4 units of 4090 Build model crash during rendering.
4 units of 4090 Build model crash during rendering.
Recently, I created a rendering setup to aid my projects. Yet, using Redshift in Houdini with Microsoft Windows 11 caused crashes. The system would freeze the screen or shut down abruptly. Details: OS is Microsoft Windows 11 Home, 64-bit CPU from AMD Ryzen Threadripper PRO 5955WX, 16-core 4.0GHz processor. Motherboard is Asus Pro WS WRX80E-SAGE SE WIFI, CPU cooler is Noctua NH-U9 TR4-SP3, RAM is 4x Kingston 32GB DDR4 ECC at 3200MT/s. Storage includes a 1TB M.2 NVMe SSD and a 4TB SATA III SSD from Samsung. Power supply is a Thermaltake Toughpower GF 3 1650W with 80 Plus Gold. GPUs are four TUF Gaming GeForce RTX 4090 OC Edition 24GB GDDR6X units. I updated the BIOS and checked for Windows updates, but the problem continues. Unfortunately, I can’t send screenshots because the PC stops working. Temperatures are normal per HWMonitor readings. The rig is air-cooled and open, maintaining safe temps. To fix, I reduced GPUs from four to three to test stability. I also verified power draw stays within limits. Any advice or solutions would be greatly appreciated. If others have faced similar issues, please share your experience. Thank you for reading, and I look forward to your help!
Hey there, glad you reached out! I've got a setup with two power supplies—A and B. Supply A feeds the motherboard and connects two GPUs via a dedicated cable, while supply B powers another two GPUs plus an extra input for the PCI lanes. Right now I'm running a 3xGPU configuration, which hasn't caused any crashes yet. I'll stick to the same load conditions as before. The PSU B was blowing hot, so I added a motherboard splitter to calm it down. Let me know if you need any more details!
Reviewed several 4090 rendering discussions noting GPU power draw around 300-320W. If a boot error occurs, it might point to a PCIe issue rather than a PSU fault. Possible causes include signal integrity issues when voltage spikes near the power supply or close proximity to MB PCI-e slots. Clarifying earlier, the focus now is on verifying PSU A can support two GPUs and the motherboard. A search revealed many similar cases, including one here: https://forum.level1techs.com/t/help-wit...ver/190742
Thank you for your feedback. I’ll proceed with testing using 3 and 2 GPUs. Also, reviewing the PSU voltage output could help identify any issues. Would you like me to suggest a program for logging information before crashes? It should write data to a log file.
Please note the PCI lanes are Express 4.0, which should be compatible with the 4090 GPUs. The PCI extenders are set to x16 4.0 to match the board’s slots.
Interesting point from Jeppes about bios at 3.0. Clarification on why 3.0 is better than 4.0 in this context. Remake from Likwid on signal interference shows tight cable connections and close proximity to the PSU. In typical setups, shielding is used to guard against interference.
Signal quality depends on maintaining perfect connections at higher frequencies. Reducing to 3.0 has proven effective for many setups. Riser cables also benefit from this adjustment. For rendering devices, 3.0 and 4.0 are less critical. Consider adding a small gap between risers to prevent them from touching, as this improves stability. Be mindful of right angle cables and adapters, which often fail easily.
Thanks for the helpful response. I'm a bit amazed by how dependable the right angle cables are. I expected them to perform better. I might have to test some of the points mentioned earlier. I'll keep an eye on the updates and let you all know soon. If anything new comes up, please advise me to fix the build to prevent these problems.