A prominent open-source dev publishes their findings as to what's going on with Starfield's performance, and it's pretty darn strange.
According to Hans-Kristian Arntzen, a prominent open-source developer working on Vkd3d, a DirectX 12 to Vulkan translation layer, Starfield is not interacting properly with graphics card drivers.
People really have no idea about anything in game development. I agree it should have seemless planet travel, but it is not something that an engine “can just do.” It takes so many complicated systems to make that function. There’s no engine that does it out of the box.
Basically any engine can do it, but it requires it to be built. The land must be deterministic at all points, it must be able to create chunks accurately for all points (which gets really weird at the poles, but any latitude above 0 because your chunks shouldn’t be square anymore), and they must be able to be streamed in to their correct position seemlessly.
It is quite complicated, and there’s no reason the engine developed for an arena shooter (Unreal) would be able to handle it any better than any other engine. It just has to be built.
There’s a reason Hello Games wrote their own engine for NMS. We all know that it was pretty bad gameplay-wise at launch, but under the hood NMS was (and still is) something of a technical marvel. No loading screens except for a disguised one when jumping between systems is quite impressive.
Also, IIRC, NMS doesn’t have different gravities, right? Been a year or two since I properly played, but I don’t remember ever really jumping higher or being forced to the ground. That’s one of the sacrifices for seamless landing.
I don’t buy this. Plenty of games allow you to adjust gravity on the fly using console commands. All they would have to do if you enter a new planet’s atmosphere, is adjust the gravity value.
Source engine has allowed this forever, changing gravity on the fly. No reason it can’t be implemented in other engines.
I guess though I mean it is expected at this stage of game development for this genre to have something like seamless planet travel for a space game. Like it didn’t have to be NMS or Elite Dangerous, they could’ve copied something like how Jedi Fallen Order did it, where basically your ship takes off from the planet, jumps to hyperspace and loads the next one during hyperspace and lets you know when you’re ‘arriving’ (aka when the destination is loaded) and you then take an action and land on the loaded planet. It ends up being the same thing as what Starfield basically does but handles it much more deftly.
Idk, just saying there’s better ways they could’ve handled it even if the engine couldn’t handle seamless planet travel in a traditional sense.
I have no game dev experience but I have a math and software background. I’m just curious about what “it gets weird at the poles” means. If I wanted to (abstractly) generate tiny square chunks of a large sphere, I would generate them as (proper) squares and then pass them through an explicit diffeomorphism to the associated region of the sphere, relying on the relative smallness to guarantee that the diffeomorphism doesn’t change things too much. From a game dev perspective, what approach do you take that causes issues at the poles?
Imagine trying to find the intersections of a line or region as it crosses multiple cells of a non-euclidian “grid” near the poles where an entire axis can flip from one cell to the next.
Are you suggesting using a stereographic projection? That seems like a bad idea. You wouldn’t want your projection to depend on the coordinate system. Am I missing a reason why you wouldn’t use proper, nonsingular spherical coordinates?
Games, support libraries, and engines don’t really support spherical coordinate systems. If you don’t want to write everything from scratch, you gotta go Cartesian.
Sure, I guess, but constantly mapping between them gets complicated and adds overhead. Plus, now you are dealing with curves instead of lines when checking for intersections, and that gets far more expensive to compute when you are trying to do thousands if not millions of checks per frame when trying to run at 60 or 120 frames per second.
I’m not saying it isn’t possible, just that games haven’t traditionally been written that way, so you can’t build on what they have already figured out. That makes it harder to find people who have game dev experience in that kind of math.
People really have no idea about anything in game development. I agree it should have seemless planet travel, but it is not something that an engine “can just do.” It takes so many complicated systems to make that function. There’s no engine that does it out of the box.
Basically any engine can do it, but it requires it to be built. The land must be deterministic at all points, it must be able to create chunks accurately for all points (which gets really weird at the poles, but any latitude above 0 because your chunks shouldn’t be square anymore), and they must be able to be streamed in to their correct position seemlessly.
It is quite complicated, and there’s no reason the engine developed for an arena shooter (Unreal) would be able to handle it any better than any other engine. It just has to be built.
There’s a reason Hello Games wrote their own engine for NMS. We all know that it was pretty bad gameplay-wise at launch, but under the hood NMS was (and still is) something of a technical marvel. No loading screens except for a disguised one when jumping between systems is quite impressive.
Impressive for sure. They had to choose to not have a lot of things to do it though. They knew what they wanted and did it, which is smart.
Also, IIRC, NMS doesn’t have different gravities, right? Been a year or two since I properly played, but I don’t remember ever really jumping higher or being forced to the ground. That’s one of the sacrifices for seamless landing.
I don’t buy this. Plenty of games allow you to adjust gravity on the fly using console commands. All they would have to do if you enter a new planet’s atmosphere, is adjust the gravity value.
Source engine has allowed this forever, changing gravity on the fly. No reason it can’t be implemented in other engines.
I guess though I mean it is expected at this stage of game development for this genre to have something like seamless planet travel for a space game. Like it didn’t have to be NMS or Elite Dangerous, they could’ve copied something like how Jedi Fallen Order did it, where basically your ship takes off from the planet, jumps to hyperspace and loads the next one during hyperspace and lets you know when you’re ‘arriving’ (aka when the destination is loaded) and you then take an action and land on the loaded planet. It ends up being the same thing as what Starfield basically does but handles it much more deftly.
Idk, just saying there’s better ways they could’ve handled it even if the engine couldn’t handle seamless planet travel in a traditional sense.
I have no game dev experience but I have a math and software background. I’m just curious about what “it gets weird at the poles” means. If I wanted to (abstractly) generate tiny square chunks of a large sphere, I would generate them as (proper) squares and then pass them through an explicit diffeomorphism to the associated region of the sphere, relying on the relative smallness to guarantee that the diffeomorphism doesn’t change things too much. From a game dev perspective, what approach do you take that causes issues at the poles?
Imagine trying to find the intersections of a line or region as it crosses multiple cells of a non-euclidian “grid” near the poles where an entire axis can flip from one cell to the next.
Are you suggesting using a stereographic projection? That seems like a bad idea. You wouldn’t want your projection to depend on the coordinate system. Am I missing a reason why you wouldn’t use proper, nonsingular spherical coordinates?
Games, support libraries, and engines don’t really support spherical coordinate systems. If you don’t want to write everything from scratch, you gotta go Cartesian.
You can still use local Cartesian coordinates.
Sure, I guess, but constantly mapping between them gets complicated and adds overhead. Plus, now you are dealing with curves instead of lines when checking for intersections, and that gets far more expensive to compute when you are trying to do thousands if not millions of checks per frame when trying to run at 60 or 120 frames per second.
I’m not saying it isn’t possible, just that games haven’t traditionally been written that way, so you can’t build on what they have already figured out. That makes it harder to find people who have game dev experience in that kind of math.