One of the Ark developers spoke highly of the latest version of Epic Games‘ technology.
When Epic Games launched Unreal Engine 5.7 (UE 5.7) a month and a half ago, it appeared to be a significant improvement, at least on paper. Although still in the experimental stage, the Nanite foliage feature looked very exciting, and MegaLights promised to increase the number of dynamic shadow-casting lights developers could add to games without compromising frame rates. But would these additions and changes be just as beneficial in an actual game? WCCFTech asked Studio Wildcard co-founder, co-creative director, lead designer, lead programmer, and development director Jeremy Stieglitz that very question.
When asked about the change in reception of Ark: Survival Ascended since its early access release two years ago and how the latest user reviews finally achieved a “mostly positive” rating on Steam, Stieglitz cited performance improvements as the main reason. Stieglitz then revealed that UE 5.7 is already in internal testing and expected to be released in March. He said it is close to being called a miracle cure in terms of improved performance.
“When the game was first released, it didn’t run very well. Frankly, it still doesn’t run as well as it should, but performance has improved by about 50% since the initial release of the game on the same hardware. While this is a significant improvement, it’s not ultimately where we want to be. Where are we going to get more performance from? A big part of it is Epic‘s latest version of Unreal Engine, 5.7, which introduces great, performance-oriented improvements and systems. Version 5.6 did as well, but we’re still on version 5.5. The performance improvements of 5.6, combined with the new features of 5.7 will gain us another 30 to 40% on both the GPU and CPU. UE 5.6 optimized the render thread, making it work much better in parallel with the game thread, GPU, and CPU. UE 5.6 has major architectural changes, and UE 5.7 has an entirely new system for drawing large foliage, as demonstrated in The Witcher IV demo.
We can actually utilize that directly. We have it working internally. We’re just preparing it for release. It will take a few months to work out the kinks in the upgraded parts of the engine, but we can already say that we’ve achieved a third more performance on our existing maps thanks to the new Nanite tessellation and culling system that Epic introduced in UE 5.7 for foliage. We’re really excited about that. It’s as close to a magic bullet as you’re going to get at this point. It’s one of those rare cases where it’s a clear win. There are no real caveats or asterisks. It works much better than the previous version, and we don’t have to sacrifice anything to make it work. It was the closest thing to a magic solution for some of those performance problems. We expect the UE 5.7 upgrade for Ark: Survival Ascended to be available by the end of March of this year,” said Stieglitz.
When asked if UE 5.7 would help eliminate the stuttering players endure in Ark, and other Unreal Engine games, Stieglitz replied that it would, but cautioned that players shouldn’t expect it to disappear completely.
“It should be better, but I don’t know if it will be completely eliminated because of two factors. First, in Ark in particular, some of the stuttering is due to the initiation of actors that cannot be multi-threaded. When playing an online game, if you enter a new area with a giant base built by another player, the base must load on the client instantly as soon as the data is received. We could try multi-threading it, but UE5 is poor at network actor instantiation and multi-threading. I don’t know if they have an inherent system for that.” In other words, when the client receives an actor, such as a dinosaur or a building, over the network, it immediately spawns it to start receiving the associated network data. Now, multiply that by 50,000. Receiving 50,000 structures in a big batch of data in a single frame will result in a hitch, as you can imagine. This has nothing to do with graphics or background loading of the world. The issue is simply that 50,000 game objects were received over the network, and they need to be prepared for play because network data is about to be received.
That’s a type of stuttering unique to Ark. It’s probably the main type of stuttering that people experience when playing online, especially when building large bases. Another kind of stuttering is more associated with background loading of the world. Much of that has been eliminated in more recent versions of Unreal Engine, but I think one issue is particular to Unreal games in general: shader caching. We don’t do pre-shader caching, so I think it generates shader code on the fly from HLSL-based intermediate shader code and prepares it for uploading to the GPU. This process is specific to your GPU hardware and can result in stalls. Many Unreal games eliminate this issue by pre-caching all the shaders they can for your hardware at startup. I don’t like doing that with Ark because we have so much content that it wouldn’t be practical. Second, we do modding, which makes it even more unpredictable what materials and shaders we’ll need to compile.
In practice, I think modding would render it ineffective, even for the base game content. There would still be stalls because, for one thing, we allow people to create custom cosmetics when they play online. This is an option that people can turn off, but by default, these cosmetics are downloaded dynamically as people play. Therefore, you may encounter players with random cosmetics on their characters or bases while playing the game. Those cosmetics can have unique materials and therefore unique shaders. There’s no real way to pre-cache those because we don’t know what they are in advance. Over time, I think Epic might be able to optimize these things even better than they are now. However, I do think hitching in a game that doesn’t pre-compile all the shaders will always exist to some extent. I will say that it’s improved significantly since earlier versions of Unreal Engine 5, and I expect to see continued improvements as Epic continues to refine the engine. We expect to upgrade the engine at least once a year going forward. We got a good sense of how to do it reasonably efficiently with our upgrade from 5.2 to 5.5 last June. We’ll do it again with UE 5.7 in March, which we’ve already started, and then with whatever version Epic has in early to mid-2027,” Stieglitz added.
Version 5.7 of Unreal Engine is a significant improvement, but it’s not perfect yet.
Forrás: WCCFTech




Leave a Reply