Research & Papers

VaultxGPU slashes blockchain proof plotting from 45 min to 45 sec

GPU acceleration delivers 59x speedup for Proof-of-Space consensus plotting.

Deep Dive

Researchers Samuel Taiwo Fatunmbi, Om Amit Gandhi, and Luke Logan have released VaultxGPU, a GPU-accelerated extension of the VaultX Proof-of-Space plotter. Traditional Proof-of-Space (like Chia) replaces energy-intensive computation with storage, but plot generation is still bottlenecked by CPU hashing throughput. VaultxGPU offloads the Blake3 hashing pipeline entirely to the GPU using custom kernels—CUDA for NVIDIA hardware and SYCL for AMD and Intel GPUs. By keeping the entire Table 1 in GPU VRAM and fusing sort and match stages into a single kernel, data movement is minimized.

The results are striking: on a K=31 plot (roughly 100 GB of disk space), the SYCL GPU implementation completes in just 45.4 seconds, a 59.2x speedup over the single-threaded CPU baseline (2688 seconds). Even against the best 384-thread CPU configuration, the GPU solution dominates. This breakthrough confirms that GPU acceleration is the correct path for scaling Proof-of-Space plotting beyond what CPU parallelism can realistically achieve, promising faster farmer onboarding and more efficient blockchain consensus with lower energy overhead.

Key Points
  • VaultxGPU achieves 59.2x speedup over single-threaded CPU for Proof-of-Space plotting.
  • K=31 plot completes in 45.4 seconds on GPU versus 2688 seconds on single-threaded CPU.
  • Supports NVIDIA (CUDA) and AMD/Intel (SYCL) GPUs with fused sort/match kernels for minimal data movement.

Why It Matters

Faster, more energy-efficient blockchain consensus plotting could reduce barriers to decentralized storage networks and lower power consumption.

📬 Get the top 10 AI stories daily