On November 9, 2026, Phoronix reported that a new 16-block AVX-512 function for Linux kernel ChaCha was queued in Eric Biggers’ libcrypto-next branch. Phoronix reported up to 99% improvement in ChaCha performance on AMD Ryzen AI Max Strix Halo (Zen 5) with the 512-bit path, and up to 39% on Zen 4 with a 256-bit path. The figures describe ChaCha benchmarks, not WireGuard tunnel throughput.
What changes in the ChaCha path
ChaCha is a symmetric stream cipher. The patch explanation describes a wider AVX-512 path that can process more blocks of cipher data in one invocation. The existing AVX-512VL path handles at most eight blocks in 256-bit YMM registers; the proposed path uses 512-bit ZMM registers to handle 16.
| ChaCha path | Registers and width | Blocks per invocation |
| Existing AVX-512VL path | 256-bit YMM | Up to 8 |
| Proposed AVX-512 path | 512-bit ZMM | 16 |
The new path is selected for inputs larger than eight blocks, according to the patch explanation. Martin Willi, identified by Phoronix as a StrongSwan developer, described the wider route as suitable for newer AVX-512 processors, where the down-clocking concern that had motivated the narrower path on Skylake-X was absent.
What the AMD benchmark results measure
The reported maximum depends on both the processor and the data path used. Phoronix gave these separate ChaCha results:
| Test platform | Data path | Maximum reported ChaCha improvement |
| AMD Ryzen AI Max Strix Halo (Zen 5) | 512-bit | Up to 99% |
| AMD Zen 4 | 256-bit | Up to 39% |
The Zen 5 figure belongs to the named Ryzen AI Max Strix Halo system and its full 512-bit path. The Zen 4 result uses a 256-bit path. They describe different conditions, rather than a direct Zen 4-versus-Zen 5 comparison.
Keep the 99% figure attached to its subject: ChaCha performance on the reported Zen 5 system.
What the results mean for WireGuard
Those percentages do not give a WireGuard speedup. They measure ChaCha performance; a VPN tunnel’s end-to-end throughput is a separate outcome, so the benchmark figures cannot quantify how much faster a WireGuard connection would run.
At the time of Phoronix’s November 9 report, the patch was queued in libcrypto-next, and submission for the Linux 7.5 merge window was expected by the end of 2026.
