Performance Analysis: PlayStation 2 emulation on PS4
How much faster can Sony's emulator run classic PS2 titles?
We've already established the basics. Sony's new PlayStation 2 emulation technology for PS4 runs original software with a number of key enhancements: resolution gets a ballpark 4x increase and trophy support is also added. But as we put the final touches to our initial look at the three Star Wars 'classics' revealed so far, one outstanding issue remained: just how much faster can the emulator run PS2 titles compared to original hardware?
With a sample of just three somewhat mediocre PS2 titles to experiment with, it's safe to say that we can't draw too many conclusions right now. Regardless, there is a lot of interesting data here - principally because all three titles run with an unlocked frame-rate, and each possesses a wildly variable level of performance running on original hardware. Given enough computational horsepower, all of these games can theoretically run at a completely locked 60 frames per second. PlayStation 4 doesn't quite hit that target, but for much of the duration, it gets very close, and the experience in those titles is transformative.
Two of the titles we have to examine - Jedi Starfighter and Bounty Hunter - run with v-sync enabled, and both are double-buffered. Essentially, what that means is that while one frame is scanned out to your display, the next is being rendered internally - the idea being that when your screen next refreshes, the new frame is swapped in. In optimal conditions, that provides a slick, smooth 60fps. The only problem is that both of these titles often run over their render-time budgets, frequently missing the next display refresh, meaning a sudden, often sustained drop down to 30fps or lower. The end result is some pretty unsightly judder as the performance level switches erratically between very different frame-rates, plus very different levels of response from the controls at any given point.
Very few modern titles perform like this. A 30fps frame-rate cap levels out the experience, or other techniques like triple-buffering or adaptive v-sync are used to level out performance dips when a new frame runs over its rendering budget. However, in ascertaining just how much faster the PS2 emulator is, these titles' sub-optimal set-up actually allows Sony's new PS4 tech to shine. We are at 60fps for most of the duration, though it's interesting to note that graphically heavy effects - such as alpha transparencies - can still impact frame-rates. There's a clear and sustained uplift, but there's not a bottomless pit of additional performance available.
As interesting as the results are with Jedi Starfighter and Bounty Hunter, it's not really the full story. A hard, double-buffer v-sync like this is effectively causing the GPU to stall, limiting performance on the original PlayStation 2 hardware and possibly exaggerating just how much faster the PlayStation 4 emulation really is. And that's why the third Star Wars title we have to look at - Racer Revenge - is actually the most illuminating game in the pack.
For this title, the original developers at Lucasarts decided to opt for a smoother refresh and more responsive controls. To facilitate this, adaptive v-sync was utilised. To put it simply, if the engine can't produce a new frame to match the refresh of the display, it'll swap in the new image as soon as it is ready - as the screen is in the process of updating with the new frame. This produces smoother gameplay, but it comes at a cost: you'll see on-screen tearing. In the case of a racing game, opting for adaptive v-sync is the right call - but it has an extra special bonus for us with our performance testing. The code isn't trying to synchronise with the display refresh so there are no GPU stalls - Racer Revenge is running flat-out. It's the best opportunity we have to see just how much faster the emulation actually is.
The short answer is that it's a lot faster. Racer Revenge can see the PlayStation 4 emulation running the title at 60fps while the equivalent scene limps along at 33fps on original PS2 hardware. In fairness, that's one of the widest differentials we saw - but more regularly we are looking at a 20fps delta, so in many scenes, Racer Revenge is handing in a 50 per cent increase in frame-rate. That result is actually skewed a little in the PS2's favour too - curiously, the emulator seems to be ignoring the original game's adaptive v-sync set-up, with just some very rare single torn frames on-screen. However, even as a baseline, that's still impressive.
Questions remain about how representative these games are of overall emulator performance - especially in more challenging titles, possibly those where the CPU is put through much more of a thorough work-out. However, the possibilities here are intriguing, even if the GPU uplift is less pronounced on other games compared to these old Star Wars titles. Shadow of the Colossus operates with a hard double-buffer v-sync, just like Jedi Starfighter and Bounty Hunter - just how much would it benefit running under emulation?
And then there are the adaptive v-sync titles on PlayStation 2 - games like Gran Turismo 4 and the God of War releases. If there is a decent amount of additional GPU performance available there, potentially these titles could run at a locked 60fps. Combine that with the tangible increases to image quality the emulator hands in, we'd love to revisit those titles - and comparisons with PS3-era remasters could be intriguing too...