一位高级客户最近询问了在测试强制的“高”动态电源管理状态而不是默认的“自动”模式时,如何测试开源Radeon驱动程序的性能。这里有一些基准。
这是关于强制 /sys/class/DRM/card 0/device/power_DPM_force_performance_level 为“高”而不是“自动”,以确保GPU被绑定到其最高性能状态,而不是基于负载动态改变性能状态。有些人把这归因于更高的性能,虽然它在过去或旧的硬件中可能有,但我并没有发现它对新的图形处理器和最近的内核版本有太大的好处。
但是当一个高级支持者提出请求时,我做了一些快速测试。测试是用Linux 4.17内核和普通的Radeon RX 580图形卡完成的。
只有在非常特殊的情况下,比如Dota 2和Vulkan,强制的高性能状态才会对这个现代内核上的RX 580产生任何影响。
所有的数据都可以在这里找到,但长话短说,与我运行的其他测试类似,至少在最近的AMD GPU硬件上使用了最近的内核,切换 power_DPM_force_performance_level 对于默认行为并不太有用。如果您遇到了不同的结果,请务必让我们知道硬件和您在论坛中的结果。
A premium patron recently asked about testing the open-source Radeon driver performance when testing the forced "high" dynamic power management state rather than the default "auto" mode. Here are some benchmarks.
This is about forcing /sys/class/drm/card0/device/power_dpm_force_performance_level to high rather than auto, to ensure the GPU is bound to its highest performance state rather than dynamically changing performance states based upon load. Some have attributed this to greater performance, and while it may have in the past or with older hardware, I haven't really found it to be too beneficial for newer GPUs paired with recent kernel releases.
But as the request came in by a premium supporter, I ran some quick tests. Tests were done with a Linux 4.17 kernel and the common Radeon RX 580 graphics card.
Only in very select instances like Dota 2 with Vulkan did the forced high performance state make any difference with the RX 580 on this modern kernel.
All the data can be found here but long story short similar to other tests I have ran, with at least recent kernels on recent AMD GPU hardware, toggling power_dpm_force_performance_level hasn't really been found to be too useful over the default behavior. If you've encountered different outcomes, be sure to let us know the hardware and your results in the forums.