Re-Testing AMD Ryzen 9 3900X Performance
Naturally, many of you want to see the new data and how it effected the scores achieved by our Ryzen 9 3900X. The reality is that almost nothing changed in most tests where the application was heavily multi-threaded. There are all kinds of things changed in the AGESA code, and as we said, performance didn’t necessarily improve even if the clock speeds did. We went ahead and reran all the tests anyway, to see what if any difference the update made.
There are some comments floating around out there. Some people have said you can basically add 10% to the single-threaded scores. That’s not what we experienced, although there are cases where the numbers did change quite a bit. There were tests where we did see significant differences, but as we said these weren’t always positive.
Testing methodology is unchanged. We are only comparing BIOS 1.1 to BIOS 1.3 for the MSI MEG X570 GODLIKE motherboard using the newer AGESA code 1.0.0.3 patch AB. At the time of this writing, this version is not available to the public yet.
Sandra Memory Bandwidth
In this test, we can see that BIOS version 1.3 had very little if any impact. This variance could simply be chalked up to standard testing variances.
Sandra CPU Dhrystone
The difference is within an acceptable variance and is essentially meaningless. The point being that in this test, our BIOS update did not change the result.
WinRAR – Multithread
In this test we see our first significant difference. This was a reduction in performance that was absolutely repeatable. The difference is tiny, but it was consistent. We lost a little performance here.
WinRAR – Single Thread
This is another example of a test where nothing changed.
wPrime v2.10
The wPrime test showed us our first significant improvement from the AGESA code update and one that was again repeatable. The result isn’t an Earth shattering difference, but again there was a change.
POV-Ray
Once again, we see nearly no change in performance. What we do see is statistically insignificant.
V-RAY
Once again it looks like we lost performance, but again the result is almost too small to mention. If you didn’t graph it out like this, you wouldn’t really see anything here.
Cinebench R20 – Multithread
In our Cinebench multi-thread test, we saw another slight reduction in CPU performance. Again, this was a consistent result which we could repeat easily.
Cinebench R20 – Single-Thread
Here is where we see one of the only positive changes so far. It was a definite although still largely insignificant improvement which I can really only attribute to the slightly improved clocks.
After Effects CC- Puget Systems
This was a highly unusual result in which we saw a substantial improvement in the test. Strangely, this was also a multi-threaded test.
Blender
Unfortunately, we saw a slight reduction in performance once again.
Gooseberry
In this test we saw another improvement. Again, its almost so small as to be something we could just as easily attribute to the margin of error.