The Pulse
Microsoft’s Magnetar Proves Code Up to 39 Times Faster
Microsoft Research unveiled Magnetar, a zero-knowledge proof system that checks programs compiled for RISC-V and relies on standard hash functions. In tests across 15 workloads, it outpaced three competing systems, with reported average spe
AI.info Team · Written with AI ·
Microsoft Research says its new proof system, Magnetar, can verify program executions without rerunning them and can prove workloads at speeds up to 39 times higher than comparable systems in the team’s tests. The system is designed for ordinary programs compiled to the RISC-V processor architecture, and bases its security on standard hash functions such as SHA-256 or SHA-3. Microsoft Research published the announcement on October 15.
Magnetar targets everyday code
Magnetar is a zero-knowledge virtual machine, or zkVM: it produces a cryptographic proof that a program ran on particular inputs and returned a particular result. A verifier can check that proof in milliseconds rather than repeat the computation, while the zero-knowledge property can keep private inputs concealed. Microsoft Research says developers can write programs in Rust, Go or C, compile them as usual for RISC-V, and prove their execution without translating the program into a custom cryptographic circuit.
“As a result, proving stops being a specialist activity,” writes Srinath Setty, senior principal researcher and manager at Microsoft Research. Setty points to business policies, data pipelines and interpreters for the Ethereum Virtual Machine as examples of programs that could be proved as written. The announcement describes possible uses, not a general commercial release or deployment.
15 workloads, three comparisons
Microsoft’s technical paper evaluates Magnetar on 15 application workloads, including cryptographic hashing and token transfers in a small Ethereum Virtual Machine interpreter. On an M4 Max MacBook Pro using 12 performance cores, the paper reports geometric-mean proving-throughput gains of about 39 times over SP1 Hypercube, 18 times over Jolt-Dory and 9.6 times over Jolt-Akita. Those systems use different cryptographic approaches, so the comparisons are not simply speed rankings divorced from security assumptions.
The paper reports Magnetar’s throughput at 7.2 million to 8.7 million cycles per second, with proofs of about 600 KiB that take 11 to 13 milliseconds to verify. The Microsoft Research blog gives a broader summary of 10-to-39-times average speedups against the fastest existing zkVMs in the same class on the same machine. An optimized version reaches 9.5 million to 12.8 million cycles per second, according to the blog.
Standard hashes shape the security trade-off
Magnetar’s security rests on standard hash functions rather than elliptic curves, lattice assumptions, special-purpose hash functions or a trusted setup, Microsoft says. That choice is intended to avoid relying on cryptographic assumptions that could be threatened by sufficiently powerful quantum computers, while also avoiding newer constructions with less extensive scrutiny. The researchers frame the design as a post-quantum approach; the benchmark results do not establish how the system performs in every deployment or application.
The trade-off is visible in the proof sizes: the paper says Magnetar’s proofs are about 4.6 times smaller than SP1 Hypercube’s, while Jolt’s proofs are under 100 KiB but rely on elliptic-curve or lattice assumptions. Magnetar also makes proof verification an efficient workload within its own design, allowing computations to be proved in pieces and combined. In the reported tests, those proofs measured about 600 KiB and verified in 11 to 13 milliseconds.