跳到正文
GamesIndustry.biz · 全球游戏产业· 来源· Alex Forbes-Calvin·· 2 小时前AI 评分68

EVE Online 开发商 Fenris Creations 开源 CORTAO 渲染技术

Eve Online maker Fenris Creations is adding a rendering technique to its open-source Carbon engine that promises more accurate shading and greater depth

正文 · 原文

Having open-sourced its Carbon engine earlier this year, Eve Online creator Fenris Creations is making a sizeable addition to the tech in the form of CORTAO.

This stands for Circular Occluder Ray-Traced Ambient Occlusion, a new rendering technique that provides more accurate shading and greater depth, while claiming to keep rendering costs low.

It initially started out as a personal project for Fenris graphics programmer and rendering engineer Daniel Isheden and – like all great innovations – started from a wealth of feeling about a particular problem. The feeling, in this case, was "spite" for the existing state-of-the-art rendering technique. Specifically, Isheden was trying to solve some issues around vegetation, notably something that Eve Online does not have much of.

Daniel Isheden
Daniel Isheden | Image credit: Fenris Creations

"The existing state of the art is called Ground Truth Ambient Occlusion (GTAO)," he explains. "I was trying it out and noticing a lot of issues with it. I just realised it didn't actually converge to the ground truth. I started trying to figure out how to actually get something that would converge to the correct thing, given that the information was actually there."

So how does CORTAO work? It's a Screen Space Ambient Occlusion technique (SSAO), which, for the less technically minded, is about calculating how shadows should appear based on environmental lighting.

"These techniques are screen-space limited, so they are only operating on the information that is on the screen," says Isheden. "So it's a very rough approximation in that way. But it doesn't know of anything that is invisible on the screen at the moment. But CORTAO, I believe, is now at least the technique that is able to utilise this information to calculate something that is extremely close to what the correct result should be, given that you knew exactly how the scene looked and not just what was on the screen."

So what makes CORTAO that much better than other techniques? Isheden says that existing techniques focus on "horizontal angles", which causes problems for non-flat surfaces.

"It's a completely different approach that normally shouldn't really be doable"

"Basically, [existing techniques] calculate how much light is coming into a specific pixel by looking at the horizon angles around it," he explains. "The problem is that when you calculate those for sloped surfaces, it simply isn't correct. It's not able to do that correctly given the screen space limitation.

"My technique instead is based on screen space ray tracing. It's instead shooting rays on the screen, up to 32 per pixel, and then, from the number of rays that are occluded or unoccluded, it is able to calculate how much light is reaching that point. It's a completely different approach that normally shouldn't really be doable for performance reasons. The main contribution, or the main part of my work, was actually getting this to run in real time and be fast enough to be useful."

A step up

According to Isheden, CORTAO produces the same result as other rendering techniques, but with about a quarter of the square error. "You can either reduce the quality of CORTAO and still get the same level of accuracy as existing techniques, or you can match the quality and then get better performance instead," he says.

CORTAO graph showing a simple scene
"The simple scene is GTAO's best-case scenario," explains Isheden. "While both techniques converge, CORTAO converges much faster and ends up having a 4x lower error at low sample counts." | Image credit: Fenris Creations

The technique represents a big step forward. "It's basically both faster and cheaper," says Isheden. "That's the long story short. But the real strength of CORTAO is that it manages to achieve a level of quality that previously wasn't possible with these kinds of screen space techniques. It handles depth discontinuities and things like that much better."

There are two versions of CORTAO launching, both open source. The first is Isheden's own iteration, the official reference implementation – releasing under MIT license – while the second is the version coming out via Carbon, which is already open source.

"That is going to be a bit more specialised towards Eve requirements, with the huge depth ranges that we have in space scenes, and being able to have these ambient shadows both on a tiny ship right in front of the player, but also these mega structures, hundreds of kilometres away," Isheden says.

CORTAO complex scene
"The advanced scene shows that GTAO doesn't actually converge to the ground truth, with the error only decreasing up to a certain point and then decreasing again," says Isheden. "This shows that CORTAO is able to achieve a level of accuracy that previously required hardware raytracing and was impossible to achieve with only screen space information." | Image credit: Fenris Creations

As mentioned earlier, CORTAO started off as a pet project for Isheden, something he worked on outside of office hours – and Fenris is "absolutely" open to similar innovations coming from its staff, says senior development director Ben Hunter.

Ben Hunter
Ben Hunter | Image credit: Fenris Creations

"It goes hand in hand with open source, and I think I mentioned this before: the rising tide raises all ships," Hunter says. "We believe that we can share and collaborate on things, and all benefit from them. We saw the potential in Daniel's side project and are very happy that he wanted to find a good use case for it. That's actually part of the process. You can have great innovation, and this is definitely an innovative approach.

"But it also needs to be battle-tested and hardened within a real environment. That's ultimately what Daniel wanted out of this, and then of course we want to benefit from the innovative approach and also being able to contribute to this going forward. And then both of us were very much aligned on having this be open sourced, so others can benefit from it too."

There's one similar project in the works – an innovation from Fenris staff – but Hunter is keeping his lips sealed about it for the time being, save for the fact that it is "very early days" and is more to do with innovating using LLMs.

"That is one of the value adds of LLMs, like the ability to have multiple side projects without affecting current priorities or direction," Hunter explains.

"We can harvest a lot of ideas very quickly, and there are a lot of people doing some interesting things at the moment. But nothing that we're elevating to the point where we can talk about it."

Open-source journey

The last time we spoke with Hunter, around four months ago, Fenris was in the process of making its Carbon engine open source. So far, Hunter thinks that things are going well, although the biggest change is the fact that the company is now developing in the open.

"We are directly submitting into our repositories that can be inspected, and we've had a lot of forks of the repositories," Hunter says.

"I think just on the rendering side alone, which is probably the highest interest, we have at least 70 plus repositories that have been forked off, and people doing various things.

Eve Online
Eve Online's Carbon engine went open source in the summer | Image credit: Fenris Creations

There seems to be a big interest in trying to utilise the rendering side in web applications. Around Eve Online, a lot of the community want to show their ship skins or have various side applications that help them visualise things within the game.

"We've also had around 60–65 community submissions into Carbon – various bug fixes, improvements, and security updates, as well as one or two small feature-related things. We have a couple of very passionate open-source developers who are very much trying to push towards adding Linux support, which is really interesting to see.

"Overall, the response has been really good. It is what we expected: We would have a small cohort of people that would want to contribute and level up what we already have."

The next step is making the tools open source. This was initially set for the end of 2026, but has been bumped into the first quarter of next year.

"Pretty much every day, we're getting a new game team showing us a tool they've just created"

"The main reason for that is that we took a step back and looked at how people will use the tools," Hunter continues. "One of the things we felt quite strongly about was that in the world of LLMs, people are going to want to modify and extend the tools, and we want to ensure there's a structure in place to be able to do that easily. We've spent the past couple of months working on a plugin extension architecture for the tools, which has actually been really, really interesting and very exciting internally, because we never thought that internally we would get so much interest from the game teams who are now extending the tools so much.

"Pretty much every day, we're getting a new game team showing us a tool they've just created in this new plugin architecture. This is exactly what we wanted, because it's showing us internally that the interest is there. A lot of the developers adding these plugins are not engineers; they're also using LLMs. Mainly the artists, but also some of the game designers."

He concludes: "That's giving us validation that this is the right approach. So when we do release them early next year, we know that the community will be able to do the same thing."

来源:GamesIndustry.biz · 全球游戏产业 · gamesindustry.biz