Magnetar speeds up proofs for ordinary code without erasing the trust bill

Microsoft Research has released Magnetar, a transparent zkVM for programs compiled to the RV64I instruction set. A proof can attest that a program ran correctly without repeating the full computation and, depending on its use, without revealing sensitive inputs.

RV64I turns a cryptographic circuit into a programming target

The design bases security on standard hash functions such as SHA-256 or SHA-3 and requires no trusted setup. Its Rust implementation targets 120 bits of stated security. On a MacBook Pro with an M4 Max and 12 performance cores, it sustained 6.5 to 7.7 million cycles per second. Proofs were about 600 KiB and verified in 10 to 12 ms.

Across 15 application workloads, the authors report 22 to 57 times the throughput of SP1 Hypercube. Against two Jolt variants, the reported range is 5 to 33 times depending on the system and workload.

A conventional toolchain lowers the entry cost for developers

RV64I matters more than the record alone. Developers can compile Rust or Go toward a general instruction set rather than manually translating an application into a specialized cryptographic circuit. That makes verifiable cloud jobs, computation audits and blockchain applications plausible targets.

The operating model also changes. One party performs the expensive computation once, while recipients inspect a compact proof. If integration stays close to ordinary software development, proofs could become a deployment component rather than a separate research project.

One benchmark win does not settle a production choice

The figures come from the authors on a particular machine. zkVM comparisons move with the workload, compiler, memory, parallelism and security profile. A smaller proof does not automatically mean a cheaper system. Jolt produced proofs below 100 KiB in this comparison, while Magnetar spends about 600 KiB on a different mix of assumptions and performance.

Post-quantum here describes security rooted in standard hashes. It says nothing about the security of the application, compiler or guest code running inside the zkVM.

Deployments will expose the memory, cost and portability bill

The useful next evidence is reproduction on x86 and ARM servers, memory use on long jobs and cost per generated proof. Independent teams also need to run existing Rust and Go programs without extensive changes.

Production integrations will show whether faster cryptography removed the bottleneck or merely moved it into compilers, operations and audits.

Lilith's verdict

Magnetar hands the verifier a 600 KiB receipt instead of the whole computation. Independent teams now need to confirm that the total still adds up outside Microsoft’s lab.