What Happened? BIOS Issues?
The question at this point becomes, why didn’t we see the full 4.6GHz clock speeds during our review of the Ryzen 9 3900X? The answer is complicated. The simple answer is that the platform is new and needs time to mature. While it was stable and worked very well out of the box, there are still small things that need to be addressed. AMD’s AGESA code is largely to blame for this as some user’s reported lower boost clocks on the initial firmware.
With CPU launches like this one, reviewers had little time with these CPU’s and a whole lot of testing and writing. The launch also occurred after a holiday weekend and ended up going live on a Sunday, a day you don’t typically see prompt BIOS updates. We were acutely aware of the fact that boost clocks are not guaranteed, so we rolled with it. Essentially we weren’t sure if this was something specific to our setup, or elsewhere till after the launch. The clock speed did reach 4.5GHz clocks with Precision Boost Overdrive and the 200MHz offset. We knew the CPU’s were capable of it. You chalk these things up to an immature BIOS and you move on until it can be addressed.
As to the technical reason, AMD’s AGESA Combo PI code version 1.0.0.3 was what was built into the UEFI BIOS we used for our MSI MEG X570 GODLIKE motherboard. We used version 1.1 for our testing. Evidently, this is a board that had more trouble with boost clocks than some of the other options reviewers were given.
According to MSI, version 1.2 of the UEFI has AGESA 1.0.0.3 patch A, which may yield slightly different results. MSI also gave us an internal copy of v1.3. This UEFI uses AGESA 1.0.0.3 patch AB. This latest version seems to raise the boost clocks of the processor and we can confirm that we were able to hit pretty close to the 4.6GHz mark in single threaded applications. We also have to say that this did absolutely nothing for the “all-core” boost speeds and nothing for overclocking. Our test CPU was limited to a manual overclock of 4.3GHz, and that didn’t change.
MSI’s Response
MSI’s Complete Email response with explanation is posted below for your information. This is directly from MSI.
The boost frequency or boost behavior is determined by AMD Combo PI. Frankly speaking, our BIOS cannot touch this. For each Combo PI version, AMD fixes some bugs (may bring new bugs), and optimizes some things like performance, power consumption and this is why the boost behavior especially single core boost and performance may be slightly different for each Combo PI version.
Like Gavin from Anandtech mentions, Combo PI 1.0.0.3 patch A performs better in some workloads, unchanged in some and negative in some. So, not every benchmark gets benefits from new Combo PI.
For GODLIKE we have the following BIOS with different Combo PI versions. PI 1.0.0.3 patch AB is the latest. Per my quick test with 3800X, for Cinebench R15 single-core performance of PI 1.0.0.3 is the same as PI 1.0.0.3 patch A. But latest PI 1.0.0.3 patch AB’s Cinebench R20 single-core performance is 0.7% lower than PI 1.0.0.3 patch A. 1.0.0.3 patch AB’s load Vcore looks a little bit lower too. My 3800X can hit the max turbo frequency 4.5GHz while running Cinebench Single Core benchmark, both PI 1.0.0.3 and PI 1.0.0.3 patch A. I use HWiNFO64 to check the CPU clock, it’s easier to see the CPU frequency. I’m using Corsair H150i water cooling.
BIOS 110 (v1.1 on MSI website) – Combo PI 1.0.0.3
BIOS 120 (v1.2 on MSI website) – Combo PI 1.0.0.3 patch A
BIOS 1.3b1 (131) – Combo PI 1.0.0.3 patch AB, not available on MSI website yet