GPU Low Profile & Single Slot RTX 4060/4070 (Frankenstein) from Taobao

[DevinKit]

Seems disabling ASPM in the bios has stopped my crashing. This felt like a wild goose chase for me, hope it helps someone else
Going to tinker with the other drivers, currently using the vendor one linked by sicksorrow

 

[DevinKit]

Something that helped me diagnose this was running a nividia-smi query for watts drawn continuously and writing to a log file while installing the driver.
was only showing 30w before "gpu is lost"

 

[REVOCCASES]

Got three specific drivers to get to the desktop.
The vendor, thank you Sicksorrow
Luke's NVcleanermethod
Frankendriver

All lead to the same result, a crash about 3 seconds on the desktop. Not really sure where to start until I get that replacement adapter.
I don't crash on the iGPU if I boot with HDMI connected there first.
For now I'm just going to do what I can for now
Update bios
Update windows (idk why this would be the issue)
Try different cables, DP and HDMI
Last if needed, use a linux distro like Luke suggested

If you have any ideas they would be appreciated! Hope I can help someone in the future!

When I ordered my A2000m(8GB), the manufacturer told me that their cards may be not compatible with newer Intel desktop CPUs. Do you have a system with AMD CPU to check if the card would run properly?

 

[LukeD]

deleted - replied to old message

 

[rhinca2026]

Strange issue with Abovetop 4070 GPU, any pointers or clues ? Is this a hardware issue

• Card: Abovetop-branded desktop PCIe conversion card built around an NVIDIA GeForce RTX 4070 Max-Q / Mobile GPU die (chip: AD106M)
• PCI device ID: 10de:1731
• Subsystem ID reported by the OS: Acer Incorporated [ALI] Device 1731 (suggests the GPU die was sourced from an Acer laptop model)
• The card is recognized correctly at the PCI level by every operating system and driver tested — the device ID matches, and NVIDIA's drivers correctly identify it as an AD106M RTX 4070 Max-Q. The failure occurs after recognition, during hardware initialization.

Summary of the fault​

The GPU never successfully initializes under any NVIDIA Linux driver tested. nvidia-smi consistently reports either "No devices were found" or fails to communicate with the driver at all. The kernel log shows the same class of low-level hardware initialization failure recurring across every configuration tested:

NVRM: kbusVerifyBar2_GM107: MMUTest BAR0 window offset 0x70e000 returned garbage 0x0
NVRM: nvAssertOkFailedNoLog: Assertion failed: Generic memory error [NV_ERR_MEMORY_ERROR] (0x00000072)
NVRM: RmInitNvDevice: *** Cannot initialize the device
NVRM: RmInitAdapter: RmInitNvDevice failed, bailing out of RmInitAdapter
NVRM: GPU 0000:xx:00.0: RmInitAdapter failed! (0x24:0x72:1220)
This is a hardware-level self-test performed by NVIDIA's own driver during device initialization (verifying that a diagnostic memory window through the GPU's BAR0 register aperture reads back correctly). It is failing consistently, indicating the GPU itself is not responding correctly to this internal check.

Operating system and driver stack. Tested under two completely independent environments:

1. TrueNAS SCALE 25.10.6, using NVIDIA's open-source kernel module (version 580.173.02).
2. A clean, official Ubuntu 26.04 LTS installation (Live environment), using NVIDIA's proprietary/closed-source driver (nvidia-driver-580), freshly compiled via DKMS specifically for that system's kernel.

 

[LukeD]

Try Batocera linux. And see If the GPU works there. They did some work (equivalent to nvcleanstall) but on linux. Maybe you can life the driver from batocera and use it on other linux distorts ?