Two weeks passed after our AMD test weekend. The 7/7 highlight was the extensive coverage of the new Ryzen 3000 series CPUs, the new Navi GPUs of the Radeon RX 5700 series and our complete overview of the X570 chipset motherboards.
One of the things that did not go as planned in the hectic time before Sunday was BIOS booting. Always a delicate issue with new platform launches – BIOS versions are often heavily developed until a new platform is delivered – and there was a situation where several BIOS versions with performance differences appeared on some boards. And while Moore's law may be dead, Murphy's Law is alive and well, so the BIOS we ran our first Ryzen 3000 tests was not the best BIOS for the platform, of course.
Therefore, we would like to clarify the timing of events, how we tested them for the first time, what we re-tested, and if and how the new BIOS behavior could change our original Ryzen 3000 series conclusion.
To get a quick look behind the scenes and talk about AMD's sampling process, the company initially focused on providing examiners with a choice of 4 different X570 motherboards. Among them was MSI's flagship, the MSI MEG X570 Godlike, which we already had in the lab along with the MSI MEG X570 Ace for use in our first round of motherboard testing. In an unfortunate moment, our gods could not reach us, and we were never able to get it up. After another comedy of logistical issues where we got the CPUs just days before the start date – and therefore not enough time for a spare board – we turned to our other X570 board, MSI's MEG Ace.
In turn, the MSI X570 MEG Ace is still one of MSI's high-end motherboards and is usually a good choice for a test board. The danger of using Ace in this situation, however, is that it is not one of the startup boards that AMD has worked with. So there was no explicit and well tuned test BIOS like Godlike.
Due to the time constraints and ignorance of the new Ryzen 3000 boosting behavior, we were finally unable to detect any problems in the board's BIOS (or were warned of problems) until we got to A / B and tested a new audience BIOS from MSI after the review.
In a custom frequency test, we were able to verify that AMD's new UEFI Collaborative Power and Performance Control (CPPC2) interface on the board was not working properly because it was not raised to the higher frequencies, but above all did not increase frequencies in the AMD promised to increase 1-2ms periods, but in quite a slow ~ 500ms.
Re-testing and updating our numbers
Since then we have been able to re-test both the Ryzen 9 3900X and the Ryzen 7 3900X and have updated the test article accordingly with the new numbers. In addition, for those who have already read the first test, we wanted to publish a summary of the changes in the various workloads we saw on the 3900X:
The biggest change was in tests tied to the performance of a single thread. These tests mainly consist of either multiple threads with only one heavy thread or just one thread period. Here we have seen the effect of the new BIOS in action, which allows the CPUs to approach their announced maximum boost speeds, and we have seen the biggest increases in the Web tests between 4% in SPEC and 7-9% ,
Interactive tests such as WebXPRT made major changes as the frequency increases were boosted by the faster rate of increase in frequency, resulting in increases above the 5.8% higher boost frequencies we could verify.
* Note: The 3900X AppTimer result is probably an outlier.
In the system, rendering, and encoding tests, performance changes and improvements were heavily dependent on the multithreaded behavior of the workload. Agisoft's Photoscan test is similar to the Web test in terms of enhancements, but no changes were detected during extensive multithreaded testing in the System Suite. Some tests have bottlenecks on single thread components, leading to intermediate improvements, averaging 2 to 3%.
The changes to the gaming benchmarks were slightly different in terms of outcomes, especially as we saw bigger losses. The main reason for this is that some titles have multiple threads, but only a limited number of threads. In those titles and situations where we are not just tied to the performance of a single big thread, it's likely that the CPU clock in the BIOS of the new version is slightly lower compared to what we originally tested. Our results have been consistent across multiple runs, so this is not an artifact of normal run to run variations.
Conclusion: Better single-threaded results with the same positioning
Overall, MSI's updated startup BIOS improved our original numbers most in areas where the new Ryzen 3000 has already excelled: office and productivity applications. Particularly exclusive single-threaded workloads experienced a greater boost, making the new Zen 2 all the more impressive. In the meantime, the revaluation of game results seems to indicate that AMD's boosting algorithms need to be improved. None of this changes our overall performance analysis or recommendations, but if the difference between Intel and AMD is sometimes a customs clearance, it's not surprising that everyone is interested in the slightest change.
AMD noted that the results may vary from manufacturer BIOS to manufacturer BIOS. In this context, we will continue to monitor the BIOS changes over the next few weeks and see if the new Precision Boost 2 mechanism contains more comprehensive behavior updates.
However, if you want to buy a new Ryzen 3000 CPU, you should keep up to date with the latest BIOS, as the versions shipped with the new X570 cards are unlikely to be able to provide the full power of the new CPUs.