UPS optimization - say Substations once more...

[Belter]

Subsations vs Poles vs UPS optimization

- I was interested what is the real impact of different electricity wiring setups
- Using 5x100 lanes of Iron Ore -> Iron Plate setups, 3 x 50.000 ticks for benchmarking

Wires-anim-.gif (13.5 MiB) Viewed 4862 times

Map-Wireconfigs-.png (36.4 KiB) Viewed 4862 times

Results - TL;DR: just use poles

- Unless you create a mess poles are OK. Substations often won't fit anyway to optimized layouts
- There might some errors in my benchmark (the two slowest), but I don't expect any major finding here
- We're close to the margin of error
- Saves attached

UPS.png (85.47 KiB) Viewed 4862 times

#1 Substations Min

- High hopes, optimized # of substaitons per row to 2

#2 Substations Norm

- 4 substations per row, modular BP

#3 Poles Min

- Removed some cables from the normal wires setup (#4) to see the difference
- OK this was a pain as region cloner won't keep this
- Results might be wrong

#4 Poles Norm

- Normal pole setup - optimized of course to minimal number of poles - 18/row
- Modular BP

#5 Poles Max

- Added extra wires to #4

#6 Poles Mess

- Added poles to all empty spaces just for testing

[coppercoil]

First thought: wtf???
Second thought: lets make the graph correct.

Offset_0

corrected.png (18.55 KiB) Viewed 4849 times

[Belter]

First thought: wtf???
Second thought: lets make the graph correct.

First thought: You're right.
Second thought: Do you see the 1% mark with yellow on that graph?

[coppercoil]

No, I missed that. I did not found yellow color on the graph and skipped the following text . You know, browsing nowadays is about skipping 95% unnecessary information, this works on a driver layer

Nevertheless... 1% is still unexpected.

[disentius]

Dave's version is still faster:)
The difference between min-max in my previous test bothered me, so
- I shut down all other programs and windows on my PC.
- left it alone during the test.
it seems to prevent "outliers"

2022-02-15 17_24_51-bm_ smelting results.csv - LibreOffice Calc.png (77.94 KiB) Viewed 4706 times

[DaveMcW]

It's possible you both are correct.

After disentius made the map 13 times smaller and shut down everything else, he was able to fit all the active entities in CPU cache and never touch RAM. So it is not affected by storing extra entities in RAM.

Belter was forced to use RAM access, and all the extra entities slowed it down.

I wonder if we could optimize the RAM layout, by building the poles after everything else? This is not really practical in a normal game.

[disentius]

I made all belter's saves from the last post 100 belts, and ran them all 10 times together with davemcw's with substations, and belters V6.75
No activity on PC while testing
Results: Dave still wins (with my modifications)

[disentius]

2022-02-15 21_17_10-bm_ smelting results.csv - LibreOffice Calc.png (99 KiB) Viewed 4643 times

[Belter]

TY for testing!

It is pretty interesting to have different results, even if the diff is low. For now we see diff output for 100 and 500 lanes. Havig the savegames is crucial and TY for attaching.

On the weekend I'll do a scalability testing w/100/200/300/400/500 lanes to understand the difference. Any chance to run 500 lanes test justs to see it somewhere else as well? I'll do a 100 lane test as well. I've chosen 500 lanes as on my old computer it runs around 130FPS w/UI, so its "playable" and minimizes the "fixed" UPS cost of the game I assume. Better BPs should perform better whatever lanes we're testig with though. It is faster to test w/100 lanes, yes.

Also why substations helps for DaveMcW's BP, but same or worse for me?

I really appreciate the effort and interest here and the smelting thread - hope we won't end up trying to prove which BP is faster. Wait, I've just said I'll do more tests.. for science

[disentius]

Welcome!
yes, the why of it all is interesting. I am not sciency/knowledgeable enough about the technical side of factorio/computers to make a good hypothesis.
(That is why I also included the original logs) I hope we get a true nerd to help...

What I can do, is consistent testing. I ll see If I can run another one today. Do you have the original blueprint of the versions you want tested?
And a feature request: keep the timers apart, I ant to use 1 timer per map of 100/300/500 furnaces)
Also the reported UPS is a derived value made in the script I use, so i tend to use the execution time, min/max/avg as reported by the benchmark command as leading

[disentius]

There is something with the "max" values. the ones from your builds are twice as high as the ones from davemcW's
Maybe you can plot avg/max/min over time for both saves?

[Belter]

And a feature request: keep the timers apart, I ant to use 1 timer per map of 100/300/500 furnaces)

OK, let's agree on this.

Also the reported UPS is a derived value made in the script I use, so i tend to use the execution time, min/max/avg as reported by the benchmark command as leading

The scirpt also uses execution time (1000 * $ticks / $executionTime), which is calculated form the avg ms numbers.
Honestly its simpler to use this UPS for graphs as it is there.

[Belter]

Lets continue at viewtopic.php?p=561883#p561883 where we came from..