No need to wait, that AMD CPU performance update is available right now in Windows 23H2, and even Zen 3 and Zen 4 chips should receive a boost

You'll never hear me complaining about free performance, but you might hear me complaining about performance that was left on the table and shimmied in after the fact. This seems to be the case with the latest AMD Ryzen 9000-series chips, which have been underwhelming so far but are now able to gain some free performance from an optional Windows update.

And it's not just Zen 5 chips that are invited to the party—testing has shown that older Ryzen chips look to be getting a tasty performance bonus, too.

While we previously reported that Zen 5 chips would gain performance from the upcoming Windows 11 24H2 build, you can now install the update in the current 23H2 build (via Wccftech). No having to be a «Windows Insider» or any of that malarkey: You can just install the relevant update like normal.

To be clear, though, it's still an optional update. As AMD tells Wccftech, this means you'll have to go to Windows update > Advanced options > Optional updates. There, you can install the KB5041587 update. This should improve the branch prediction of not just 9000-series processors, but 5000-series and 7000-series ones, too. 

Which complicates AMD's messaging a little, in my opinion, because in a previous blog post AMD suggested that branch prediction would be improved for Zen 5 chips with a Windows update, and this was pointed out in response to backlash against lacklustre Zen 5 performance that didn't match AMD's original performance claims.

The problem is, while much of this backlash revolved around CPU performance claims compared to Intel chips, some of it was about AMD 9000-series performance compared to the 7000-series. So, if this update is a rising tide for all AMD chips, this still won't signal much of a generational improvement.

Negative spin aside, however, in absolute terms this performance boost is welcome news. While branch prediction changes probably won't improve heavily multi-threaded or purely single-threaded performance, we can expect significant