ZDoom

Sorry for replying on this old topic, but i felt it was necessary.

I'm amazed that the devs are actually responding to this, i mean you can get great performance if you tune the video settings. And lets be honest, for a map this silly some measurements must be taken.

Running gzdoom on an old dualcore AMD, no problem:

nuts has the same problem as my raytracing brute force demo, the problem isn't the render and the monsters being activated, but yes the math behind the code to check if a monster see and hear the player, it's not a 1+1 like math, it plays with sqrt, pow and even other demanding methods where using for 1000 cases is ok, but not for 100000000 cases. (and you can do whatever magic you want to save performance, but at the end you'll end up just saving 1 - 3fps)

I'll just do a guess, but gzdoom may be just faster because of the render overhead that is not done 100% by the cpu

Actually, that won't help that much really. The main problem is not A_Chase but what it calls internally, down to P_TryMove and PIT_CheckThing

Yes, that is true, but even a simplified A_Chase will still perform better than the current one.

Idea: A_NativeChase and A_NativeVileChase? For lack of a better term.

A scripted A_Chase is a lot slower than the native version, though, so this won't do much good.

The problem is A_Chase and A_Look. A_Look is not as costly, but you will notice even if you run "notarget" and play the map it does not have as big of a performance impact as letting the monsters run wild.

If you really absolutely must have a playable nuts.wad - you will need to recreate all the monsters it uses using a more simplified (probably just copied from the original DS and translated to ZScript) A_Chase function. However keep in mind that even this will not magically solve all your problems - there's still the overhead of other parts of the engine which will bottleneck the performance in various ways.

Yes, Graf is right that the sprites themselves do cause problems - but they're only part of the bigger problem in running nuts.wad on ZDoom.

Enjay wrote:I actually played it the other day using the bullet time mod. For some reason, all the time freezing kept things reasonably playable and I kept going until I had 100%:kills.

Interesting. Could it be that because the world is frozen, the monsters' Tick() methods are not being ran, therefore saving up on CPU cycles?

Nevander wrote:Why does nuts tank the performance?

I think it's becouse all monsters are activated at the same time.

I actually played it the other day using the bullet time mod. For some reason, all the time freezing kept things reasonably playable and I kept going until I had 100%:kills.

What I'd like to know is what is it about nuts that causes the massive FPS plummets whereas Holy Hell MAP05 runs pretty well for me only dipping to about 40 or so which makes sense. Why does nuts tank the performance? Is it because of the monster types and no walls to close off the visible areas? If the Cyberdemon horde was swapped for Zombiemen would the performance not tank?

nuts runs fine on PrBoom+ for my computer. But do not compare PrBoom+ with ZDoom. The logic between the two - while similar - are vastly different. I'm no programmer. But from my understanding, ZDoom has a lot more "checks" to go through for every actor. You know. Features that you all requested. It takes a toll on the engine when you decide to throw more monsters at the screen than you can shake a stick at.

Couldn't prboom handle it at a decent frame rate on high end hardware? I remember seeing a video of it.

I hope I live long enough to experience nuts.wad at full 60 FPS

If someone can magically increase max. clock frequency to 8 GHz it should be doable if renderer and game logic can also be separated onto different cores.
The single biggest performance issue here is still the sprite rendering, not the monster AI.

I mean people never thought nuts would ever be playable on vanilla, though I guess that's more due to the speed of a single core doubling every six months. Unless a theoretical port dedicated an entire core to monster AI, I'm guessing it will probably never happen without cpu speed becoming 50x faster than what's available today.