Black Myth: Wukong - to what extent could PS5 Pro deliver an improved experience?
And how could Game Science improve the existing PS5 code?
Black Myth: Wukong is one of the most interesting console games I've seen in a while. There's a clear PC focus, with the PlayStation 5 version featuring a trio of awkward graphics modes - none of which quite hit the target. I also get the sense it's a game tailor-made for the PlayStation 5 Pro, assuming that the features of the hardware are put to good use. Using the PC version of the game, I decided to look at 'BMW' on a number of fronts. Could the resources of the standard PS5 be better utilised for smoother gameplay? To what extent would PS5 Pro be a game-changer? And what of Series S: if Game Science had issues with the PS5 version, can Microsoft's junior Xbox hack it?
First up, let's quickly recap what BMW is doing on PlayStation 5. Quality mode targets 4K and has a dynamic resolution between 1224p and 1584p. Frame-rate looks to be unlocked to my eye, so I'm curious how the dynamic resolution system may be working. Even so, I think this is the most competent mode in the current game. Next up, the balance mode: this looks like a hard-set 1080p resolution with no DRS but an utterly bizarre 45fps cap which makes it jerky on any display likely to be connected to the PlayStation 5. Finally there's the performance mode - this is using frame generation from 30fps to 60fps. It's fundamentally a bad idea as one of the key ideas behind a performance mode is to reduce lag, not increase it as is the case here.
For my tests, I used the Digital Foundry Frankenstein's PC - a Windows unit built on a motherboard hosting the actual Xbox Series X CPU, paired with the Radeon RX 6700, an AMD graphics card that has much in common with the PS5 GPU. It's a mish-mash of console-equivalent parts, yet it's proven to be capable of delivering very console-like results. By using the balance mode with its fixed resolution, paired with graphics options matched to console equivalents, we can benchmark our set-up to see how close to PS5 performance we get. It turns out that the console is slightly faster - around seven to 10 percent in some scenarios. So, when we carry out our tests, we could likely expect console performance to actually be a touch faster.
PS5-Equivalent | Optimised Lower-End | Optimised High-End | |
---|---|---|---|
View Distance | High | High | High |
Post Effects Quality | High | High | High |
Shadow Quality | High | High | High |
Texture Quality | Low/Medium | Very High | Very High |
Visual Effects Quality | High | Medium | Medium |
Hair Quality | High | High | High |
Vegetation Quality | High | High | High |
Global Illumination Quality | High | High | Very High |
Reflection Quality | High | High | Very High |
So let's begin by talking about how BMW's quality modes could be improved. Honestly, the precedents have been set by countless other games but I think the most recent, best examples of mode selection are found in the Ubisoft Massive titles: Avatar Frontiers of Pandora and Star Wars Outlaws - a 30fps quality mode, a 40fps balanced mode than runs in a 120Hz container for consistent frame-pacing and finally a 60fps performance mode.
Quality mode first: I think this is actually the easiest way to improve BMW for PS5 owners. Right now, there's no 30fps cap. I think it may actually have the same 45fps cap as the balance mode actually, based on my own captures - it's just that the higher resolution keeps frame-rates down into the 30s. A consistent 33.3ms per frame would make all the difference, along with a switch from FSR upscaling to Epic's own TSR. It's an easy win and makes the overall experience much smoother. I honestly think this is a pretty simple tweak that Game Science could introduce to the existing game.
Balance mode next, where I think Ubisoft Massive's solution of opting for a 40fps mode within a 120Hz container - 25ms per frame - is the best solution. What I found is that the higher you try to push frame-rate in BMW, the harder it is to maintain consistent performance, even if you reduce resolution. That's what I had to do, dropping output resolution to 1728p (80 percent of 4K) with TSR upscaling from 50 percent of that (864p). It sounds low, but looks fine overall - and console could use unused GPU cycles for increased resolution via DRS. Even so, the toughest scenes still dip beneath 40fps. I think this would likely work, but may not be completely stable.
Performance mode is tricky. Game Science's solution was to take its balance mode and to add frame generation to cap at 60fps. At best this is running internally at 30 frames per second, with frame gen boosting that to 60fps - and it just doesn't work on a number of levels. First of all, it's easily the laggiest way to play the game on PlayStation 5 - completely at odds with the basic notion of what a performance mode should be adding to the game: sharper, crisper response. Secondly, with a 16.7ms per frame persistence, the deficiencies of generated frames are far more obvious to the human eye than they should be. Remember, AMD is mandating a 60fps minimum, but I personally think a 40fps minimum with an unlocked frame-rate works well for VRR screens, as seen recently in Immortals of Aveum on consoles. I found that targeting a 1080p output resolution upscaled with TSR from 720p was the best result overall - no frame-gen required.
I do think frame-gen could play a role though - I increased output resolution to 1440p, upscaled from the same 720p and found that with a 120Hz output, we could hit 70fps to 90fps, with the same dips below 60fps in the most challenging scenes. With 120Hz output, we are effectively running unlocked and paired with a VRR output, the results are fine.
This is clearly a super-demanding game and I think it's tailor made for PS5 Pro, assuming Game Science properly deploys the new console's features. And those features are roughly 45 percent more raw GPU power than PS5, much enhanced ray tracing features and a machine learning block that should - hopefully - produce results in the ballpark of Nvidia DLSS. So what GPU in the PC line-up ticks all those features? Not an AMD one, curiously, as there is no ML hardware like PS5 Pro and no enhanced RT. I settled on the Nvidia RTX 3070 Ti, which seems to offer much the same feature set mooted for the Pro and does indeed tend to run around 45 percent to the better up against a base PS5. Let's be clear here, we're talking about broadly similar projected performance here - but even so, I'd describe this as transformative.
Looking first of all at a quality mode targeting 30 frames per second, we get a far, far superior result. I can run this at a mixture of PS5's high along with very high global illumination and very high reflections - so slightly better than our own optimised settings selection. However, the big deal here is really the improvement to image quality brought about by machine learning-based upscaling. DLSS is class-leading and we've no idea if Sony's PSSR can match it, but using a 1440p base resolution delivers startling good quality. RT? We need a better idea of the Pro's capabilities there before making any projections. We're running at 30fps with horsepower to spare - so, again, console DRS could make the difference.
For performance mode, I set up a 1728p custom output resolution and set a base resolution of 50 percent, using DLSS to upscale. Settings here are kept at the same high preset as the PlayStation 5, but of course, we get improved texture fidelity. This is certainly a good improvement over the base PS5's frame generated extravaganza and certainly eclipses the upscaled 1080p alternative I posited earlier as a suggestion for the base machine, but we still do have some minor dips beneath 60 frames per second with the arrangement you see here. If you recall, the PC version on matched settings did run a bit slower than the console - so maybe that's enough to make up the balance.
Performance drops don't happen that often and the drops are fairly minimal but we are talking about an 864p base resolution here - it's an example of how difficult it is to scale up performance on Black Myth Wukong. And it's also a reminder that even though we should expect to see some impressive upgrades from the PS5 Pro, certain elements of the architecture won't scale amazingly well.
Finally, I wanted to see whether Xbox Series S-level horsepower could handle BMW. The truth is you can get a pretty decent experience with this game even on something like an Asus ROG Ally - but we've also seen with the PlayStation 5 version that Game Science is prioritising visual settings, and may be reticent to bring them down. I simulated Series S performance here with a Radeon RX 5500 XT - 22 compute units vs Series S at 20 - and I'm achieving a similar four teraflops of GPU compute by dropping clocks down to 1420Mhz. Even with my PS5-class performance mode settings, I could not achieve a stable 30fps and had to drop down to medium settings. And even here, the game still ran slower relative to the RX 6700 targeting an upscaled 4K 30fps.
To be clear, I do think that BMW will work on Xbox Series S but there are two challenges here. First of all, Game Science may need to cut deeper into fidelity and resolution than we may like. And secondly, even though we can simulate relative GPU compute with the 5500XT, we can't simulate the overall lack of memory Series S has. I'll be interested to see how this game is adapted to Series S. Series X? I wouldn't expect Game Science to have any more problems than they've had with PS5.
This proved to be an interesting experiment, playing around with the limits of both PS5-level hardware while approximating what PS5 Pro can bring to the table. I genuinely think that it's the machine learning upscaling that makes the Pro viable. Without it, the bump in GPU horsepower alone would not justify a new product, let alone one at that price-point. As to whether we may have underestimated the Pro's capabilities… well, the games will do the talking in due course and we'll be there to cover them.