Hey Midnight community 
We just open-sourced a new piece of infra we think is directly tied to whether programmable privacy feels fast in practice:
nocy GPU Proof Server - a Midnight proof server adapted for GPU proving via CUDA, benchmarked on an NVIDIA GeForce RTX 5090.
Repo: https://github.com/Nocy-io/nocy-gpu-proof-server
Why this matters
In privacy-native apps, proof generation latency is one of the biggest UX bottlenecks. Speeding up proving doesn’t change the privacy model, it lowers the “privacy tax” so real flows feel responsive.
Who benefits:
-
Users: shielded actions feel lighter (less waiting on proofs)
-
Builders: privacy apps stop feeling “proof-heavy”
-
Infra providers: a clearer path to run high-performance proving as part of a stack
Benchmarks (from our initial setup)
The README includes a comparison between RTX 5090 (GPU) and AMD EPYC 7642 (CPU), averaged over 10 runs:
-
k17: 4.4s (GPU) vs 12.3s (CPU)
-
k18: 8.8s (GPU) vs 24.6s (CPU)
That’s ~2.8× faster proving vs CPU-only in this setup.
API routes
The server exposes:
-
POST /prove— generate a zero-knowledge proof -
POST /check— validate preimages -
GET /health— health check
Related open source (already public)
We’re doing staged OSS and already have wallet infra public:
What we’d love from the community
If you have different GPUs / driver versions / server setups, we’d really value:
-
benchmark comparisons (and circuit sizes tested)
-
build notes (what worked cleanly)
-
issues/PRs for stability + reproducibility