup.
Best For: Local LLM workstations and multi-GPU inference nodes.

Calculating total wattage for dual RTX 5090 AI workstations

Calculate your power budget by adding the CPU's peak draw to the combined transient spike values of both GPUs. You must include a 30% buffer above these peak numbers to keep the PSU in its efficiency sweet spot. This headroom prevents voltage sag and unexpected shutdowns during heavy compute loads.

Stop looking at the sticker TDP. It is a lie for your purposes. If you are building a rig with two RTX 5090s, you are building a high-density compute node, not a gaming PC. You need to account for transient spikes—those millisecond bursts of power that can trip a lesser unit's over-current protection.

Don't make the mistake of reaching for a Corsair RM1000x. It lacks the digital telemetry and the current capacity required for the 12V-2x6 rails on these 2026-spec cards. The HX1600i is the baseline here. It handles sustained, heavy wattage without the voltage instability found in mid-range gear.

Running a dual-PSU setup? Utilize an Add2PSU switching module. It synchronizes the startup sequence. Without it, your secondary unit might try to pull current before the primary unit has stabilized the system rails, and that is a recipe for a dead motherboard.

The Verdict: The gold standard for high-wattage, high-stability AI workstation builds.
Best For: Multi-GPU builders requiring massive transient headroom.

How to wire an Add2PSU relay for synchronized ATX startup

An Add2PSU relay triggers the secondary power supply's internal rails the instant your primary motherboard signal is detected. This hardware-level handshake ensures both units are active before the CPU begins its POST sequence.

If you're running a dual-PSU rig, you can't just plug the second unit into the wall and expect it to play nice. Without a relay, your secondary unit sits idle while your primary unit tries to boot. You need a physical bridge to pass the 'Power On' signal from the motherboard to the second rail.

Step 1 — Prepare the trigger signal

Find a peripheral connector on your primary PSU that isn't doing anything vital. SATA or Molex are your standard targets here.

Step 2 — Connect the Add2PSU relay

Plug the SATA-to-Molex trigger cable from the Add2PSU kit directly into your primary Corsair HX1600i.

Expert Note: Watch your pinouts. If you're running modern 12V-2x6 ATX 3.1 hardware alongside older relay logic, an incorrect SATA-to-Molex voltage configuration will cause a signal mismatch. Check your rail voltages before you flip the switch.

Step 3 — Connect the secondary PSU

Take the 24-pin ATX cable from your secondary unit and plug it into the Add2PSU relay. This allows the relay to pass the 'Power On' signal straight through to the second unit.

Step 4 — Verify synchronization

Flip the switch on the primary PSU. Hit the case power button. Both units should kick in at the same time.

The Verdict: Ensures both power supplies engage at the exact same millisecond to prevent component damage.
Best For: Users building dual-PSU multi-GPU nodes.

Common mistakes in dual PSU ground sharing and voltage spikes

Dual PSU configurations fail when the two units lack a shared electrical ground. This creates a potential difference that forces current through sensitive motherboard traces or PCIe lanes. To prevent hardware death, plug both units into the same high-quality PDU.

If you're running two power supplies, don't assume they're talking the same language just because you used a relay. Most cheap Add2PSU modules only sync the startup signal; they fail to electrically bond the grounds. If PSU A is on one wall circuit and PSU B is on another, you've just built a bridge for stray current to hunt for a path through your motherboard. That's how you kill an RTX 5090.

Keep it simple. Plug both units into the same PDU. This ensures they share a common ground reference from the start. And for god's sake, don't try to be clever with your cabling. Never use a single cable to bridge power between the two units. Every component must pull its full current from its dedicated supply. If your GPU is pulling massive wattage, split the PCIe power cables between the two units. Let the motherboard handle the ground reference through its own architecture.

Expert Gotcha: Modern 12V-2x6 ATX 3.1 standards are less forgiving than the old stuff. If you're using a basic SATA-to-Molex trigger, you're asking for timing issues. Use a dedicated Add2PSU delay relay. You need that primary PSU to stabilise its rails before the secondary unit starts drawing load. Otherwise, you're just inviting transient spikes into your high-end silicon.

Testing power stability under peak VRAM load

Run a sustained, high-current workload that stresses both the GPU tensor cores and the VRAM simultaneously. You need to simulate the aggressive power fluctuations seen during LLM inference. A standard benchmark won't cut it.

Once your dual HX1600i setup is wired, don't trust it. You haven't tested it until you've pushed it. If you're running heavy VRAM-intensive workloads, like loading massive model weights, you need to pair that with a heavy GPU compute kernel or a tool like mprime. Use Corsair iCUE or a digital multimeter to monitor your voltage rails.

Watch the 12V rail closely. If it dips below 11.4V during a spike, your capacity or your synchronization is failing. System reboots usually mean a transient spike just hit the combined OCP of your units. Or, you've got a faulty relay connection.

A common expert-level failure mode in dual-PSU builds is an improper trigger signal; ensure your Add2PSU delay relay is utilizing a SATA-to-Molex safe trigger voltage to prevent the secondary unit from attempting to draw current before the primary unit's rails have stabilized.

Recommended hardware for dual PSU multi GPU setups

Running two RTX 5090s requires separate power rails to prevent transient spikes from tripping overcurrent protection. You need two ATX 3.1 compliant power supplies and a physical relay to sync their startup sequences. This configuration splits the heavy lifting between two units to maintain stability during massive inference loads.

The Verdict: The essential component for any professional multi-GPU build.
Best For: High-end AI workstations.

The Verdict: A low-cost, high-reliability solution for dual PSU synchronization.
Best For: DIY workstation builders.

The Verdict: Essential for preventing connector melt during high-load RTX 5090 tasks.
Best For: Next-gen GPU owners.

The Verdict: The most reliable way to power a multi-GPU AI node.
Best For: Serious MLOps professionals and systems architects.

If you're serious about running local LLMs without a hard reboot killing your inference task, stop trying to cram everything onto one power supply. The math doesn't work. The RTX 5090 draws enough current to choke most single-unit setups during peak transients.

Use two Corsair HX1600i units. Connect them via an Add2PSU relay. This setup uses a SATA-to-Molex trigger to ensure the secondary unit only kicks in once the primary is stable. It's a messy build. You'll deal with extra cables and a larger footprint. But it's the only way to handle the power spikes inherent in high-context computing.

"Disclaimer: All third-party product names, logos, and brands referenced in this article are trademarks or registered trademarks of their respective holders. Use of them does not imply any affiliation with or endorsement by them. Features, pricing structures, and specifications are subject to change over time. Systems architects should verify exact parameters directly with current vendor documentations."

Disclaimer: The information in this article is provided for general informational purposes only. Any instructions, settings changes, or technical steps carry inherent risk. Always back up your data and follow the manufacturer's guidance before modifying device settings or configurations. Results may vary based on your specific hardware, device firmware, software version, and environment. You are solely responsible for any changes you make. The author and publisher accept no liability for damage, data loss, or device malfunction arising from following this guidance. Amazon product links are affiliate links — the author may receive a commission on qualifying purchases at no extra cost to you. Prices and availability are subject to change; check Amazon directly for current pricing.