Proper multi CPU core support in TF2?

[Kreny]

Hi, does TF2 use CPU cores better than TF1 or it is the same - main computing done one first core and some side computing on second core so game become pretty slow on large maps woth decent number of vehicles.

[Dan]

You're describing literally *all* multi-core games.
Games are not a symmetric-parallel compute task. Games have a main thread, and delegate certain tasks to other cores, where the out-of-order execution of those tasks won't break the game logic (things like physics simulation)

What you're imagining isn't how games work... ever.

[Mackintosh]

I doubt they had the time to improve the engine that much since the multi (well, dual) core patch was released for TPF. In fact, I suspect that TPF patch was released at the same time they forked the code for TPF2. We'll find out soon enough. I'm not holding my breath.

[CunkFeatures]

I feel its going to be same old.

It could maybe be improved, much like the X games managed to get their sim to go from a 4 core hell fest to a 12 core 80% utilization improving game performance greatly. Or even CiM to CS.

End game though will always be the same. Think of that as opening program after program on your computer, at some point it will stop.

[Mackintosh]

Pretty sure most of the performance improvement in CS vs CiM came from their "fudging numbers". They only simulate a limited number of agents, and the rest is guesstimation. TPF is almost exactly like CiM2 and Simcity (the EA one) in its agent simulation. Even multicore wouldn't save it, though it would certainly mitigate the eventual meltdown, perhaps even prevent issues on smaller maps.

[CunkFeatures]

Originally posted by Mackintosh:

Pretty sure most of the performance improvement in CS vs CiM came from their "fudging numbers". They only simulate a limited number of agents, and the rest is guesstimation. TPF is almost exactly like CiM2 and Simcity (the EA one) in its agent simulation. Even multicore wouldn't save it, though it would certainly mitigate the eventual meltdown, perhaps even prevent issues on smaller maps.

It did feel quite fake, but without source codes we will never know for sure.

I'm sure the same can be done for people and traffic in TpF if not done already. But as we all know population rises and CPU demand rises quite dramatically.

Yeah we need to keep reminding people, end game will always drop. No matter what.
But with a map editor, we can make maps huge with less going on for later game if need be. But I never go bellow 20fps anyway.

I will be checking out the game on my home server once out to see how it handles 24 threads. If not already tested by someone. But looking at Urbans video, I doubt much has changed as that looked laggy once he had a few things going on.

[2025 Subaru Crosstrek]

Why we all talking about Team Fortress 2??

lol joking, but yeah honestly they could do better in many areas, but personally I think they need to improve in the FPS department most of all.

[Dan]

The problem is, unlike most other games of this type, the entire economy and underlying mechanics of Transport Fever is fully deterministic based on the movements of each and every agent in the game. As it is currently designed, they cannot afford for a single result to come in out-of-sequence, or it would break the simulation... so even for things that are offloaded to other threads, the main thread presumably has to lock and await results before progressing to the next timestep.
The only way to make significant performance improvements would likely be to abandon some of the simulation complexity, and make agents more of a cosmetic layer, as in Cities Skylines, rather than being directly tied to the simulation.

In short... Transport Fever has always been ambitious and focused singularly on emergent simulation rather than faking it for visual appeal... and with that comes performance costs.
They've done an amazing job already considering what the game actually achieves.

[Mackintosh]

Agreed, while I'm no programmer, I suspect you're right with respect to multi-threading. I just hope they managed to offload the graphical bling somehow so as not to impact the simulation calculations, otherwise we're up the proverbial creek.

As things stand, agent-based simulations are inherently performance-sinks. Unfortunately the only way to improve performance in the end-game is to put a cap on agents, and to play on smaller maps. At the end of the day, the more powerful your CPU, the better your chances at staying off the inevitable. Unfortunately, the inevitable will get you in the end. That's what happened in CiM, CiM2 and Simcity (EA). It's what happens in all agent-based simulations that allow for unrestricted agent spawn. Can't have the cake and eat it too.

[Dan]

Originally posted by Mackintosh:

Agreed, while I'm no programmer, I suspect you're right with respect to multi-threading. I just hope they managed to offload the graphical bling somehow so as not to impact the simulation calculations, otherwise we're up the proverbial creek.

As things stand, agent-based simulations are inherently performance-sinks. Unfortunately the only way to improve performance in the end-game is to put a cap on agents, and to play on smaller maps. At the end of the day, the more powerful your CPU, the better your chances at staying off the inevitable. Unfortunately, the inevitable will get you in the end. That's what happened in CiM, CiM2 and Simcity (EA). It's what happens in all agent-based simulations that allow for unrestricted agent spawn. Can't have the cake and eat it too.

I should think offloading graphical bling was the first thing they did. I mean, most is implicitly offloaded anyway - to the GPU.
The performance is likely entirely bottlenecked by the underlying simulation at this point.

[Kreny]

Main reason for me - when it comes to buymg the game - is performance improvement in compare with TF1. I understand it can´t use all the cores efficienly so I would sacrifice some agents, eg have a cap for them per each city and then just multiply them by city size.

Honesty, if you have 100 and 1000 agents, they will distribute almost identicaly (statistics) so why to compute 1000 if 100 is good enough and rest can be just multilicated.

[CunkFeatures]

It could be simplified, but then everyone will be up in arms because they cant follow someone all the way home.
Id prefer a game about transport than sims (agents).

[Heinekn]

All point to a good reason to prefer an i-9 system. Whether particular software is specifically "built" with the capabilities of using "unused" cores or not...with an i-9 processor...you have one of your cores specifically tasked with "spreading around" some of the work load.

With my i-9 system configuration, it's a rare piece of software that makes my system hiccup.

But as Dirty Harry once said..."A man's got to know his limitations". smiles.

[Thoddy]

They told us about that question at the Berlin event: Yes. They have worked on it.

[Bruno]

You guys got me worried. What kind of performance can I expect on i5 6600k, gtx 1070, 16GB ram and ssd? On cities skylines, when population in my last town increased to 12K, my fps rate would go below 30 fps, and I didn't even use any mods. I play on 1080p.

[Mackintosh]

Depends on how large your population and cargo grow. Since everything is an agent (cargo too), as the number of instances grow, so does the demand for CPU calculations. You can completely ignore your GPU, as long as it meets minimum specs. This game is all about raw CPU horsepower. On my 9900K the game chocked at a population of around 100000, with first issues noticeable at around 75000. Since the population grows almost exponentially, performance decreases inversely. That said - 100000 is a very large population on a large map with many cities. Conversely, by capping overall population to 30000-40000 you could play the game for many hours without running into major issues.

Now, while all this might hold true for TPF1 and TF, we're just speculating here. That's why it's imperative for developers and streamers to show us late-game gameplay on an average CPU. I am slightly miffed that they have not, and it's what's fueling this whole discussion.