The official GPU prover implementation for SP1, written in CUDA.

Profiling

Jaeger

Setup Jaeger:

sudo docker run -it --rm -d -p4318:4318 -p4317:4317 -p16686:16686 jaegertracing/all-in-one:latest

Run a benchmark:

RUST_LOG=debug cargo run --release -p moongate-perf -- --program fibonacci

Nvidia Nsight Systems

Run a benchmark:

RUST_LOG="debug" nsys profile --trace=cuda,nvtx cargo run --release -p moongate-perf -- --program fibonacci --trace nvtx 

Server

Build the server image:

sudo docker build -f Dockerfile.server -t moongate-server .

DOCKER_BUILDKIT=1 docker build -f Dockerfile.server -t jtguibas/sp1-gpu:v4.0.0-rc1 --ssh default=${SSH_AGENT_AUTH_SOCK} .

Run the server:

sudo docker run -e "RUST_LOG=debug" -p 3000:3000 --rm --runtime=nvidia --gpus all moongate-server

Docker Images

Our Docker images are automatically built and pushed to Amazon ECR Public using GitHub Actions. The process is defined in the .github/workflows/docker.yml file.

Build and Release Process

1. The Docker image is built on every push to the main branch and for all pull requests targeting the main branch.
2. The image is always tagged with the Git commit SHA.
3. If the commit has a Git tag:
   • An additional image is pushed with that tag.
   • The image is also tagged as latest.

Accessing the Images

You can browse and pull our Docker images from the Amazon ECR Public Gallery:

https://gallery.ecr.aws/succinct-labs/sp1-gpu

To pull the latest tagged release:

docker pull public.ecr.aws/succinct-labs/sp1-gpu:latest

To pull a specific version (replace <tag> with the desired version):

docker pull public.ecr.aws/succinct-labs/sp1-gpu:<tag>