Expander-GPU: Blasting Past CPU Limits with a 10x Speed Force

Announcing Our New CUDA-Powered Backend: 7–10x Faster ZK Proofs for Ethereum

"The future of blockchain is not just decentralized, it's fast. We're building that future today."

1. The Sound Barrier for Zero-Knowledge Proofs

Zero-Knowledge (ZK) proofs are the engine of blockchain privacy and scalability. But for too long, that engine has been stuck in first gear. Complex proofs on powerful CPUs take precious seconds, even minutes—a lifetime in the world of high-speed finance and seamless user experiences.

To break this barrier, you don't just need a bigger engine; you need a different kind of power.

Enter Expander-GPU: our new, high-performance proof generation backend, engineered to unleash the massive parallelism of NVIDIA GPUs.

2. Meet the Expander-GPU Core

Expander-GPU is more than just an update; it's a ground-up redesign for the age of accelerated computing. Powered by CUDA, it transforms proof generation from a sequential task into a massively parallel process.

Key Features That Redefine Speed:

• Massive Parallelism: We harness thousands of GPU cores to process countless proof computations simultaneously.
• Revolutionary MPI Merge: Our innovative merge feature can compress proofs from thousands of independent computations into a single, compact proof. We've hit a staggering 16,384:1 compression ratio, turning a data flood into a trickle.
• Broad Field Support: Optimized for multiple cryptographic fields, including BN254, Goldilocks, and M31Ext3, making Expander exceptionally versatile.

3. The Main Event: BN254 Performance That Crushes Records

For the Ethereum ecosystem, performance on the BN254 curve is what matters. This is where most ZK applications for Ethereum are built, and it's where the computational bottleneck hurts the most.

Our GPU backend doesn't just inch past the competition; it delivers a knockout. Running on a next-generation NVIDIA 5090 GPU, Expander is 7–10 times faster than our own highly optimized AVX512 backend running on a top-of-the-line AMD 9950X3D CPU.

This isn't a theoretical improvement. It's a practical leap that makes complex, Ethereum-friendly ZK applications faster and more cost-effective than ever before.

Performance Across the Board

The speedup isn't limited to BN254. Our benchmarks show staggering throughput across all supported fields:

4. Why This Is a Game-Changer for Web3

• Unlocks L2 Scalability: Faster, cheaper proof generation means Layer-2 roll-ups can achieve higher TPS and lower fees.
• Enables Complex dApps: Developers can build more sophisticated privacy-preserving applications on Ethereum without worrying about performance bottlenecks.
• Future-Proofs the Ecosystem: With GPU acceleration, the ZK ecosystem is ready to handle the demands of mainstream adoption.

5. Join the Acceleration

Expander-GPU is now available in a binary release. We are committed to pushing the boundaries of what's possible in the ZK space and invite you to join us.

• Try it out: Contact us for access to the binary and run your own benchmarks.
• Learn More: We are grateful to the ICICLE team, whose work on GPU field operations helped make this possible.

The era of waiting is over. The future is accelerated.