Ryzen Platform Affected by RTC Bias; W8/8.1/10 Not Allowed on Select Benchmarks

In a statement issued by the Head of Moderation Christian Ney, we have confirmed that the AM4 platform is affected by the Windows 8/8.1/10 RTC bias. The bias occurs when adjusting the reference clock at run-time and will affect the Windows timer, causing benchmarks to perceive time slower (or faster) than it really is. This results in benchmark scores affected in such a way that the benchmark scores reported do not reflect real performance.

The RTC bias is referenced in the ROG Crosshair VI Hero Extreme Overclocking guide available on Overclocking.guide: "Timer is skewed when changing REFCLK in Windows 8+. Additionally the default systimer has issues with OS ratio changes unless HPET is enabled. To summarize, always enable HPET on this platform."

We described the behavior of the RTC bias in an article published on August 18, 2013 (see below). We also issued rules updates for the Skylake platform in an article published on November 5, 2015.

“The concept of ‘time’ on a PC configuration is, if not synced via network or internet, an arbitrarily defined constant designed to ensure that the configuration is running in sync with the real world. In other words: hardware and software engineers ensure that ‘one second’ on your PC equals ‘one second’ in real time. One of the reasons why it’s so important to have the PC’s timer line up with the real world time is to ensure that your PC can produce accurate measurements and predictions.” The points we brought up in that editorial are relevant again. To ensure that the arbitrarily defined constant of ‘time’ is the same on everyone’s benchmark system, we rely on the OS and hardware. This worked quite well, until Windows8 came around.

The problem builds on the problems we faced with Heaven. When downclocking the system under Windows8, the Windows RTC is affected as well. The biggest difference between Windows7 and Windows8 is that now all benchmarks (no exception) are affected.

Let us make this more practical. On our Haswell test system we downclocked the BCLK frequency by about 6% from 130 MHz to 122MHz. Using a CPU ratio of respectively 32x and 34x, the resulting CPU frequency remains 4160MHz. Then we ran comparison benchmarks.

With immediate effect, we no longer accept AM4-based overclocking result submissions with Windows 8/8.1/10-based Operating Systems for benchmarks listed in the General Rules, Section 1.6. You are allowed to use those operating systems with approved benchmarks such as the entire 3DMark suite, GPUPI, HWBOT X265 Benchmark, Y-Cruncher, Realbench and CPU-Z. We will keep you updated on any changes to this list.

For reliable performance measurements with at run-time overclocking, we recommend enabling the High Performance Event Timer (HPET). Alternatively you can opt to use a Windows 7 based operating system. Note that at run-time overclocking using the CPU multiplier will not result in RTC bias (under investigation).

We are also investigating the impact of the "Ryzen Sleep bug" possibly affecting benchmark integrity.

The HWBOT Staff.


Please log in or register to comment.