Neutral evaluation of diverse #zkVM implementations reveals insights into their design and performance. ZisK analysis completed. @jbaylina @ziskvm 🧐 In addition to the amazing emulator, ZisK has the following features: 🔹 CPU/GPU Parallelism: #CPU handles witness generation while #GPU manages proof generation, running concurrently. 💻✨ 🔹 Near 100% GPU Utilization: Thanks to independent circuits (main, mem, ram, etc.) and task distribution via multi-threading + multi-streaming. 🚀 Top 3 GPU Kernels (Time Consumption): 🔹 computeExpressions: 40%. ⏱️ 🔹 br_ntt_8_steps: ~20%. ⏱️ 🔹 linear_hash_gpu_coalesced_2: ~15%. ⏱️ Optimization Opportunities: 🔹 For br_ntt_8_steps: · Reuse twiddle factors across kernels (currently regenerated each time). ♻️ · Add a native-to-bit-reverse NTT kernel to eliminate redundant bit-reverse steps during commits. ⚙️ 🔹 For computeExpressions: · GPU resources are fully utilized. Future gains may require optimizing the proving key's ops for efficiency. 🔍 Guess which zkVM we'll analyze next? 🤔 @eth_proofs