spn-node calibrate fails because the background moongate-server crashes on startup with an RTX 2080 Ti

[echowill]

Environment:

GPU: NVIDIA GeForce RTX 2080 Ti
Operating System: Windows 11 with Docker Desktop
NVIDIA Driver Version: 571.96
spn-node Image: public.ecr.aws/succinct-labs/spn-node:latest-gpu
moongate Image: public.ecr.aws/succinct-labs/moongate:v5.0.8 (This is the image used by the sp1-gpu container)
Bug Description:
When attempting to run the benchmark using the spn-node calibrate command, the process consistently fails. The client (spn-node) hangs for approximately 60 seconds before panicking with a timeout or connection error.

The root cause has been identified as the immediate crash of the background proving server container (sp1-gpu, which runs the moongate-server image) upon startup.

Steps to Reproduce:
The crash of the moongate-server can be isolated and reproduced directly with the following command, without needing the spn-node client:

docker run -it --rm --gpus all public.ecr.aws/succinct-labs/moongate:v5.0.8 moongate-server

bash

Expected Behavior:
The moongate-server container should start successfully, initialize the GPU, and begin listening for RPC connections from the client.

Actual Behavior:
The moongate-server container crashes immediately upon launch. The only output to the console is a fatal runtime error:

fatal runtime error: Rust cannot catch foreign exceptions, aborting

This indicates that the server process is encountering a low-level (likely C++/CUDA) exception during the GPU initialization phase, which the Rust runtime cannot handle, leading to an immediate abort.

Additional Context & Diagnostics Performed:

The issue is not related to networking. The crash occurs even when both client and server containers are run with --network="host".
The issue is not with the spn-node client. The server container fails even when run standalone.
The issue appears to be a hardware-specific incompatibility between the moongate-server's GPU initialization code and the NVIDIA RTX 2080 Ti under a Windows + Docker environment.