Many gamers love competitive online multiplayer games, While others prefer story driven single player experiences jump and runs, fighting, racing or rhythm games. But no matter what you prefer, one thing that we all have in common is that we want games to feel responsive and smooth. For a very long time the gaming industry was laser focused on Frames Per Second as the most important metric to quantify how responsive a game is.
Most gamers were not aware of the fact that in their pursuit of maxing out their graphics card they actually made a game less responsive when they ended up being GPU bound. The reason why a maxed-out GPU causes the system latency to increase significant. Is because the CPU then runs ahead of the GPU and piles up work in the render queue.
When the render queue is full then the CPU must wait around for the GPU to finish what the CPU put in the render queue. So that is where the backpressure and the increased system latency is coming from when you are GPU bound. In 2019 AMD released Anti-Lag and from NVIDIA we got NULL, short for nvidia ultra low latency. Both are features of the graphics driver which try to mitigate the delay increase caused by the backpressure in a GPU bound scenario. However, since the driver tries to pace the game from the outside it cannot affect the pipeline in a way that would lead to a significant delay reduction when you are GPU bound. In a video I did back in 2019 I could show that both Anti-Lag and NULL reduce system latency by a few percent, while a good in-game frame rate limiter reduces system latency by 40%.
The reason why a good in-game frame rate limiter can achieve such a massive improvement. is because the game engine paces itself, basically creating a CPU bound scenario. In that state the render queue is always almost empty as the CPU cannot run ahead of the GPU then. The GPU is just sitting there waiting to render what is prepared by the CPU. So in games which use the Unreal, Source or Frostbite engine as well as in Overwatch I have seen the in-game frame rate limiter successfully prevent the massive system latency increase that you would otherwise get when your system is GPU bound.
That video caused quite a stir back then as all it took to completely destroy both Anti-Lag and NULL was to avoid getting GPU bound. And that was not hard to do as you just had spent some time to find a frame rate value your PC can maintain most of the time and then set your in-game frame rate limiter to that value. One year later NVIDIA presented the Reflex Software Development Kit which allows developers to quite easily implement support for the Reflex Low Latency feature which does not disable the render queue, but instead aims to keep it empty.
So very simply put you can think of the Reflex Low Latency feature like a very good in-game frame rate limiter. But with the major advantage that the Reflex Low Latency feature does not lock you at a specific frame rate, nor does it require any configuration as you just have to enable a single setting. So what do you need to take use of the Reflex Low Latency feature.
First of all, you do not need a G-Sync monitor, nor do you need a monitor which comes with a Reflex Latency Analyzer. The reflex low latency feature works with any monitor from any brand. What you do need is a game where the developer added support for the Reflex Low Latency Feature, and an NVidia graphics card where all GPUs starting with Maxwell, that’s the GeForce 900 Series, are supported. In the description you can find a link to a site which lists all games that support the Reflex Low Latency feature. So, in a game where the developer added support for the Reflex Low Latency feature you will find 3 settings. Off On And Boost.
Off and on are self-explaining. The Boost option overrides the power saving features of the graphics card which causes the GPU clocks to stay high even when you are CPU bound. Compared to the on setting, the boost option can lead to an additional slight latency reduction, but it comes at the cost of the graphics card consuming significantly more power.
Which means more heat and most likely more noise. So in addition to enabling the Reflex Low Latency feature, the boost setting actually just does the same as manually setting the Power Management Mode in the nvidia control panel to “prefer maximum performance”. So while NVIDIA decided to prevent the system from getting GPU bound by allowing developers to easily implement a dynamic frame rate limiter which requires no configuration from the users. AMD went in a very different direction their Radeon Boost feature. For select games Radeon boost offers dynamic resolution in response to your mouse or controller inputs Leading to higher frame rates and as a result quicker responses to your input Both Radeon Anti-Lag and Radeon Boost can be enabled via the Radeon software that comes with your graphics cards And neither of them require a proprietary API or game integration to provide latency reducing benefits. Radeon Boost can also provide an even more dramatic reduction in input lag thanks to the higher frame rates it provides.
You might know the render scale slider offered in games like Fortnite or Overwatch which allows you to change the resolution at which the engine renders the gameworld while the user interface is rendered at the resolution you selected in the display options. The resolution slider is a very handy feature as you can use it to have the engine render the game at a higher resolution which leads to less aliasing and a sharper image. Or you can reduce the resolution at which the engine renders the game to increase the frame rate while the user interface stays sharp. Very simply put it seems like Radeon Boost dynamically adjust the render scale setting to give you a frame rate boost in fast paced situations. I say it seems like it because in my tests I have not seen the resolution change at which the user interface is rendered, so it appears to only affect the resolution at which the engine renders the game world like the resolution slider in many games does. So while it works nothing like NVIDIA’s Reflex Low Latency feature, Games must also be supported to use it.
In the description you can find a link to a site which lists all games that are supported by the Radeon Boost feature. Besides a game which is supported by Radeon Boost feature you also need an AMD graphics card where GPU’s starting with the RX 400 are supported. Now when you enable the Radeon Boost feature after you launched the game then you must restart the game for the feature to get active. You will also notice that you can not use Anti-Lag and Boost at the same time. Once it is active you will not notice any difference, that is until you move the mouse which then causes the engine to render the game world at a lower resolution. At the default setting the resolution can be lowered by up to 50%.
Depending on how fast you move your mouse and depending on the game you will see the frame rate go up as you are moving your mouse. You will also notice that the image quality is reduced and it looks like there is a blur applied as well. That said please keep in mind that you watch a gameplay capture that has also been butchered by YouTubes encoder – so the quality won’t be great either when Radeon Boost is disabled. So I spent some time playing with Radeon Boost enabled, and I have to say that it is an interesting idea to get a significant FPS boost by reducing the render resolution while I am moving my mouse.
Question is if the image quality loss bothers you as the difference is noticeable at 1080p and sometimes it takes Radeon Boost a bit to switch back to the full resolution if you abruptly stop moving your mouse. I can also show you the difference in image quality here in the menu. Now lets see how these features affect how responsive a game feels. To measure the delay from pressing the left mouse button to the action triggered by that input showing up on the monitor I use NVIDIAs system latency measurement tool called LDAT.
The contacts of the switch from the left mouse button are directly connected to LDAT, this is how it senses that a button press has occurred. Once I press the left mouse button it starts to count the time until the muzzle flash shows up on the monitor which is detected by the sensor inside LDAT facing the monitor. So LDAT does not know what kind of graphics card is used, or which latency mitigation feature is enabled. It simply measures the time between the button press and the muzzle flash showing up on the monitor. So lets start with Overwatch The monitor is set to a fixed refresh rate of 360Hz which means that the refresh rate is always the same and does not affect the results. With a GTX 1050 maxed out I got between 66 to 68 frames per second And an average system latency of 86.67ms in this GPU bound scenario
With just the in-game reduce buffering option enabled the average delay was reduced by about 12ms With just the Low Latency Mode set to ON inside the nvidia control panel the average delay increased by about 5ms With the Low Latency mode inside the nvidia control panel set to ULTRA also know as NULL the average delay was reduced by about 8ms Then with just the in-game frame rate limiter set to 60, which means that I lost 6 to 8 frames per second The average delay was reduced by about 34ms. When you combine the in-game frame rate limiter with the in-game reduce buffering feature then this has a minor impact on the delay. While when you combine the in-game frame rate limiter with the Low Latency Mode set to ON or NULL in the nvidia control panel then these do not further reduce the delay. So this again to shows that pacing a GPU bound game from the outside does not do much, while using a good in-engine frame rate limiter leads to significantly less delay.
Now lets move on to the Results of the Reflex Low Latency feature When the system is GPU bound I measured an average delay of 86.67ms With the frame rate limiter set to 60FPs and reduce buffering enabled in the Overwatch the average delay is down at 51.59ms Then with just the Reflex Low Latency feature enabled the delays are as low as with the in-engine frame rate limiter, while we gained about 3 to 4 frames per second.
When I also enable the reduce buffering setting then this has a small impact on the delays, while the Reflex Low Latency Boost option does not affect the delay, which is not a surprise though considering that we have a GTX 1050 here. When I also enable the reduce buffering setting then this does not have much of an impact. So pacing a GPU bound game from the outside with NULL does not help much while the Reflex Low Latency feature performs just as well as a good in-game frame rate limiter while it requires no configuration and even gives you a few FPS more than you would get with an in-game frame rate limiter set at a fixed value.
So when the 1050 is maxed out I get between 66 to 68 fps and I measured an average delay of 86.67ms With the AMD RX580 I got between 62 to 68FPS and measured an average delay of 84.21ms Now I do not want you to compare these results, what I want you to pay attention to is the difference between The 86.67ms And the 78.81ms when NULL is enabled As well as the difference between the 84.21ms And the 78ms when Anti-Lag is enabled.
So with Anti-Lag and NULL we only get a small delay reduction as these driver features attempt to pace the game from the outside. I think we all understand now that this will not give us the delay reduction that we want because as I explained back in 2019 we can simply use an in-game frame rate limiter instead of NULL or Anti-Lag to get much less delay. So lets focus on nvidia Reflex and Radeon Boost.
For Radeon BOOST I had to do 2 separate tests, For the first one I measured the delay when I just stood still and did not move the mouse. In this tests Readon Boost does not do anything as it only reduces the render resolution when you move your mouse. Despite that, the measured average delay was slightly higher which is bad news for everyone who has their sight on a player waiting for him to peek out of cover. It is also bad news for those who are slowly tracking a target as the frame rate does not change much when you slowly move your mouse. Then for the next test I had to measure the delay while moving the mouse.
So I tried to always move the mouse at the same speed and over the same distance. And when I reached the right most position I pressed the left mouse button. During that test the frame rate went up from 62 to about 121 Frames per second And the average delay was reduced by 16ms That frame rate increase from 62 to 121FPS while aiming is significant. Sadly it does not result in a big system latency decrease as we are still GPU bound. Furthermore, when you look at how far apart the standard deviation markers are you can see that there is a lot more variation – which means that the delays are not consistent. That is not really a surprise though because as a result of the changing frame rate we also see that the frame times are all over the place.
And lastly at 1080p on a 24 inch monitor you do notice the lower render resolution while you are moving your mouse. Then with a GTX 1650 Super and an RX 5500 XT I got about 144 frames per second in my tests. The Reflex Low Latency feature again reduces the delays significantly.
While standing still with Radeon Boost enabled, I measured slightly less delay on average but a lot more variation. While moving the mouse the frame rate went up to 210FPS, but sadly the average delay was only reduced slightly as we are still GPU bound. Back to the 1050 and RX580 in Fortnite Reflex Low Latency cuts the system latency in half Radeon Boost while standing still reduces the average delay by 8ms and while moving the mouse by 17ms.
However, the delays are all over the place while moving the mouse as indicated by the standard deviation markers. With the 1650 Super and RX 5500XT the frame rates are up to about 144FPS. Reflex Low Latency continues to perform as well as a good in-game frame rate limiter. With Radeon boost enabled the average delay did not change while standing still.
While when I moved the mouse the delay was reduced by about 6ms. Also my PC froze during every test run after I fired about 90 shots. So it does not seem to be perfectly stable yet. And I did test not only with the latest recommended Driver but also the latest optional driver. Next up is apex legends where Reflex Low Latency again leads to a massive delay reduction.
While standing still Radeon Boost did not reduce the delay What it did do is make the game unplayable because this happens when I move the mouse with Radeon Boost enabled. Again I tested this with the latest recommended and the latest optional driver as well as with a different PC and during a Duo’s match. Currently it seems that Readon Boost does not work with Apex Legends. Moving to the 1650 Super and the RX5500XT we see the expect delay reduction with the Reflex Low Latency feature enabled. Radeon Boost did not affect the delays while standing still But the game also became unplayable with the RX 5500 XT when Radeon Boost was enabled, even though it is officially supported by that feature.
So a system that is GPU bound is less responsive. In Overwatch as well as games that use the Unreal, Source or Frostbite engine the in-game frame rate limiter will prevent your system from getting GPU bound and so avoid the system latency increase that you will otherwise be subjected to. But it requires some trial and error to find a frame rate value your PC can maintain most of the time and you are locked at a fixed frame rate then. AMD’s Anti-Lag and NVIDIA’s NULL try to pace the game from the outside and so they only have a very, very small impact on system latency. NVIDIA’s Reflex Low Latency feature requires the game developer to add support.
But it not only performs as well as a good in-engine frame rate limiter it also comes with the massive advantage that it does not require any configuration from the user while it squeezes as many frames per second as possible out of your system. AMD’s Radeon Boost on the other hand tries to make games more responsive by increasing their frame rate during fast paced action. Which is achieved by reducing the resolution the engine renders the game world at. Radeon Boost also only works with specific games that are supported.
And the game still appears to be GPU bound with Boost active which is why it does not decrease the system latency by much. Also when you don’t move your mouse or when you move it very slowly then Boost does not increase the frame rate. But when you move your mouse fast enough Radeon Boost can double your frame rate Sadly that comes at the cost of a worse image quality and when you abruptly stop moving your mouse it can take Radeon Boost a bit to render the game at the full resolution again. Also your frame times and the system latency are all over the place as a result of boost increasing and decreasing your frame rate which could also have a negative impact on your aim. Stability also seems to be a problem at least at the time I made this video as Apex Legends became unplayable in my test and while testing Fortnite my PC froze a couple of times when Radon Boost was active.
So based on my tests NVIDIA’s Reflex Low Latency feature is the better solution to combat the system latency increase that you will otherwise suffer from when your system is gpu bound. It does not affect image quality nor frame times in the way Radeon Boost does. It also appears to be much more stable because since the release of this feature I have never seen my PC freeze or experience stutter that made a game unplayable. So I am very happy that AMD got more competitive in terms of processing power as competition is great for gamers and it drives innovation. I hope that AMD will also step up their game when it comes to reducing system latency, which NVIDIA surely has committed to with their Reflex Low Latency platform. I also hope that more game developers will add support for the Reflex Low Latency feature, and not just in competitive online games. I mean, Heartstone might not benefit much
from Reflex Low Latency but I would love to see it supported by games like Outriders, Warframe, The Division, Ghost Recon Wildlands and many more as system latency is a concern in almost every game, not just competitive online shooters. And that is it for today. As always, I want to thank my patrons for their awesome support Because without them I would have had to abandon this channel a long time ago.
So if you find my content useful and would like to support me then you can find a link to my patreon in the description down below. If you enjoyed this video then please give it a like subscribe for more ring the bell to get notified when I upload my next video and I hope to see you next time! Until then, have a nice day and take care, my name is Chris and this was Battle(non)sense
2021-03-19