Factorio Roadmap for 1.0.

[MeduSalem]

However the more they optimise the game the bigger bases will get, and while an i5 is an excellent balanced processor you really need an i7 for raw power.

As long as clock speed is same, going with I7 over I5 makes no perceiveable difference for Factorio as extra "virtual" cores is all i7 has over i5.

Factorio has one thread that is slowest and every other thread is bottlenecked by it so adding more cores won't help.

I'm already looking forward to how AMD's Zen will perform with Factorio as I'm probably getting one of them for my next rig since Intel's price policy is turning me off already. Let's see how that one will deal with single-threaded performance. (I know they are going to brand it as Ryzen, but I already hate that name so I'll be referring to it as Zen throughout its lifetime)

At least I haven't gotten myself a new computer in 7 years. I'm still running a first generation Nehalem Core i7 860, which in my opinion was the last true micro architecture enhancement that didn't solely rely on SIMD extension brutality to achieve more hypothetical throughput.

Somehow always when AMD is heading for a comeback Intel somehow lost its edge due to lack of serious competition. I really looked forward to Skylake already but then it proved only to be on par with Haswell/Broadwell at best... so not really worth it.



That said... didn't the devs already write that they are trying to have somewhat better multithreading support with one of the upcoming updates? At least I remember vaguely that there were some words about it in one of the FFFs.

[hoho]

I'm already looking forward to how AMD's Zen will perform with Factorio

I'd love it if AMD got their act together and become a serious competition again. Sadly, after K8 their promises haven't really been all that truthful. I hope with Zen it's different.

At least I haven't gotten myself a new computer in 7 years.

I upgraded this spring from a CPU that was released in the beginning of 2007, q6600 to i5 6600
CPU per-core performance has changed quite little in past decade.

That said... didn't the devs already write that they are trying to have somewhat better multithreading support with one of the upcoming updates?

Yes, they have. Still, as long as their main thread gets to have one core all to itself and it's uses 100% of that core, adding extra cores for the other threads will not help much. Only way adding cores could help is if those other threads combined take a LOT more processing power than the main thread. I'm willing to bet that this is not the case with Factorio.

[FunMaker]

What about programmable assemblers - will they be included in 0.15? That would be some extreme fun with combinators - just imagine building a rocket just by one assembler (other assembling machines excepted of cause).

[theRustyKnife]

What about programmable assemblers - will they be included in 0.15? That would be some extreme fun with combinators - just imagine building a rocket just by one assembler (other assembling machines excepted of cause).

Check out Crafting Combinator. Does pretty much what you said

[MeduSalem]

I'd love it if AMD got their act together and become a serious competition again. Sadly, after K8 their promises haven't really been all that truthful. I hope with Zen it's different.

Yeah, Bulldozer wasn't really that much of a success. I don't blame it on the architecture idea itself but rather on its lousy implementation... because IBM will be doing something similar with one of their upcoming Power CPUs and there it seems to turn out rather well. So the idea of having cores sharing certain resources would basically work, if implemented well.

CPU per-core performance has changed quite little in past decade.

If one neglecting SIMD computation power that is... because that stuff is scaling up in 2^n rate... soon there will be AVX-512 and I bet they might even go for AVX-1024 and more ridiculous stuff.

Yes, they have. Still, as long as their main thread gets to have one core all to itself and it's uses 100% of that core, adding extra cores for the other threads will not help much. Only way adding cores could help is if those other threads combined take a LOT more processing power than the main thread. I'm willing to bet that this is not the case with Factorio.

If the main thread is the problem then the more they are able to outsource/seperate routines from the main thread to other threads the better. But I don't know how many optimization possibilities there are for that because I guess there might be quite a lot of stuff in the main thread that can't be done in parallel, that have to be done exactly in that order or things wouldn't work at all.

I mean yeah hypothetically a lot of things might work in parallel but since they have to stay deterministic for the sake of multiplayer they can't be done in an asynchronous fashion... so the multithreading capabilities are capped at how much you can parallelize without losing a deterministic state.

[FunMaker]

What about programmable assemblers - will they be included in 0.15? That would be some extreme fun with combinators - just imagine building a rocket just by one assembler (other assembling machines excepted of cause).

Check out Crafting Combinator. Does pretty much what you said

Thanks, i searched for such a mod but did not find it - nice! But i hope the devs will include something similar in trunk - well - they are investigating

[hoho]

IBM will be doing something similar with one of their upcoming Power CPUs and there it seems to turn out rather well.

Yes, Bulldozer wasn't bad because of shared resources. It was bad for having slow cache, high memory latency and low throughput of basic integer calculations.

Power being good in the area its used doesn't show much. Itanium is (was?) also great in what it did while it was horrible in running daily stuff. Similarly Bulldozer was pretty darn awesome once you crammed 8 of them to single motherboard. Intel topped out at 4.

Bulldozer simply was designed in a way that vast majority of "normal" desktop applications/games didn't benefit from.

If one neglecting SIMD computation power that is... because that stuff is scaling up in 2^n rate... soon there will be AVX-512 and I bet they might even go for AVX-1024 and more ridiculous stuff.

I know. SIMD is nice but it's rather limited in practice. I'd be quite surprised if Factorio uses it in any significant way, if at all. It's awesome for SpecFP/Int benchmarks and other synthetic benchmarks for sure

If the main thread is the problem then the more they are able to outsource/seperate routines from the main thread to other threads the better.

Of course but in modern day when few people have less than 4 cores, it's quite unlikely that they can get enough stuff out of that main thread to fully load the rest. If they can't do that then going above 4 cores won't really speed up the game.

multithreading capabilities are capped at how much you can parallelize without losing a deterministic state.

https://en.wikipedia.org/wiki/Amdahl's_law

I would imagine that in Factorio, the limiting factor is the part of code that synchronizes-merges the results from all other threads in order to "finalize" a tick. Stuff like that can't really be parallelized.

[SQLek]

Sooo Factorio isn't multi threaded already

CPU usage captured in nest clearing situation.

Cores occupied pretty evenly

factorio_cpu.png (232.67 KiB) Viewed 11192 times

Tested on FX-6350 Linux Debian 4.7.6-1

[daniel34]

Sooo Factorio isn't multi threaded already

See https://www.factorio.com/blog/post/fff-151, section Multithreaded update

[hoho]

Sooo Factorio isn't multi threaded already

It is multithreaded but for whatever reason, the main thread keeps jumping around your cores. If it was kept on a single core, you'd see that one core under 100% load and others much lower.

Image like that will give a false impression as if Factorio doesn't use CPU properly. The CPU core load is usually integrated over a period of time, say 1s. If within that 1s the main thread jumps around, say, 3 cores, it'll mean each core will get ~33% load while in reality, the thread is CPU bottlenecked.

[Loewchen]

Sooo Factorio isn't multi threaded already

CPU usage captured in nest clearing situation.

factorio_cpu.png

Tested on FX-6350 Linux Debian 4.7.6-1

This is just load balancing.

[SQLek]

Sooo Factorio isn't multi threaded already

It is multithreaded but for whatever reason, the main thread keeps jumping around your cores. If it was kept on a single core, you'd see that one core under 100% load and others much lower.

I got false impression that Factorio one thread gets distributed to multiple cores for some reason. Other CPU hungry, single threaded apps occupy single core toped at 100%

Image like that will give a false impression as if Factorio doesn't use CPU properly. The CPU core load is usually integrated over a period of time, say 1s. If within that 1s the main thread jumps around, say, 3 cores, it'll mean each core will get ~33% load while in reality, the thread is CPU bottlenecked.

As long as i have FPS/TPS 60/60 i'm not complaining. Just curious what tricks are happening under the hood to show up as a core-jumping-thread. Thanks for another idea for investigate

[hoho]

I got false impression that Factorio one thread gets distributed to multiple cores for some reason.

Technically, that's kind of what is happening but different cores aren't running that thread simultaneously, the thread only ever executes on one core. When one core is done with it, another one picks it up again.

Other CPU hungry, single threaded apps occupy single core toped at 100%

If I had to guess, I'd say it's because factorio main thread occasionally yields to wait for other threads to catch up in order to not just have an busyloop. OS takes that as a hint that it can use that core for some other tasks and the main thread gets offloaded.

As long as i have FPS/TPS 60/60 i'm not complaining.

And you shouldn't
If the game is at 60/60, I think not even the main thread can take up all of a single core. When I'm talking about the game being limited by that single thread, I'm talking about situations where TPS can't be kept at 60 any more.

[Sparkysam]

As long as I have FPS/TPS 60/60 I'm not complaining.

And you shouldn't
If the game is at 60/60, I think not even the main thread can take up all of a single core. When I'm talking about the game being limited by that single thread, I'm talking about situations where TPS can't be kept at 60 anymore.[/quote]

I wish I could get that with my bases but with bobs, it becomes impossible late game, I get more jumps than a Mexican jumping bean.

[Mendel]

Have you tried using train wagons as warehouses? Just put a few trackpieces to where you need a buffer, put wagon on top, add stack inserters and presto, you have a warehouse wagon!

You're right that gameplay should decide what happens with this idea, but I disagree with the implication that there are not gameplay reasons to nerf chest stack counts and have larger warehouse type building(s). It would also incidentally 'solve' the storage density 'imbalance' between barreled oil in chests and storage tanks as a side effect. The gameplay reason to want larger warehouse buildings is that you can put more inserters around them, which is handy for making high throughput buffers without having to fuss with combinator magic to make sure it loads them in a balanced fashion (as is currently needed for an 'ideal' design). A single 3x3 warehouse could work as a buffer for a line with up to 6 stack inserters worth of item throughput with no combinators required, and you have the added simplicity of everything being in a single container (which makes it easy to move item stock in and out of the system via that single warehouse rather than messing with multiple chests and likely futzing up the balancing between them, which will lower system throughput until the imbalance is corrected).

[hoho]

I wish I could get that with my bases but with bobs, it becomes impossible late game, I get more jumps than a Mexican jumping bean.

Bob's mods don't really tax the game any more than regular vanilla game. All it does is to force you to make a bigger base with more stuff in it. It's the size of the base that slows you down, not bob's mods

Though there are some other mods that do slow the game down, some quite substantially. You might be able to see the bigger slowdown-creators in the detailed debug menu (F5).

[generral87]

hello
my english is not good course i am german so i try to make it understandable

the nuke power plant is very good also the steam turbine and i can´t wait for using them
but i am thinking of other alternative (green) energy sources like:
solar tower powerplant
geothermal energy

and better (bigger) akkumulators

for the endgame because the need for large spaces by the old ones

thanks for reading

[Xenomorph]

Update (26. August 2016):
Factorio 0.15
Planned:

• Nuclear power
• Liquid wagon + universal barrelling

How you get those features in the game? There're some mods which I'm using to have this features allready. Did you implement only mods (maybe with some adjustments) like

https://mods.factorio.com/mods/Choumiko/RailTanker for the liquid wagon
https://mods.factorio.com/mods/Simdezimon/Nucular for the nuclear power
https://mods.factorio.com/mods/GotLag/Omnibarrels for the universal barreling

or did you made an own solution?

[daniel34]

How you get those features in the game? There're some mods which I'm using to have this features allready. Did you implement only mods (maybe with some adjustments) like [...] or did you made an own solution?

Both will be implemented using their own solution.
For details on the implementation of nuclear power see here:

• Friday Facts #164 - Nuclear power
• Friday Facts #167 - Reactors Operational

[r0nalxd]

Factorio 0.15
Planned:

• Nuclear power
• Liquid wagon + universal barrelling

Factorio 0.16 (This is planned to become 1.0 when stabilised)
Planned:

• Dirty mining. (Way to get more resources from mining posts at a cost of additional industry, logistics requirements and investment)

Looking forward to these new features. Got some questions to the devs/community:

• Regarding universal barreling, will factories accept regular liquid pipelines and the new barreled liquids?
• Assuming dirty mining is mining for unidentified ore, will there be installations comparable to gold mining industry to clean and separate ores? Perhaps also waste that needs to be processed?