The bench
The same network, weights and input, run twice: as the scalar twin on today's instruction set and as the vector mind on the SIMD-0714 prototype's ops. Two transactions you can check. CU is compute units, Solana's measure of work.
Not Solana mainnet. A fork we run.
DRAW OR TAKE AN INPUT FIRST
READING THE CHAIN
Run both. The same 256 bytes go to each, one transaction each.
Run both. The same 256 bytes go to each, one transaction each.
NOT RUN YET
NOT RUN YET
NOT RUN YET
NOT RUN YET
THE STOCK SOLANA CLI READS BOTH. ONLY THE URL CHANGES
Each run prints its line here: solana confirm -v, the signature, and the public RPC of this chain.
MEASURED ON OUR MACHINE
AMD Ryzen 9 9950X 16-Core Processor, AVX-512. Agave v4.5.0-alpha.4 with temporalxyz/sbpf rodata-pair at 585f893. Last run 2026-11-18 20:47:54 UTC
SIMD-0714 DOCUMENT: ANOTHER MACHINE
AMD EPYC 9534 (Zen 4). Another machine. Published in the draft, not measured by us. We ran the draft's prototype and measured it. The draft and its reviews