RTX 4070 eGPU Working on RTX Spark HP OmniBook

Hardware Hub Laptops & Netbooks
, ,
1

I successfully connected NVIDIA GeForce RTX 4070 to my RTX Spark HP OmniBook X using the AOOSTAR AG02 eGPU dock over USB4.

I’m sharing my setup and workaround for anyone experimenting with external GPUs on Windows 11 ARM64.

TLDR: I extracted NVIDIA’s official 621.01 ARM64 Spark driver with 7-Zip, manually bound the driver to the external RTX 4070, and resolved Device Manager Error Code 43 by setting RM1774520 to DWORD 1 in the external GPU’s driver registry key.

1. Hardware and software

Component Configuration
Laptop HP OmniBook X 14-ks000
Internal GPU NVIDIA RTX Spark N1X
External GPU NVIDIA GeForce RTX 4070, 12 GB
eGPU dock AOOSTAR AG02
Connection USB4
Operating system Windows 11 ARM64
NVIDIA driver 621.01 WHQL
External display Connected directly to RTX 4070

Driver package:

621.01-rtx-spark-notebook-win11-arm64-international-whql.exe

Important: Use the Windows ARM64 NVIDIA driver package, not the standard x64 GeForce desktop driver.

2. Manually installing the driver

With the RTX 4070 installed and powered in the AOOSTAR AG02, I connected the dock directly to the laptop’s USB4 port and connected my external monitor to the RTX 4070.

I downloaded the NVIDIA ARM64 driver package and extracted it using 7-Zip.

The external RTX 4070’s hardware ID was:

PCI\VEN_10DE&DEV_2786

The driver INF I used was:

Display.Driver\nv_hp_woaig.inf

I selected this model from the INF:

NVIDIA RTX Spark N1X (6144-core Blackwell RTX GPU)

Install Driver with Device Maanger

  1. Open Device Manager → Display adapters.

  2. Locate the external Microsoft Basic Display Adapter.

  3. Check Properties → Details → Hardware IDs and confirm PCI\VEN_10DE&DEV_2786.

  4. Select Update driver → Browse my computer for drivers → Let me pick → Have Disk.

  5. I chose nv_hp_woaig.inf in the display driver folder from the extracted package since I have HP.

  6. Select the NVIDIA RTX Spark N1X model.

  7. Accept the compatibility warning if you choose to proceed with this unsupported binding.

Warning: Make sure you’re modifying the external GPU, not the working internal Spark GPU or NPU.

After installation, the driver successfully bound to the external RTX 4070, but Windows reported Error Code 43.

3. Resolving NVIDIA Error Code 43

The workaround that resolved Code 43 on my system was adding this registry value:

Value name: RM1774520
Type:       REG_DWORD
Value:      1

This value needs to be placed under the external GPU’s specific display-driver registry key.

Finding the correct registry key

In Device Manager:

  1. Open the external GPU’s Properties.

  2. Navigate to Details.

  3. Select Driver key from the dropdown.

  4. Identify the corresponding key under:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\

On my system, the complete path was:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\
{4d36e968-e325-11ce-bfc1-08002be10318}\0002

Do not assume your external GPU uses **0002**. The driver key may be different, particularly after reinstalling drivers.

Before proceeding:

  • Verify the registry key belongs to the external GPU.

  • Export the key as a backup.

  • Add RM1774520 as a REG_DWORD with value 1.

  • Restart only the external GPU device.

After applying this change and restarting the external adapter, Code 43 disappeared.

If your eGPU is already working and appears correctly in nvidia-smi, there’s no reason to apply this workaround.

I did not need to modify any INF files, downgrade the driver, change firmware, or disable Secure Boot, driver-signature enforcement, or Memory Integrity.

4. Verification and results

After applying the workaround, I ran some functional checks.

Test Result
RTX 4070 detected by nvidia-smi PASS
12 GB VRAM identified PASS
Windows device problem code 0
CUDA arithmetic kernel PASS — 64/64 correct
Direct3D hardware shader test PASS — 4,096/4,096 pixels
External monitor through RTX 4070 2560 × 1600, ~120 Hz
Internal display through Spark GPU 2880 × 1800, 60 Hz
Internal Spark GPU and NPU Working
Detection after system reboot PASS

Both the internal Spark GPU and external RTX 4070 appeared in nvidia-smi.

The CUDA and Direct3D tests confirmed that the external GPU could execute compute workloads and render graphics, rather than merely appearing in Device Manager.

Device Manager naming issue

One interesting detail: Device Manager shows the external RTX 4070 as “NVIDIA RTX Spark N1X” .

However, NVIDIA’s runtime correctly identifies the physical GPU as an RTX 4070.

Don’t rely solely on the Device Manager display name to distinguish the internal and external GPUs.

Planned testing

Linux Compatibility

I’d like to investigate whether the same AOOSTAR AG02 and RTX 4070 setup can work under Linux on the NVIDIA RTX Spark N1X platform.

Local AI - SLM and LLM Inference

I’d like to explore running Small Language Models (SLMs) and Large Language Models (LLMs) locally using the internal Spark GPU and external Nvidia GPUs (3090,5090)

References and resources

  • CasonPointLLC Reddit post - Snapdragon Surface Pro 11 + 5060 ti - /r/eGPU/comments/1xekn5m/arm_egpu_in_windows_11_with_public_drivers/
  • AleCyriaco/nvidia-error43-fix on github