Test Setup

Performance Results

We ran our full suite of games and applications on the B350 platform with both the air cooler and water cooler installed.

While games showed no performance differences, application performance results had some surprising data points.



The chart above plots relative performance against the data from our Ryzen 9 3900X review in all applications. If performance is the same, the point gets put at "0%". If performance is better on B350, a positive value is used, and if performance is worse, that value goes into the negative.

While all results are within random variation, two results stick out immediately. These are for Blender and Corona—two heavily threaded rendering applications. It looks like when watercooling is used, they're losing a lot of performance. This is also true to a lesser degree for KeyShot and Photogrammetry, which are multi-threaded applications as well.

The explanation is that the air cooler has the benefit of creating some airflow around the CPU socket, which reduces VRM temperatures due to it picking up some of the heat on the VRMs. Enthusiasts often use watercooling to maximize their cooling performance for higher CPU overclocking, which is the root of the problem. With zero airflow around the socket area, VRM temperatures just keep climbing, which activates some safeguard that slows down the CPU to avoid physically damaging the motherboard, CPU, or other components.

We'll look at VRM temperatures in more detail on the next page.

Overclocking

The overclocking experience for the Ryzen 9 3900X using B350 matched exactly what we saw during our testing with the ASRock X570 Taichi. The biggest hurdle with Ryzen 9 3900X is its massive heat output that managed to overwhelm our 240 mm AIO watercooling, even with slightly reduced voltages. In order to ensure the processor doesn't overheat with heavy multi-threaded loads, we used the same CPU voltage as in our original 3900X review: 1.225 V. While that is quite low, it's the best we can do with the 240 mm AIO.

Next, we dialed up the clocks and reached stability at 4 GHz all-core, 25 MHz higher was unstable—exactly the same as on X570.

Setting the memory timings was mostly trivial, too. Initially, the board didn't want to run 3200 MHz CL14 and kept resetting down to 2400 MHz fail-safe. Manually setting 2666 MHz, booting successfully with that, and switching up to 3200 MHz fixed it, and the system never complained again about memory speeds.