AMD AFMF 2 TESTED – AMD Fluid Motion Frames 2 LATENCY + FPS BENCHMARK
AMD first launched their in-driver frame generation tool, AMD Fluid Motion Frames, back in October last year, although I tested it in February this year after it had some updates and fixes in, and if you want to know more about how this generally works, or why you might not want to use this in anything even remotely latency sensitive – think any first person shooter, rhythm games, and probably things like racing games too – check that video out in the cards above. This video though is all about testing the second generation of AFMF, aptly named AFMF 2. I’ll be running the same tests with my open source latency testing tool – which of course I sell over at OSRTT.com, linked in the description – and I’ll be using AMD’s in-driver performance monitor to keep an eye on the FPS data, as tools like PresentMon and even AMD’s own OCAT won’t see any difference as the frame generation is all done in-driver. Right, let’s dive in!
The main thing that’s new here is a revision to how AFMF works under the hood, as AMD now claims that AFMF 2 runs 28 percent lower latency than AFMF 1, which is a pretty massive improvement. We will, of course, be testing that. This also means AFMF now works on not just DirectX titles, but Vulkan and OpenGL titles too, which is pretty huge. The other major change is the addition of two new controls, one for an “AI-Optimized Search Mode”, and the other for performance mode. By default both are set to auto, with the search mode being what happens when AFMF disables itself during fast motion, and how smoothness is handled there. The options are auto, standard and high, where standard is really only for 1080p, versus high for 1440p and up – which is what I’ll be testing at here. The performance mode is really only for those of you with AMD’s APUs – basically this just turns AFMF down a bit so you still get some frame generation benefits, but will less performance overhead, hence why the options are either “quality” or “performance”, and “auto” which defaults to quality save for on those lower power graphics cores. That’s pretty much it, so let’s see how much of a difference it makes – this time testing with an RX 6900 XT as both AFMF 1 and AFMF 2 are supported on 7000 and 6000 series cards.
Starting with CS2, a game I really really hope you don’t enable AFMF with, and looking at the FPS data first, it seems like AFMF 1 didn’t fancy running here as the frame gen lag was zero – as reported by AMD’s own driver – and the performance isn’t double, in fact much like I saw in the original AFMF video it’s actually worse than without. Now this might be an issue with CS2 potentially running in Vulkan mode – especially with the introduction of Anti-Lag 2, that video will be in the cards too actually – and Anti-Lag one doesn’t work for that, although it did say it was active so I’m not so sure. What I do know is that AFMF 2 did work, pumping the FPS way, way up, from around 500 FPS to 730 FPS, and even more impressively the frame generation lag average was only 2.6 milliseconds. Compare that to the 11.2 milliseconds from the last round of testing and that’s mighty impressive – plus it actually showed a performance improvement here which is great news. Now if we switch to the latency data you’ll see why you really, really shouldn’t enable AFMF in any sort of competitive game like this. The average latency jumped from 7.6 milliseconds to 15 milliseconds – literally double. By enabling AFMF here you are making yourself wait twice as long as your competitors to see new frames, and that is a really bad idea.
What about in a less competitive game – and a more demanding one at that – like Cyberpunk 2077? Well, looking at the performance data I’m happy to say that it’s great! While I am testing on an RX 6900 XT so my base performance is already fantastic at medium settings, at 155 FPS average, enabling either AFMF version pegs it to 260 FPS average, making for an exceptionally smooth experience – actually making use of the 240 hertz OLED I’m testing on here. The real star here is the frame gen lag, which dropped from 12.5 milliseconds with AFMF 1 to just 5.4 milliseconds with AFMF 2. That is a major difference, and that’s actually reflected in the latency data too, where with no AFMF enabled you’re looking at around 20 milliseconds of end to end latency, and with AFMF 2 on you only add around 7 milliseconds of end to end latency. While that is a 36.5 percentage increase over no AFMF, if you don’t happen to have an RX 6900 XT and would rather get decent chunk more performance at the cost if a handful of milliseconds of latency, this finally feels like a tradeoff that might be worth making. AFMF 1, even on this 6900 XT, still takes an unbearably long time to render inputs – 33.6 milliseconds, compared to 19.7 with no AFMF. That’s not good enough, but AFMF 2 is.
Finally we have Starfield, an arguably more intensive game than Cyberpunk, as even the 6900 XT on full low settings still nets less FPS than you get in Cyberpunk on medium – so a perfect candidate for AFMF, right? Well if the FPS data is anything to go by, definitely! Going from 137 FPS average to 265 FPS or so is astonishing – hell the 1 percent lows are actually higher with AFMF on than the average without either version enabled, that’s how much more performance you are getting here. Plus, much like Cyberpunk, the frame generation lag reported by AMD’s driver dropped from 12.6 milliseconds down to just 5.4 milliseconds, which is fantastic. Looking at the latency data that improvement is clear again. With neither mode on it takes around 23 milliseconds for an input to start showing up on screen. With AFMF 1, it’s not too far from double that at 38.4 milliseconds, but with AFMF 2 it’s a lot closer at 31.8 milliseconds on average. That is still an additional 8.7 milliseconds, which for things like accurate gunplay would make it more difficult, but if you’re only getting 30 or 40 FPS with a mid to low end AMD card right now, you’re already not having a great time and a smoother experience is going to make a much bigger difference for you.
I’ll also say that the qualitative experience – ie actually playing some games with AFMF 2 enabled – it’s definitely a bit smoother and less jittery, especially in fast motion. While I still highly recommend you make sure it ISN’T active in competitive games like CS2, Rainbow 6 Siege, Valorant or Warzone, for anything single player – and especially if you have a lower end GPU that isn’t performing too well in the games you play – I think I can finally recommend you turn AFMF on. AMD has done a great job improving this, especially in the latency department, and now the tradeoff is considerably less impactful than with the original version. For strategy games, simulators, and really anything single player, AFMF 2 is a great way to fake-it-till-you-make-it your way to better performance and a smoother gaming experience, with a considerably lower latency cost than before. Great job AMD! Of course like I said if you want to know more about how it works, or why there will always be added latency, check out the first video which explains that in detail – and if you want to know if you should enable Anti-Lag and Anti-Lag 2, check that video out in the cards too.
