PhysX, CPU or GPU?

[Jeq]

Does it support GPU based physX instead of forced CPU? I am trying to guess how smooth gameplay we will have.

[shdwlrd]

Don't know. Nothing has been said about that.

[NoMrBond]

It seems unlikely that they'll work on integrating the CUDA branch of PhysX over the integrated CPU-only branch of PhysX as this would only benefit PC players, and then only those with suitable nvidia cards

Could get HAVOK though?

[K^2]

CPU PhysX only, but we might get some of the multithreading optimizations from recent Unity versions.

Havok's unlikely, as that requires ECS/DOTS setup, and it sounds like Intercept is building KSP2 on GameObjects, same as original KSP. It would be amazing for both stability and performance, but seems unlikely.

[Incarnation of Chaos]

Unity's implementation of PhysX is CPU only, so absolutely no on GPU accelerated physics.

But they have plenty of other ways to get it smooth.

[Jeq]

Sounds like we get same or worse performance as original KSP. Worse if physx is same as now and there will be more calculations done. Physics needs be totally rewritten or this will be disappointing sequel.

Edited January 15, 2021 by Jeq

[Incarnation of Chaos]

7 minutes ago, Jeq said:

Sounds like we get same or worse performance as original KSP. Worse if physx is same as now and there will be more calculations done. Physics needs be totally rewritten or this will be disappointing sequel.

I don't know why everyone thinks the only approach is to either put it on the GPU or multiple threads.....

They've spoken at length about a physics LOD system that will take large numbers of parts that aren't changing much between cycles and "Weld" them into a single part for the purpose of those calculations until each one individually is needed for the calculations again (Like a collision). That alone will massively reduce the CPU load, as the amount of calculations needed as part count increases isn't linear.

There's also much, much they can do just from the backend, and reducing the amount of poorly optimized code that's updating when it isn't needed or likewise.

Basically, there's no reason to think KSP2 will be a performance hog unless they absolutely fail on all counts. We'll know for sure once it releases.

[K^2]

Well, nobody actually really does physics on GPU anyways. Yes, I know that was supposed to be one of the big selling points of PhysX, but nobody wants to use GPU resources for it, and most of the stuff you'd want to do massively in parallel for physics, GPUs weren't terribly good at. With RTX, that can actually change quite a bit, because BVH accelerators for RT can be used for collision checks as well. But if your options are to have ray tracing or GPU physics, which one are you going to chose? Pretty much every game out there goes with more graphics and does all the physics on CPU as that's usually enough anyways.

Even KSP isn't bound so much by the amount of physics that needs to be done as by how bad Unity is at physics. Newer versions of Unity are better at utilizing CPU resources, so we should see improvements in physics even just from that. And yes, there are a lot of optimizations that can be taken. I am a bit concerned about how these optimizations were talked about early on, because that was happening at the same time Intercept had an opening for a senior physics engineer, and we haven't heard anything about it since that position got filled, but even replicating what KSP did with some Unity improvements they get "for free" should be a good start. If they can combine some modules into larger rigid bodies, that will help a lot, and shouldn't be too difficult to do, so fingers crossed.

Just please, remember to use parallel axis theorem correctly. If I had a nickel for every time I have to go into a physics engine and fix how the moment of inertia tensor is built, I'd have a few nickels, which isn't a lot, but it's weird that this keeps happening.

[Incarnation of Chaos]

7 hours ago, K^2 said:

Well, nobody actually really does physics on GPU anyways. Yes, I know that was supposed to be one of the big selling points of PhysX, but nobody wants to use GPU resources for it, and most of the stuff you'd want to do massively in parallel for physics, GPUs weren't terribly good at. With RTX, that can actually change quite a bit, because BVH accelerators for RT can be used for collision checks as well. But if your options are to have ray tracing or GPU physics, which one are you going to chose? Pretty much every game out there goes with more graphics and does all the physics on CPU as that's usually enough anyways.

Even KSP isn't bound so much by the amount of physics that needs to be done as by how bad Unity is at physics. Newer versions of Unity are better at utilizing CPU resources, so we should see improvements in physics even just from that. And yes, there are a lot of optimizations that can be taken. I am a bit concerned about how these optimizations were talked about early on, because that was happening at the same time Intercept had an opening for a senior physics engineer, and we haven't heard anything about it since that position got filled, but even replicating what KSP did with some Unity improvements they get "for free" should be a good start. If they can combine some modules into larger rigid bodies, that will help a lot, and shouldn't be too difficult to do, so fingers crossed.

Just please, remember to use parallel axis theorem correctly. If I had a nickel for every time I have to go into a physics engine and fix how the moment of inertia tensor is built, I'd have a few nickels, which isn't a lot, but it's weird that this keeps happening.

Yeah, last time i saw it was Borderlands 2.....

And it only helped performance if you were massively CPU bound....

[Guest]

Quote

Intercept had an opening for a senior physics engineer

And I'm pretty sure that Squad, the Mexican multimedia company, didn't have access to that kind of resources when they developed the physics of KSP1.

I'm no expert and there's probably some bottleneck down the road on how much you can optimize KSP2 physics, but I bet that KSP1 is nowhere near that level of optimization.

[K^2]

52 minutes ago, Master39 said:

And I'm pretty sure that Squad, the Mexican multimedia company, didn't have access to that kind of resources when they developed the physics of KSP1.

I'm no expert and there's probably some bottleneck down the road on how much you can optimize KSP2 physics, but I bet that KSP1 is nowhere near that level of optimization.

A lot of the work that HarvesteR did would fall under the responsibilities of physics programmer, so I would argue that Squad did, in fact, have these kind of resources. Availability of said resources to focus on physics tasks is a separate matter, however.

KSP physics performance was abysmal. Part of it was certainly due to Squad having to work with limited resources, and part due to just the state of physics on Unity being in a really bad place, especially, at the time. KSP2 is in a better starting place just from having a more up-to-date version of Unity. So we should have better physics performance right away. However, Squad also spent a lot of time tuning joints to get stability, and the only footage I've seen of KSP2 where you can see physics shows they haven't replicated that at that point. It's been a few months, and hopefully, they're further along, but it does point to the fact that they haven't even began core physics improvements. And that doesn't leave a lot of time to actually work on such features. So I'm concerned about how much optimization will actually get done. Does it have to be done to ship the game? Not really. So long as they at least do the same stability pass that was done on KSP to reduce cases of Kraken, it'll be fine on PC and next gen. PS4/XB1 performance is where it's really going to hurt. It's also going to put some limits on how big we could build on any given hardware. But since I don't expect KSP2 to be that much heavier on physics, maybe that's good enough?

[Incarnation of Chaos]

2 hours ago, K^2 said:

A lot of the work that HarvesteR did would fall under the responsibilities of physics programmer, so I would argue that Squad did, in fact, have these kind of resources. Availability of said resources to focus on physics tasks is a separate matter, however.

KSP physics performance was abysmal. Part of it was certainly due to Squad having to work with limited resources, and part due to just the state of physics on Unity being in a really bad place, especially, at the time. KSP2 is in a better starting place just from having a more up-to-date version of Unity. So we should have better physics performance right away. However, Squad also spent a lot of time tuning joints to get stability, and the only footage I've seen of KSP2 where you can see physics shows they haven't replicated that at that point. It's been a few months, and hopefully, they're further along, but it does point to the fact that they haven't even began core physics improvements. And that doesn't leave a lot of time to actually work on such features. So I'm concerned about how much optimization will actually get done. Does it have to be done to ship the game? Not really. So long as they at least do the same stability pass that was done on KSP to reduce cases of Kraken, it'll be fine on PC and next gen. PS4/XB1 performance is where it's really going to hurt. It's also going to put some limits on how big we could build on any given hardware. But since I don't expect KSP2 to be that much heavier on physics, maybe that's good enough?

tbh if they haven't gotten Physics LOD working, then that's going to severely limit the scope of their game.

They could work around it by using much larger, single parts to replicate the functionality of multiple, smaller ones. But if that was absent on launch, I'd be tempted to not buy KSP2. And it doesn't seem like something they could patch in later with how deeply that would be ingrained into the game itself. PS4/XB1 Performance was always going to blow chunks, the question was just how many.

It comes back to where I've been the entire time, if they've spent the majority of the time on the CORE of the game then KSP2 will likely be a mess on launch....but fixable. If instead they never managed to get these enabling features/optimizations working? Well then it's never coming back after they go live.

It's why I've repeatedly said i don't give a **** about fancy graphics, or clouds or anything much else in KSP2 for that matter. Because before i can admire the surroundings of the game, they need to make sure it can actually sustain the demand placed on it. Do i think they can pull it off? They might already have, we don't know what internal builds they're working with. But could it also be a massive dumpster fire? Yep

I will say; either way it'll be fun to see the post mortem.

[K^2]

9 hours ago, Incarnation of Chaos said:

It's why I've repeatedly said i don't give a **** about fancy graphics, or clouds or anything much else in KSP2 for that matter. Because before i can admire the surroundings of the game, they need to make sure it can actually sustain the demand placed on it.

It's not an either-or, regardless. It takes very different people to work on these tasks. It looks like Intercept got a good amount of art and tech art people from Star Theory, so they're running with what they have. I can't quite tell what the situation was at Star Theory, but as of formation of Intercept, they needed a physics engineer, and it took them a while to find one. And it's still not clear if they got somebody senior enough to make anything like the LoD system they've talked about.

Naturally, it's all limitation of budget. You can sink enough money into the search to guarantee you'll find people who will handle the work that needs to be done, but if it costs you as much as hiring five engineers, you're not going to ship on budget.