Home / Technology / Higher ST gains, some gains, some losses

Higher ST gains, some gains, some losses



The two weeks following our AMD review weekend culminated 7 days a week with our extensive coverage of the new Ryzen 3000 processors, the new Radeon RX 5700 Navi GPUs, and our comprehensive overview of X570 chipset. motherboards.

Among the things that did not go as planned during the hectic period that preceded Sunday, there was the issue of launching BIOS. The problem is still thorny with the launch of new platforms – BIOSes are often intensively developed until a new platform is delivered – we found ourselves in a situation where multiple versions BIOS floated, with different performance. And, although Moore's law may be dead, Murphy's law is still in the news. The BIOS on which we finished testing Ryzen 3000 was therefore not the best BIOS for the platform.

So we want to clarify the chronology of events based on how we initially tested, what we retested, and how the new BIOS behavior could change our initial conclusion of the Ryzen 3000 series.

To go a little behind the scenes and talk about the AMD sampling process, the company initially focused on revising equipment with a selection of 4 different X570 motherboards. Among them was MSI's flagship product, the MSI MEG X570 Godlike, which we already had in the lab with the MSI MEG X570 Ace during our first series of motherboard exams. By bad luck, our God-like did not manage to make it intact, and we never managed to get it started. So, after a new comedy about the logistics issues, we only get the processors a few days before the launch date – and so, without enough time to get a replacement card – we turned to our other X570 card, the MEG Ace from MSI.

The MSI X570 MEG Ace, meanwhile, remains one of MSI's high-end motherboards and would normally be a good choice for a revision card. The trap of using Ace in this situation, however, is that it's not one of the launch tips that AMD was working with. So it does not have an explicit and fine-tuned BIOS like Godlike.

In the end, due to the pressure of time and ignorance of the new stimulating behavior of the Ryzen 3000, we did not detect any problems related to the BIOS of the card (or had been warned) before informing us and testing a new public BIOS by MSI after the review article.

During a custom frequency test, we were able to verify that the new UEFI CPPC2 (Collaborative Performance and Performance Control) interface was not working properly on the card because it was not going to the highest frequencies, but it did not improve. high frequencies in the 1 to 2 ms periods promised by AMD, but rather in a relatively slow time, about 500 ms.

Re-test and update our numbers

Since then, we have been able to test the Ryzen 9 3900X and the Ryzen 7 3900X again and have updated the review article accordingly. In addition, we wanted to publish a summary of the changes made to the various workloads observed on the 3900X for all people who have already read the initial review:

Ryzen 3900X before and after: SPEC2017 and Web tests

The biggest change concerns tests related to the performance of a single thread. These tests mainly have multiple threads with a single heavy thread or a single thread period. We have seen here the effect of the new BIOS in action, allowing processors to approach the maximum recall speeds announced, and we have seen the largest increases ranging from 4% in SPEC to 7-9% in Web tests.

In particular, interactive tests such as WebXPRT experienced greater changes, with the increase in frequency being amplified by the speed of frequency acceleration, which resulted in increases greater than the frequency of change. 5.8% higher increase than we could verify.

Ryzen 3900X before and after: system tests
* Note: The 3900X AppTimer result is probably an outlier.

Ryzen 3900X before and after: rendering tests

Ryzen 3900X before and after: encoding tests

In systems, encoding rendering and testing, changes, and performance improvements heavily depended on the multithreaded behavior of the workload. The Agisoft Photoscan test behaves the same as the Web tests, while the heavy multithreaded tests in the system suite did not see any changes. Some tests have bottlenecks of single-threaded components that lead to intermediate improvements, averaging about 2-3%.

Ryzen 3900X before and after: games (720p)

The changes to the game criteria were a little more varied in terms of results, especially because we saw more significant damage. The main reason here is that some titles, while being multi-threaded, only have a limited number of threads. In these titles and situations where we are not simply bound by the performance of a single large thread, it is likely that the CPU frequency is slightly lower in the new BIOS version than we had tested at the time. origin. Our results were also consistent across several analyzes in this regard. Therefore, it is not an artifact of the normal variation from one run to the other.

Overall conclusion: better results with a single wire, but the same positioning

MSI's updated launch BIOS has generally improved our original numbers in areas where the new Ryzen 3000 has already excelled: desktop and productivity applications. Single-threaded, particularly exclusive workloads have seen a larger increase, making the new Zen 2 all the more impressive. In the meantime, the results of the games seem to indicate that improvements still need to be made to AMD's amplification algorithms. None of this changes our overall performance analysis or recommendations, but when the difference between Intel and AMD is sometimes a game of thumb, it's no surprise that everyone is very interested in the slightest change.

AMD noted that the results may vary from one BIOS provider to another. In this regard, we will continue to monitor BIOS changes in the coming weeks and see if there will be more substantial behavior updates in the new Precision Boost 2 mechanism.

What is certain is that if you plan to buy a new processor from the Ryzen 3000 series, be sure to keep up to date with what's new in BIOS, as it is likely that the versions that come with the new X570 cards do not fully represent the performance new processors are capable of.


Source link