Frame rate has been the only metric for comparing performance in games. Even at smooth frame rates, however, gaming can be uncomfortable if there is a significant amount of input lag – the game responds to controls with noticeable lag. There are countless tools for measuring frame rates, but unfortunately not for testing responsiveness. Here we introduce the specialized Frame Latency Meter recently made available by AMD.
The whole app looks simple on the outside. It only knows a few parameters when run from the command line, and there are also only a few things to set up in the context menu, and it doesn’t look complicated in the documentation either. So I wasn’t sure why the internet and YouTube aren’t already full of tutorials on how you can measure input lag in games from the comfort of your home with AMD’s free tool, and I wanted to make a quick tutorial that would be easy to follow.
And when I started digging more into Frame Latency Meter, I realized why this is probably the case. It’s not that easy to set everything up correctly, and you run into a number of problems in the measurement process. Rather than being a simple, bug-proof application for everyone, FLM is a classic open source application whose developers assume that you are familiar with latency measurement and won’t have a problem with the command line or other “hidden” settings that you can find in the .ini configuration file. And, unfortunately, it can’t be said to work seamlessly either. So far, it seems to me that AMD has laid some groundwork on which the community will have to build their own tools with more user-friendly controls if normal people are to be able to use it.
So, consider today’s article more as an introduction to the application and the basic principles of how it works in case you are one of the more proficient ones and would like to get into experimenting as well. I’ll try to look at it in more depth later, and when I get some useful and publishable measurement results, we’ll come back to Frame Latency Meter.
The AMD Frame Latency Meter application can be downloaded from the AMD gpuopen.com website. Source code, bug reports and discussion is on GitHub GPUOpen.
There are only four files in the zip file, we can extract them to, for example, the directory c:\flm
readme.pdf
flm.exe
flm.ini
users_guide.pdf
readme.pdf is a brief description of the application and a quick setup.
The users_guide.pdf contains the most detailed description of settings and functions, including keywords for keyboard shortcuts, which are then used to configure keyboard shortcuts in the ini file.
There are some important notes at the end of both the readme.pdf and the users_guide.pdf files. From them you can guess that for tuning response times for esports games or online shooters, frame latency meter is not the most suitable, and results measured on Radeons with Advanced Media Framework and GeForce or Intel with DXGI are probably not exactly comparable either. AMD comments that in user guide: ” AMF captures frames using driver level components, whereas desktop duplication is
triggered at operating system levels. The difference in captured frames is not significant to trigger measurements consistently.” without any further details. Therefore, the frame latency meter might be more useful for comparing different system and game settings on the same platform.
AMD has noted the following limitations in the documents:
Known issues and limitations
Desktop Capture codec (DXGI) performance
- Desktop capture is too slow when monitors refresh rate is set too high > 144 Hz. Try setting the monitor refresh rate to 60 Hz while you get accustomed to how FLM works, then try higher rates.
- Certain games may limit the utilization of desktop duplication screen capture, making the option to use the DXGI codec inoperable.
- DXGI capture codec must be running on the primary display, else frame capture will be too slow.
- The current capture codecs are processing SAD measurements using CPU and may be too slow on some PC
No result shown
- The default settings may not trigger latency measurements. Try adjustments to settings for ThresholdCoefficient and AVGFilter in the flm file.
- FLM will not work using remote desktop connections
Latency results shown are too high
- When games are running at high frame rates FLM measurements show higher values then those measured by
hardware devices.
With the principle on which the frame latency meter works, it can happen that it will not be able to capture all the frames and the measurement in such a case will not correspond to reality. There is a keyboard shortcut for capturing control frames to check if it can handle this, see the readme file for details. The short way to describe this is to run some application that displays a counter for each frame, capture the area with the counter, and after saving the sequence of control frames, go through them to verify that all of them are captured. You can find more detailed descriptiption of testing procedure in user_guide.pdf
