CPU performance benchmarks

[jape3]

From your "mybase.txt" results we first take the average of the "avg" results. (Performed 1000 updates in 31587.002 ms
avg: 31.587 ms, min: 19.662 ms, max: 1158.872 ms). That comes in at 34.05ms. To get ups from update time per tick we divide 1 / 0.03405 = about 29 UPS.

[azesmbog]

Thank. I get it.
Of the five measurements, we take the average avg and calculate.
With the test base it turns out roughly 1 / 0.012 = 83.33 UPS

[lexx]

If people are intersted and have some heavy up-to date save I can run some tests on Ryzen 3700X with 3200 ram.

i am interested

i am still in between getting the 3800x or the 9900k or ks with 3600 ram (ideally get latency down to 16cl from 18cl) from what i have looked at the 9900k is the better option but as i will be playing with other people my system is likely going to faster then there Sever or PC i am connected to, but the heat from the 9900 at 5ghz is a bit

probably be doing streaming as well but i leave that to the GPU to do that (2060 super with OBS using New Nvidia encoder is very good with no CPU use)

as i can't play with some people on my 1800x once it gets to mega base size unless they slow the game down,, just not sure about going the 3800x when the 9900 is definitely faster for games like KSP, factorio and oxygen not included at late game stages

[lexx]

once i get time i post my 1800x results

Also version to be used should it be version 17.59 as 17.60 and later recipes was changed witch will deactivate a lot of building in the test map used

(guess I test both see what it does)

[DustFireSky]

I wanted to try my 3900x with the newly arived 3600 RAM but I cant get the script. Any other way to test with uncapped UPS?

I changed the original Script. U can find it here. U can test the offical map or your own maps with it. Enjoy

[azesmbog]

I changed the original Script. U can find it here. U can test the offical map or your own maps with it. Enjoy

Sehr gut !!!
I tested the antique processor Core ™ 2 Quad Q8400 (socket 775), release date Q2'09 + 4MB of memory
At the top is a standard script, the results are fixed , below is a local script with your megabase.

script_ps.jpg (86.46 KiB) Viewed 8437 times

Pretty good for an 11 year old man
Thanks for the modified script. But I still had to change the path to the directory with the game to absolute:
$ MAP = "c: \ Factorio_0.17.79 \ saves \" + "HuntingValley_17.67_Save1.zip"

[DustFireSky]

But I still had to change the path to the directory with the game to absolute:
$ MAP = "c: \ Factorio_0.17.79 \ saves \" + "HuntingValley_17.67_Save1.zip"

I followed the standard. For 90% of the gamers is the path correct.

[azesmbog]

Some fresh results with
https://factoriobox.1au.us
The same processor, three different versions of the game:
Factorio 0.18.1 - 101 UPS
Factorio 0.18.0 - 101 UPS
Factorio 0.17.79 - 91 UPS
same motherboard but with old i5-6400 (not K)
Factorio 0.17.79 - 91 UPS
Oddly enough, but the result in version 0.17.79 is comparable between the old and the new processor. Sadness.
Well and still, the results on the screen, in the console, are slightly different from the same results, but on the site. It’s strange.

[BlueTemplar]

Then your bottleneck might be RAM, rather than CPU ?
https://www.cpubenchmark.net/compare/In ... 2578vs3335
50% faster single-threaded !
(Or maybe the cooling + turbo, since TDP is 50% higher too..?)

[lexx]

Some fresh results with
https://factoriobox.1au.us
The same processor, three different versions of the game:
Factorio 0.18.1 - 101 UPS
Factorio 0.18.0 - 101 UPS
Factorio 0.17.79 - 91 UPS
same motherboard but with old i5-6400 (not K)
Factorio 0.17.79 - 91 UPS
Oddly enough, but the result in version 0.17.79 is comparable between the old and the new processor. Sadness.
Well and still, the results on the screen, in the console, are slightly different from the same results, but on the site. It’s strange.

What cpu was you comparing (old and new cpu)

[azesmbog]

Then your bottleneck might be RAM, rather than CPU ?
https://www.cpubenchmark.net/compare/In ... 2578vs3335
50% faster single-threaded !
(Or maybe the cooling + turbo, since TDP is 50% higher too..?)

in my tests, the motherboard + memory is the same. Only the processors are different. Apparently the old one was well dispersed, since the result is no worse than the new one))
Or a new one - bad )
Memory - yes, here I have a bottleneck, only 2900-2930-3000 MHz. It can no longer, but it's a matter of time

What cpu was you comparing (old and new cpu)

I thought that from the latest tests it’s clear, but I can clarify:
Old cpu -
https://factoriobox.1au.us/result/f3f0f ... da0f3feae6
new cpu -
https://factoriobox.1au.us/result/31371 ... 0cf9130ff5
https://factoriobox.1au.us/result/66fed ... 3416b30b70
https://factoriobox.1au.us/result/330f2 ... 178b8acfec

[lexx]

Then your bottleneck might be RAM, rather than CPU ?
https://www.cpubenchmark.net/compare/In ... 2578vs3335
50% faster single-threaded !
(Or maybe the cooling + turbo, since TDP is 50% higher too..?)

in my tests, the motherboard + memory is the same. Only the processors are different. Apparently the old one was well dispersed, since the result is no worse than the new one))
Or a new one - bad )
Memory - yes, here I have a bottleneck, only 2900-2930-3000 MHz. It can no longer, but it's a matter of time

What cpu was you comparing (old and new cpu)

I thought that from the latest tests it’s clear, but I can clarify:
Old cpu -
https://factoriobox.1au.us/result/f3f0f ... da0f3feae6
new cpu -
https://factoriobox.1au.us/result/31371 ... 0cf9130ff5
https://factoriobox.1au.us/result/66fed ... 3416b30b70
https://factoriobox.1au.us/result/330f2 ... 178b8acfec

I7-9700k and i5-6400

Why are you using miss match ram, that sure put the ram controller into flex mode or just flat out single channel mode if flex is not been used (3x8gb 1x16gb) this can affect performance a bit

Use 2x8gb in channel A and 16gb in channel B on its own to keep dual channel mode, it should do full dual channel, as 2x8gb is dual rank and 1x16gb is dual rank (or 2x8gb)

Reason for above
8gb sticks are single rank so 2 of them in first 2 slots that is for channel A will make them dual rank for like 1 16gb stick (16gb are always duel rank so that goes in the other channel B on it's own)

Apart from ram setup you have, the 9700k is a jump from a 6400 you should be getting more ups unless its purely due to ram (but tears have shown that there is a difference between 6000 and 9000 series)

[lokki]

Some fresh results with
https://factoriobox.1au.us
The same processor, three different versions of the game:
Factorio 0.18.1 - 101 UPS
Factorio 0.18.0 - 101 UPS
Factorio 0.17.79 - 91 UPS
same motherboard but with old i5-6400 (not K)
Factorio 0.17.79 - 91 UPS
Oddly enough, but the result in version 0.17.79 is comparable between the old and the new processor. Sadness.
Well and still, the results on the screen, in the console, are slightly different from the same results, but on the site. It’s strange.

I'd test my 3600 (Stock CPU, 2*8gb 3000mhz 16-16-16 mem) - 17.79 75 UPS and 18.1 83 UPS.

CPU overclocking doesn't takes a lot of sense - OC to 4.2-4.225 didn't increase performance dramatically, 2 or 3 FPS or something like this.

IF overcloking make more sense, even with low-speed memory. I did test 1900IF and 1500(3000ddr) mem and have results a bit better that 1500IF, but near noise-level. Aida shows 80ns instead of 72ns on 1500IF, but real-world performance was about the same.

Mem-bus speed gives a lot of boost, 3000 16-16-16 has the same results, as 3800 26-26-26, and 1900if 4800mem 48-47-47 had the same 1.12-1.18s. BUT! - 3800 14-26-26 (all timings from SPD, not XMP, except Cas Latency 14) did 10.6s, about 94UPS, but my memory is a bit unstable in this setup. Ok, 3800cl15 should be not so bad - 91ups maybe? Nope, same 11.2s. So, that was a great bench-afternoon, but i'm still on the way. I want to test something like 3733cl14, it should be stable, or a bit of magic and 3800cl14 would be so wit another ProcODT or something like that. I have also an idea, that cl15 is odd, and with cl16 it would be better, but I need to test it.

[lexx]

Most ideal for ryzen is 1800 if ( infinity fabric) and 3600 lowest Cl timing possible (do make sure benchmark between changes as stable does not mean faster on ryzen)

looking at getting 2x16gb G.Skill Neo 16cl kit (might run at Cl 14, also might do 3800 cl14 as well) and a 3800x to try and get better chance at getting a cpu that can do infinity fabric at 1900 so I can tinker

if your cpu can do is 1900 if and 3800 ram speed cl16 or cl14 if possible (3600 cl14 will likely run same benchmark result as 3800 cl16)

Anything higher then 3600(1800 if)/3800 (1900 if) will run the infinity fabric out of sync and result in negative/same performance (if will go into higher 90ns if out of sync on high speed ram kits)

[azesmbog]

nevertheless, the script works a little strange:

bench184.jpg (80.83 KiB) Viewed 7886 times

What is on my screen and what is on the site. I’ve noticed this before.
Although the result is comparable to my previous result, still this version is a bit slower (subjectively . On my map, on average, ~2 ups / fps results are lower
Upd:
New version 0.18.6
slightly increased the frequency of the processor.
https://factoriobox.1au.us/result/f7b9c ... b6f3160b04
compared to the previous result:
104 ups vs 101 on the site
95 vs 93 - on the screen of my computer,
and counted with his own hands the average number - 91 ups it turned out.
So the script does not work correctly. But for comparative performance - it’s quite suitable

[Syhn]

104 ups vs 101 on the site
95 vs 93 - on the screen of my computer,

Hey! I was wondering where all these benchmarks were coming from. I'm the creator of the website.

I wasn't expecting such a variance (a steady increase in performance at that) between runs on the same machine so I didn't bother to take an average of all the runs in the benchmark script and had it just show you the UPS of the first one. Although I think I'll change that now that it's showing such a discrepancy...

[azesmbog]

Although I think I'll change that now that it's showing such a discrepancy...

Now well.
The result on the screen and on the site is the same,
just checked on 0.8.13 and 0.8.14 versions and 0.8.15 versions.

And yet, from not very pleasant for my mega base
Checked my database offline script.
The results are very mixed.

factorio200320.jpg (39.85 KiB) Viewed 7172 times

Checked my base offline script.
The results are very mixed.
UPS “floats” in the range from 38 to 28, this is a very large spread, while the real UPS is approximately 41-44

[azesmbog]

I'm the creator of the website.

Hello.
I did 20 or 30 tests today. But I don’t see a single result in the database on the site
Here is the last attempt, I copied it from the console:
https://factoriobox.1au.us/result/a3f99 ... d2ebb2883f

Benchmark Results
There was an error parsing this benchmark. I've notified my maintainers of the issue.

Hence the question.
Is it something I have broken or on your site, or do you consider my results to be invalid and reject them?)

Upd:
Hmmm ..
Judging by the fact that someone added their result after my message, the site is working fine.
What is wrong with my results then? I did it again! one try

bench270620.jpg (40.1 KiB) Viewed 6760 times

The expected result did not appear in the database: (((

Upd2:
And it's me again. You did not wait ??
I think that the script that determines the configuration of the equipment does not see my memory: ((Or does not understand.
I added another memory bar, but I had to lower the frequency of the memory - and two results were added to the database.
But I would still like to see the result in the database with one bar and at a higher frequency and with the best UPS.
Who is to blame and what to do??
Maybe something can be fixed at the conservatory? (c)

Upd3:
It took a little to learn powershell.
Here is what it gives out regarding the installed memory:

ps1.jpg (38.49 KiB) Viewed 6739 times

Well, why can't your script parse here?

[Syhn]

Hello.
I did 20 or 30 tests today. But I don’t see a single result in the database on the site
Here is the last attempt, I copied it from the console:
https://factoriobox.1au.us/result/a3f99 ... d2ebb2883f

Benchmark Results
There was an error parsing this benchmark. I've notified my maintainers of the issue.

Hence the question.
Is it something I have broken or on your site, or do you consider my results to be invalid and reject them?)

Hey, sorry about that. I don't log into the forums often. It looks you did in fact find the issue for me. Powershell returns an array of memory banks with specs on each one, but I never tested with a single memory bank. It turns out that when there is only one, it doesn't return an array, but just the single memory bank, which broke my parsing script. I've gone ahead and fix it and all of them show up now.

[orzelek]

This benchmark shows how bad Ryzen is for Factorio:
https://factoriobox.1au.us/result/f4fe9 ... 941ac1ea8f
Thats 75 UPS only compared to more then 100 for intel CPU's.

I guess that having only 3200MHz RAM doesn't help.