Since April 2, 2026, my RTX 5070 spontaneously crashes every 1–12 hours. Both displays lose signal simultaneously, GPU fans ramp to full speed, and the system requires a hard power cycle to recover. After hard reset, the system often hangs at the GDM login screen (GPU in dirty state); a second reboot recovers.
This has occurred 5+ times across both X11 and Wayland sessions, often at idle with minimal GPU load (24W, 40°C).
Root Cause: GSP RPC Timeout
Kernel logs show the GPU System Processor (GSP) stops responding to RPC calls:
NVRM: _issueRpcAndWait: rpcSendMessage failed with status 0x0000000f for fn 10 sequence 38595!
NVRM: rpcRmApiFree_GSP: GspRmFree failed: hClient=0xc1d00086; hObject=0xbeef0100; status=0x0000000f
NVRM: nvAssertOkFailedNoLog: Assertion failed: GPU lost from the bus NV_ERR_GPU_IS_LOST (0x0000000F) @ mem_mgr_gm107.c:283
nvidia-modeset: ERROR: GPU:0: Failed detecting connected display devices
nvidia-modeset: ERROR: GPU:0: Failed to query display engine channel state: 0x0000ca7e:4:0:0x0000000f
On X11 sessions, this manifests as:
(EE) NVIDIA(0): Failed to allocate push buffer
(EE) NVIDIA(0): Error recovery failed.
Fatal server error: Failed to recover from error! No XID errors are reported before or during the crash. No MCE or PCIe AER errors either.
What I’ve Tried (none prevented the crash)
Switched from X11 to Wayland — crash still occurs (rules out Xorg DDX)
Attempted switch from open to proprietary kernel module — both packages build identical open modules for Blackwell
Additional Issue: 595.58.03 Does Not Support This GPU
I attempted to upgrade to 595.58.03 (.run installer from nvidia.com). It does not recognize device 2f04:
(EE) NVIDIA(GPU-0): The NVIDIA GPU at PCI:1:0:0 is not supported by the 595.58.03 NVIDIA driver.
This is unexpected since nvidia.com recommends 595.58.03 for RTX 5070. Is device 2f04 an early revision that has been dropped from 595, or is this a PCI ID table omission?
Questions for NVIDIA
Is there a known GSP firmware issue for device 2f04 (RTX 5070) on 580.126.18?
Is 580.142 available as a .run installer for this GPU? Would it address this GSP timeout?
Why does 595.58.03 not recognize device 2f04?
Is this likely a hardware defect warranting RMA?
Attached
nvidia-bug-report.log.gz — captured immediately after a Wayland crash (April 7, 2026)
Crashes constantly within an hour, no matter the activity.
Crashes in BIOS as well.
New DP cable did not help. Using the other DP ports on the GPU does not help.
Found a solution that at least stops the crashes: connect to the onboard GPU and rely on passthrough instead. No crashes, but also no VRR, so..
Let me know if I can assist with any logs from my machine. Here’s a snippet that guided me to this post: