I inherited a Lenovo Legion tower with an RTX 4090 that had been freezing and crashing during games, especially during visually demanding moments. I moved the 4090 and other components into a new case, replaced the original motherboard with an ASRock X870 Riptide, and installed a Ryzen 9950X with 64 GB of RAM and a 2 TB NVMe drive. The problems continued even after replacing the power supply and GPU power connection.
Overwatch reports that it lost connection to the GPU, while Diablo 4 produces a D3D error. Sometimes the screen briefly goes black while the fans ramp up and the audio continues; other times the entire system freezes and needs a hard reboot. The crashes are unpredictable and often happen after several hours of play, particularly when a game loads a demanding scene. FurMark can run successfully, however, which makes the cause less obvious.
The system uses four RAM sticks, and I have not changed any BIOS settings beyond updating the firmware. When rebuilding the system, I reset Windows on the existing Lenovo installation rather than performing a completely fresh installation. Could the old OEM Windows installation be causing this? Should I test with lower RAM or GPU clocks, and could four DIMMs be unstable on this platform? I would also appreciate reliable ways to test the memory, SSD, and graphics card before assuming the 4090 needs to be replaced.
3 Answers
A FurMark run does not completely clear the GPU. It may stress the card differently from games, which can expose driver, VRAM, transient-power, or clock-stability problems. Try a clean NVIDIA driver installation using DDU, then test a small GPU core and memory underclock with a modest power limit. If that makes the crashes disappear, it points toward marginal stability, heat, or power delivery rather than proving the card is permanently defective. Also check GPU temperatures, hotspot temperature, and whether the 4090 power connector is fully inserted without sharp bends near the plug.
Four DIMMs can definitely make memory stability harder, especially on a high-end Ryzen platform. Start by disabling EXPO/XMP and running the RAM at its default JEDEC speed. If the crashes stop, test each stick and the full kit with MemTest86 or several passes of Karhu/HCI MemTest. You can also try running only two matched sticks in the recommended motherboard slots. I would avoid changing RAM voltage manually until you know whether the system is stable at stock settings.
I would not buy another Windows license just because the installation came from Lenovo. A reset may leave drivers or configuration behind, though, so a truly clean Windows installation on the NVMe drive is worthwhile for troubleshooting. Install only the chipset, motherboard, and NVIDIA drivers at first and test before adding extra utilities. Also check the SSD with the manufacturer’s health tool and run separate CPU, memory, and GPU tests. If the machine still hard-freezes with stock BIOS settings, two RAM sticks, a clean OS, and known-good power cabling, the GPU or another hardware component becomes much more likely.
The original Lenovo system used a 13900K, but the current build has a Ryzen 9950X. I’ll test the RAM and SSD separately and may do a clean installation rather than relying on the reset.
That makes sense. I’ll try two sticks with EXPO disabled first, then test the modules individually if the crashes continue.