Horizon Forbidden West Performance (Patreon)

Hey, Rich here. One topic we really didn't cover in our Horizon Forbidden West DF Direct was the frame-rate. Why? Well, for a whole number of reasons. First of all, there's nothing really interesting to talk about. It's 30 frames per second. There is a *tiny* little bit more to that story, but really, we've learned from the PS5 reveal that early promotional materials don't tell the full story and don't really deserve the full-on graphing treatment.

Ratchet and Clank: Rift Apart had a 30fps trailer, but you'll be able to play it at 60fps. RE Village had a 30fps trailer. It's a 60fps game. In fact, most of the games shown at the PS5 reveal were presented with 30fps assets and just about every game in the generation so far runs at 60fps. Whether Horizon Forbidden West will get a 30/60 toggle remains to be seen but as a cross-gen game, I'd say its chances are good.

I did run a performance test on the footage though, but I'm not putting it public for the general audience for a couple of reasons. First of all, we only had a 30fps asset - which means that if for whatever reason the game was running faster, you will never know. Secondly, the nature of a 30fps asset is that even with a 30fps game, you are missing half of the frames output by the PS5. That means that if there are frame drops, some may be missing. It also makes frame-time meaningless as a dropped frame in a 30fps feed will present as 66ms frame persistence. That's why instead of frame-time in the video, you get 'detected dropped frames' instead - if we find a dip, it's graphed. But because of the source, some drops won't be. Oh, and finally, when you don't capture the footage yourself, you can't say 100% for sure that there are not minor issues with the capture solution chosen.

So after that galactic level swathe of caveats, what does the video above demonstrate then? Well, pretty much a clean bill of health. The only small issue appears to be with scene cuts in cutscenes, or when Aloy jumps into the water. Hitches on cutscenes are commonplace in games - TAA doesn't work well with sudden camera jumps so some titles buffer a frame to allow the temporal accumulation to 'catch up'. You'll note that some scene cuts show the buffering, others don't, likely because we only have a 30fps feed and half of the actual generated output from the console.

So consider this as background research really, more a verification of what our eyes were telling us! Oh, and of course, usually we rely on lossless feeds to ensure accuracy, so how do you analyse a lossy capture ripped from YouTube? The fact it runs with v-sync solves a lot of the problem. You just compare one frame to the next. On a lossless feed, the frames should be 0% different. With a lossy feed, you set a new threshold - say, 15% - then check every dropped frame manually by eye. There weren't many of them.

And the final, final reason we didn't punt this video live? Generally, we're happy to test public-facing code but marketing materials like this... not so keen, these days. In the past we've done pre-release stuff and a good example of this is Horizon Zero Dawn from the Pro reveal, where Sony actually put out a 60fps asset for a 30fps game so we could do a full analysis with frame-time graphing. We did the perf test there and didn't really represent the final game which looked better, played better and ran more smoothly: https://www.youtube.com/watch?v=JJCKkbV80ts