Profesora Lite

Major RISC-V zkVM breakthrough! The ZisK team hit 1.5GHz trace generation—10x faster than other RISC-V zkVMs. A huge unlock for the space, and it's fully open source. With zkVMs almost all proving can be parallelised, e.g. farmed across a cluster of GPUs. The exception is the first pass of trace generation, which is inherently sequential. For low-latency proving most zkVM teams hit a wall with this minimal first-pass execution. So far teams reached up to 150MHz RISC-V execution. @ziskvm smashes that with a 1.5GHz emulator, approaching native CPU speeds. The trick? Custom x86 assembly via ahead-of-time (AOT) compilation. 3 or 4 x86 instructions per RISC-V/ZisK instruction, plus 2 x86 instructions to segment the trace into correctly-sized chunks. Let's look at block number 21,926,929 which consumes 36M gas over 230 transactions. The emulator runs at 1.5GHz on a $3K gaming workstation, generating 1.5GB of traces in 0.5 seconds. The CPU is an AMD Ryzen 9 9950X3D 4.3/5.7GHz. Further performance is expected with faster RAM and a CPU tuned for single-threading. In the short term ultra-fast RISC-V trace generation will significantly reduce the number of GPUs needed to achieve real-time proving. Longer-term we can envision ASIC trace generation being just as fast as native execution, unlocking the endgame of real-time proving.

🧵1/ Introducing ZisK — a new era for zero-knowledge proof systems. A high-performance, fully open-source zkVM stack, born at @0xPolygon, now spun out as its own project. Designed for low latency proving from the first day. Meet ZisK 👇 zisk.technology