Never let your factory run without failcheck or supervised

[Sneaky_Bugger]

So my Factorio runs on a server and therefore 24/7. I check in a couple times per day, optimise things and see if some bottlenecks were created or sth went the wrong line.
I thought, I am at the point where I just need a more constant flow of promethium even though, it´s just needed for one tech. There is still room to optimise but I haven´t had any big hickups for a long time. I still need some Spidertron squads on each planet and just finished the transition from mixed robots to all legendary robots on all planets. But I made a major mistake, I went to bed without turning robo production off. Not sure why I haven´t had a failsafe aka production control in place but somehow it happened and when I came back today I got greeted with the info, that I don´t have enough robo ports on Vulcanus to charge my bots. Wait, what?! This can´t be, I have plenty of ports to charge bots. In fact, I have way more than I have robots.
Turns out, to charge 344k bots, you need a bit more roboports than I have. Now I have to recycle like 315k bots because smelting in lava seems cruel.

edit: Forgot screenshot

https://steamcommunity.com/sharedfiles/filedetails/?id=3434090088

[PunCrathod]

Originally posted by Sneaky_Bugger:

So my Factorio runs on a server and therefore 24/7.

That is an incorrect assumption. Unless you went out of your way and specifically changed a config file the server will pause everything when nobody is on the server.

But yeah. That is one of the reasons why I always hook up the inserter inserting new bots and set it to enable when available bots is less than 10. A nice easy automated way to get just enough bots to run your factory.

[Sneaky_Bugger]

Originally posted by PunCrathod:

Originally posted by Sneaky_Bugger:

So my Factorio runs on a server and therefore 24/7.

That is an incorrect assumption. Unless you went out of your way and specifically changed a config file the server will pause everything when nobody is on the server.

But yeah. That is one of the reasons why I always hook up the inserter inserting new bots and set it to enable when available bots is less than 10. A nice easy automated way to get just enough bots to run your factory.

Oh, it´s just a server where Factorio runs on, not a MP server. I don´t want to run my PC 24/7 when I have tech that runs 24/7 for various reasons. :)

[PunCrathod]

Oh you meant that kind of server. I guess I made the incorrect assumption that you were a normal person(no offence). Though running the game on a server is certainly one way to play the game. I would still have just hosted a private command line only multiplayer server instead. And if the server running while you are not present was a desired feature then just change the config. And then just joined that with the pc I'm sitting at.

Unless that pc was somehow much slower at running Factorio than the server. Usually servers are specced for a different kind of workload so they are not going to be as fast as a gaming pc would be with Factorio.

[Sneaky_Bugger]

Originally posted by PunCrathod:

Oh you meant that kind of server. I guess I made the incorrect assumption that you were a normal person(no offence). Though running the game on a server is certainly one way to play the game. I would still have just hosted a private command line only multiplayer server instead. And if the server running while you are not present was a desired feature then just change the config. And then just joined that with the pc I'm sitting at.

Unless that pc was somehow much slower at running Factorio than the server. Usually servers are specced for a different kind of workload so they are not going to be as fast as a gaming pc would be with Factorio.

Yeh, Factorio is bottlenecked by CPU not by GPU, so in my case it´s the perfect solution. Even with the 340k Bots I had not UPS problems at all. Game runs smooth as it can. :)
And no offense taken, I am not a normal person for sure. And running games on "server hardware" is nothing new to me. All the old RPGs like the CiV series or Anno were hosted on my servers, just in that case as MP servers.

[PunCrathod]

Originally posted by Sneaky_Bugger:

Originally posted by PunCrathod:

Oh you meant that kind of server. I guess I made the incorrect assumption that you were a normal person(no offence). Though running the game on a server is certainly one way to play the game. I would still have just hosted a private command line only multiplayer server instead. And if the server running while you are not present was a desired feature then just change the config. And then just joined that with the pc I'm sitting at.

Unless that pc was somehow much slower at running Factorio than the server. Usually servers are specced for a different kind of workload so they are not going to be as fast as a gaming pc would be with Factorio.

Yeh, Factorio is bottlenecked by CPU not by GPU, so in my case it´s the perfect solution. Even with the 340k Bots I had not UPS problems at all. Game runs smooth as it can. :)
And no offense taken, I am not a normal person for sure. And running games on "server hardware" is nothing new to me. All the old RPGs like the CiV series or Anno were hosted on my servers, just in that case as MP servers.

Yeah factorio is most of the time bottlenecked by cpu and memory bandwidth/latency. But server cpus usually sacrifice single core performance in favor of more cores compared to gaming pcs, and for factorio(and games in general) single core performance is a huge factor. That being said if you have more performance available than the game needs then it doesn't really matter.

Are you streaming the game over the network or do you have a monitor, keyboard and mouse directly hooked up to the server? I'm asking just out of curiosity. I regularly use steamlink to stream my games etc from one pc to another. Sometimes from a pc that does not even have a monitor(I have a linux machine as a daily driver and a backup monitorless windows machine for games that do not run on linux). Sometimes I even stream to another room with a big tv when playing with family/quests. And while in most games the latency and video compression artifacts do not bother me in some games it is just awful.

[kremlin]

Originally posted by PunCrathod:

Yeah factorio is most of the time bottlenecked by cpu and memory bandwidth/latency. But server cpus usually sacrifice single core performance in favor of more cores compared to gaming pcs, and for factorio(and games in general) single core performance is a huge factor. That being said if you have more performance available than the game needs then it doesn't really matter.

If the server is using a server-grade CPU it's not just that they prioritize SMP performance, all of the caches on the cores are usually orders of magnitude larger and faster than a normal desktop CPU. Factorio should get a bit of a bump from the faster memory architecture even when the single core speed isn't as high.

[PunCrathod]

Originally posted by kremlin:

Originally posted by PunCrathod:

Yeah factorio is most of the time bottlenecked by cpu and memory bandwidth/latency. But server cpus usually sacrifice single core performance in favor of more cores compared to gaming pcs, and for factorio(and games in general) single core performance is a huge factor. That being said if you have more performance available than the game needs then it doesn't really matter.

If the server is using a server-grade CPU it's not just that they prioritize SMP performance, all of the caches on the cores are usually orders of magnitude larger and faster than a normal desktop CPU. Factorio should get a bit of a bump from the faster memory architecture even when the single core speed isn't as high.

A quick google on intel xeons and i7s says that server cpus have about 3 times as much L3 cache but as far as I could see the L1 and L2 cahces were the same size for both. The cache is the same type and speed on both. And the i7 supports faster ram. And on AMD side it is a similar story of only 5 times as much L3 and the lower level caches are the same on both server and desktop etc. So no "orders of magnitude" larger or faster.

Also here is a link(you have to copypaste the url as the steam content filter fails with urls that have &) to factorio benchmarks which does include server class cpus and they come nowhere near the gaming cpus performance.

https://factoriobox.1au.us/results/cpus?map=4c5f65003d84370f16d6950f639be1d6f92984f24c0240de6335d3e161705504&vl=2.0.7&vh=

You can say what you want but the data is clearly saying server cpus have significantly less gaming performance than desktop cpus.

[Sneaky_Bugger]

I dont see the 5th gen Xeon CPUs on that benchmark list, but maybe I am blind. It´s not my own server, to clarify things. It´s a server a friend administers and I am "stealing" a bit of it´s performance. And that a CAD server is slower or than any desktop CPU is just wrong. :) It´s not that we cobbled some home server hardware together. It´s one of the benefits of working in IT and having some friends in the right companies. And while the L1 cache is not that much bigger, the whole thing is way faster in my book. I´ll see if I can run the benchmark of that thing. 64KB vs 80KB doesn´t sound that much it runs smooth no matter what. On paper my Ryzen 7 9800x3d has the same latency as the Xeon cpu and maybe I should test it on my desktop but than I have to run my pc 24/7 which would be stupid. :) It´s easer to ssh into the server, get my stuff done and log out again while everything is running without me.

[PunCrathod]

Originally posted by Sneaky_Bugger:

I dont see the 5th gen Xeon CPUs on that benchmark list, but maybe I am blind. It´s not my own server, to clarify things. It´s a server a friend administers and I am "stealing" a bit of it´s performance. And that a CAD server is slower or than any desktop CPU is just wrong. :) It´s not that we cobbled some home server hardware together. It´s one of the benefits of working in IT and having some friends in the right companies. And while the L1 cache is not that much bigger, the whole thing is way faster in my book. I´ll see if I can run the benchmark of that thing. 64KB vs 80KB doesn´t sound that much it runs smooth no matter what. On paper my Ryzen 7 9800x3d has the same latency as the Xeon cpu and maybe I should test it on my desktop but than I have to run my pc 24/7 which would be stupid. :) It´s easer to ssh into the server, get my stuff done and log out again while everything is running without me.

Oh I don't doubt that the server is faster for the kinds of workloads the server is designed for. Just that not all workloads work the same way and being fast at one thing does not mean its fast at everything.

Edit: And also the ability to run the game 24/7 without having to keep your desktop on 24/7 might make using the server worth it even if it might mean you will hit the performance ceiling a little sooner.

And also yes do run the benchmark if you can. More data is always welcome and it will be interesting to see if a newer xeon closes the gap or even surpasses the desktop cpus. You never know until you actually try and measure it.

[kremlin]

Originally posted by PunCrathod:

https://factoriobox.1au.us/results/cpus?map=4c5f65003d84370f16d6950f639be1d6f92984f24c0240de6335d3e161705504&vl=2.0.7&vh=

You can say what you want but the data is clearly saying server cpus have significantly less gaming performance than desktop cpus.

Intel Xeon w7-2595X 366.2 367, 365, 365
13th Gen Intel Core i9-13900KF 365.6 388, 367, 366, 364, 363, 317, 304, 234, 224, 216

They top end benchmarks are most likely overlocked. The workstation class CPU has more and faster cache than the desktop class CPU and is clearly keeping pace with the desktop CPU. And there are drastically fewer results for them because they're usually not worth the premium if you want a highly overclockable desktop CPU versus a workstation CPU.

Intel Ark doesn't even list the particular spec I'm referring to with cache speed. You usually have to run something like memtest to see those numbers at all. That's how I learned about it, by running the tests and seeing the massive difference.

[PunCrathod]

Originally posted by kremlin:

The workstation class CPU has more and faster cache than the desktop class CPU

More? Sure. Faster? Might be but a quick google search didn't really give any numbers so can't be sure. Orders of magnitude more? evidently not. 100 is not orders of magnitude more than 33. It's only 3 times as much. Orders of magnitude means over 10 times more and they are not there yet. Orders of magnitude faster? I highly doubt it. If intel or amd had that kind of technology they would be using it on high end desktop cpus too.

Factorio is still mostly singlethreaded even tough it has got more multithreaded action nowdays.

Find me a single threaded benchmark that isn't dominated by desktop cpus. Because the top google result passmark has server cpus at the top in mulithreaded results but in singlethreaded results server cpus are way down the list. https://www.cpubenchmark.net/singleThread.html

And just in case it is not clear. I'm not asking this as an attempt to dismiss you but rather I want to make sure I'm not misinformed. You seem to know at least something I don't as I have not had the privilige to run benchmarks on server grade cpus, but I make a point about not trusting computer related performance claims without seeing at least some data to back it up.

[Sneaky_Bugger]

Pretty interesting topic, I have to admit, took a complete turn from the orignal topic (welcomed). And I totally agree, if you run a SQL server and slap Factorio on it, it will - most likely - be slower or don´t even run because of GPU limitations. In my case, it´s a server to run CAD programs on it. So there is no bottleneck. At some point - as stated before - Factorio can´t utilise the raw power of high performance server tech. LIke we have 64 cores but it makes zero difference to have 64 or 2 for Factorio. Some stuff in Factorio utilises multi threading but it´s neglectable, as far as I am informed. To me the selling points are, I don´t have to pay dime and I can run Factorio 24/7 on it. For "normal" playthroughs I´d simply use my desktop pc.

[PunCrathod]

Originally posted by Sneaky_Bugger:

I don´t have to pay dime and I can run Factorio 24/7 on it.

That deal is very hard to beat. To be honest If I had a save I wanted to be running 24/7 I too would be using a server even if my desktop could run the game twice as fast.

But I'm still curious on how a server meant for CAD fares at running factorio. Is there any possibility you could do and submit a couple runs with the factoriobox benchmark and link the results? I don't think the difference between a desktop and a CAD server would be as extreme as 2x performance. My guess would be somewhere around 5-10% or at most 25%.

[Sneaky_Bugger]

Originally posted by PunCrathod:

Originally posted by Sneaky_Bugger:

I don´t have to pay dime and I can run Factorio 24/7 on it.

That deal is very hard to beat. To be honest If I had a save I wanted to be running 24/7 I too would be using a server even if my desktop could run the game twice as fast.

But I'm still curious on how a server meant for CAD fares at running factorio. Is there any possibility you could do and submit a couple runs with the factoriobox benchmark and link the results? I don't think the difference between a desktop and a CAD server would be as extreme as 2x performance. My guess would be somewhere around 5-10% or at most 25%.

Will ask when the next maintenance window comes around but I am pretty certain it will be unimpressive. 5-10% would be my guess as well, actually more like 5-7%. When I load my save on my desktop PC I don´t feel any difference.