FM19 Performance Benchmarking Thread

[Gee_Simpson]

Not sure if I should return my i9 9900k to get it replaced with a new one. I have until the 31st January to do this, not sure if a new one will provide better times or should I not bother? 

[asapas]

So, to sum it up.... A rather generic question... Football manager 2020 (and probably the following ones), will run for example on Ryzen 3950X better or worse than on Intel I7-9700/I7-9900. 
(Assuming that we have the same other parts, for examplel the same 16GB RAM, same GPU, same SSD etc.). 

[Gee_Simpson]

1 hour ago, asapas said:

So, to sum it up.... A rather generic question... Football manager 2020 (and probably the following ones), will run for example on Ryzen 3950X better or worse than on Intel I7-9700/I7-9900. 
(Assuming that we have the same other parts, for examplel the same 16GB RAM, same GPU, same SSD etc.). 

I would imagine the i7 9700k and i9 9900k would run it slightly quicker due to the better single core performance of Intel CPU's. AMD is closing the gap but still trails on single core speeds, which is what FM favours. 

[asapas]

7 hours ago, Gee_Simpson said:

I would imagine the i7 9700k and i9 9900k would run it slightly quicker due to the better single core performance of Intel CPU's. AMD is closing the gap but still trails on single core speeds, which is what FM favours. 

Thank you very much for your help. Another question I have though, is if you want to run football manager 2020 with let's say over 50 google chrome tabs running at the same time
is it better to run it with the 3950X, or with Intels???? Because I have heard that Ryzens are better for multitasking.

[Jakobhg]

On 18/01/2020 at 03:51, Gee_Simpson said:

Well, after some tweaking, it seems I can hit a 5ghz OC afterall. 

I'm really not sure how your 3rd test is that quick?

@Jakobhg and my new i9 9900k are very close, and we both have overclocked to 5ghz all cores. I even tested stock just to see if I could match your numbers but there's no way I can get near that. Here's my benchmarks with the overclock and without it:

i9 9900k, 16gb DDR4 3200mhz RAM, GTX 1660, Corsair MP510 NVMe SSD.

5ghz OC:

Benchmark A: 2:14
Benchmark B: 4:30
Benchmark C: 7:15

Stock:

Benchmark A: 2:13
Benchmark B: 4:41
Benchmark C: 7:17

Sorry man! Haven't checked the forum in a long time. My bad!
Yeah, our CPUs sound similar now. The overclocking part does not grant much extra performance, but a little.
However, you can try to increase your clock speed on your RAM. That actually gave me quite good results. Might be the difference between our two tests. I run 3600 Hz on my RAM.

[Gee_Simpson]

On 23/02/2020 at 12:05, Jakobhg said:

Sorry man! Haven't checked the forum in a long time. My bad!
Yeah, our CPUs sound similar now. The overclocking part does not grant much extra performance, but a little.
However, you can try to increase your clock speed on your RAM. That actually gave me quite good results. Might be the difference between our two tests. I run 3600 Hz on my RAM.

What is the default speed of your RAM, and how much of a difference did it make? Did you change any of the timings or did you only change the frequency?

Edited February 29, 2020 by Gee_Simpson

[FMFutbol]

out of curiosity, has anybody done the benchmark on the new and ultra expensive Mac Pro desktops?

[Jakobhg]

On 23/02/2020 at 20:06, Gee_Simpson said:

What is the default speed of your RAM, and how much of a difference did it make? Did you change any of the timings or did you only change the frequency?

The default speed is 3600 Hz and bought like that. But you still have to unlock it in BIOS. Think it ran 1800 or 2000 Hz before that.
I did not change any timings. It is very time consuming and might not be worth the effort.
It's been awhile now, so I can't remember the exact time saved by pushing the RAM to 3600 Hz and I haven't noted it anywhere. Sorry. But somewhere between 10-30 seconds would be my guess for these kind of tests.

[Gee_Simpson]

5 hours ago, Jakobhg said:

The default speed is 3600 Hz and bought like that. But you still have to unlock it in BIOS. Think it ran 1800 or 2000 Hz before that.
I did not change any timings. It is very time consuming and might not be worth the effort.
It's been awhile now, so I can't remember the exact time saved by pushing the RAM to 3600 Hz and I haven't noted it anywhere. Sorry. But somewhere between 10-30 seconds would be my guess for these kind of tests.

Ah yeah, you mean activating the XMP Profile to run it at advertised speeds. I thought you had actually manually overclocked your RAM but that's not the case. I bought 3200mhz RAM as 3600mhz cost much more so I didn't think it was worth it. I was thinking of manually overclocking my RAM but like I decided with the processor after advice from @Brother Ben, I don't think it's worth the extra strain. I think I'll wait to overclock those when this PC starts getting old to squeeze more life out of it. 

[Brother Ben]

9 hours ago, Gee_Simpson said:

Ah yeah, you mean activating the XMP Profile to run it at advertised speeds. I thought you had actually manually overclocked your RAM but that's not the case. I bought 3200mhz RAM as 3600mhz cost much more so I didn't think it was worth it. I was thinking of manually overclocking my RAM but like I decided with the processor after advice from @Brother Ben, I don't think it's worth the extra strain. I think I'll wait to overclock those when this PC starts getting old to squeeze more life out of it. 

I've actually dialled down my own overclock, half of the fun was in the tinkering and testing involved and whilst it does/can give you great gains in FM I tend to use the same save over multiple devices which aren't as powerful so it wasn't really worth it in my own use-case.  

Mind you if I ever decide to upgrade my old rig to the kind of impressive kit that you have no doubt i'll hammer it to see what it's capable of! 

[Brother Ben]

Okay i'm definitely thinking of setting up a Benchmarking thread for FM20 but i'm wondering what tests I should have

I'm thinking maybe test what i imagine is one of the most common types of setup (England, Italy and Spain - Large database)

so maybe

1 - England, Italy and Spain all divisions active - Full detail on only the one active league

2 - England, Italy and Spain all divisions active - Full detail 3 divisions

3 - England, Italy and Spain all divisions active - Full detail 10 divisions

This would be testing the multi core and threading of processors as i think we're at the point now where I don't really see the point of running the test with all leagues loaded as the results of the above will scale upwards anyway

I have 4 computers in the house that are all different

• An overclocked desktop
• A 17 inch desktop replacement Laptop
• A Windows Tablet
• An Intel NUC that i use as a media centre through my TV

This should give a nice variation of results

I'll set up the 3 saves tonight and run them on my own kit to see how it goes first and report back

What does everyone think?

I've mentioned before though i'm crap at graphs and spreadsheets but i'm sure I can knock something up based on what has gone before

Edited March 4, 2020 by Brother Ben

[jonpt]

6 hours ago, Brother Ben said:

Okay i'm definitely thinking of setting up a Benchmarking thread for FM20 but i'm wondering what tests I should have

I'm thinking maybe test what i imagine is one of the most common types of setup (England, Italy and Spain - Large database)

so maybe

1 - England, Italy and Spain all divisions active - Full detail on only the one active league

2 - England, Italy and Spain all divisions active - Full detail 3 divisions

3 - England, Italy and Spain all divisions active - Full detail 10 divisions

This would be testing the multi core and threading of processors as i think we're at the point now where I don't really see the point of running the test with all leagues loaded as the results of the above will scale upwards anyway

I have 4 computers in the house that are all different

• An overclocked desktop
• A 17 inch desktop replacement Laptop
• A Windows Tablet
• An Intel NUC that i use as a media centre through my TV

This should give a nice variation of results

I'll set up the 3 saves tonight and run them on my own kit to see how it goes first and report back

What does everyone think?

I've mentioned before though i'm crap at graphs and spreadsheets but i'm sure I can knock something up based on what has gone before

I would be really interested to see how those scenarios would pan out. Especially the third file with 10 leagues in full detail. 

[FMFutbol]

I think you should in reality on do option #3 for the benchmark. Those that will want to know the speeds will play with full detail.

[Brother Ben]

1 hour ago, FMFutbol said:

I think you should in reality on do option #3 for the benchmark. Those that will want to know the speeds will play with full detail.

Do you mean active leagues or full detail?

Either way after testing a few scenarios i've decided to try and exactly replicate the 2019 tests in 2020 as the first 2 tests I suggested i tried a couple of variations of the theme and the results were exactly the same on 2 machines with pretty varied spec.  The amount of leagues was clearly too low to be useful.

I've messaged @SebastianRO to see if he is going to do the 2020 version and if not i'll start it ASAP

Edited March 5, 2020 by Brother Ben

[Brother Ben]

Okay he's given me the go-ahead so last night I set up 3 saves that exactly replicate the 2019 tests.  Test 1 was a similar outcome to before however test 2 took nearly 3 times as long as last years.  Haven't even tried the third test yet as that can only be even worse.

Not too sure how to proceed now, not sure if people want to be running tests that take that long so I may have to tweak the holiday time to compensate.

If anyone has any suggestions of how to set up the tests, maybe with a view to testing expected outcomes, i'm all ears

[jonpt]

I would be happy to benchmark for a longer period of time but I can understand why others would not. 

For full detail are you setting every match in every division in the countries or just the top league? Perhaps that could be the tweak, Otherwise a shorter time period would work. Ideally if the number of days were reduced the holiday would still be over a weekend so the leagues loaded would still have matches.

 

Just thoughts.

 

 

[Brother Ben]

10 minutes ago, jonpt said:

I would be happy to benchmark for a longer period of time but I can understand why others would not. 

For full detail are you setting every match in every division in the countries or just the top league? Perhaps that could be the tweak, Otherwise a shorter time period would work. Ideally if the number of days were reduced the holiday would still be over a weekend so the leagues loaded would still have matches.

 

Just thoughts.

 

 

Good points.

At the moment i'm thinking that i will maybe drop the transfer window test

so

Benchmark A - Top 2 divisions from 10 nations - hopefully to encourage people with lesser hardware to get involved

Benchmark B - All leagues loaded - Default detail (essentially means just one league full detail)

Benchmark C - Same as B but with full detail level on maybe 10 active divisions (i'll do a few tests first to make sure it doesn't take forever - hopefully find a sweet spot between worthwhile testing conditions and how long it takes)

What do you think?

[EdL]

Hmm interesting / worrying that test i so much slower in fm20, our internal performance tests show us as faster than 19 across the board, possibly this is catching some issue we are missing.

But also we have fixed some performance issues for the upcoming patch, so probably worth testing again when that is out - if its still an issue let me know and i can investigate whats going on

[Brother Ben]

50 minutes ago, EdL said:

Hmm interesting / worrying that test i so much slower in fm20, our internal performance tests show us as faster than 19 across the board, possibly this is catching some issue we are missing.

But also we have fixed some performance issues for the upcoming patch, so probably worth testing again when that is out - if its still an issue let me know and i can investigate whats going on

I think it may just be a coincidence as i was trying to replicate exact conditions from 2019 and there may just be more games on that particular gameweek leading to longer times, I may reinstall FM19 to investigate but to be honest there's so many variables between any given save that comparing different editions of FM is nigh on impossible

[EdL]

31 minutes ago, Brother Ben said:

I think it may just be a coincidence as i was trying to replicate exact conditions from 2019 and there may just be more games on that particular gameweek leading to longer times, I may reinstall FM19 to investigate but to be honest there's so many variables between any given save that comparing different editions of FM is nigh on impossible

but still nearly 3x longer seems very wrong, especially if that save is using default detail levels

[Brother Ben]

Fair point, the test was across a transfer window if that makes any difference?

ive not got the file that I created anymore but as I say I set it up the same as the one in the OP, ill try and recreate it when I get a chance.

*edit* it’s likely now that in light of what you have said I will keep Benchmark B and C For this years thread as close as possible to last years conditions so you can see if there is an issue

Edited March 5, 2020 by Brother Ben

[EdL]

20.4 was just released so worth testing with that

[Jakobhg]

On 04/03/2020 at 02:49, Gee_Simpson said:

Ah yeah, you mean activating the XMP Profile to run it at advertised speeds. I thought you had actually manually overclocked your RAM but that's not the case. I bought 3200mhz RAM as 3600mhz cost much more so I didn't think it was worth it. I was thinking of manually overclocking my RAM but like I decided with the processor after advice from @Brother Ben, I don't think it's worth the extra strain. I think I'll wait to overclock those when this PC starts getting old to squeeze more life out of it. 

Yes that is what I meant - sorry for the confusion.
I fully agree that overclocking your RAM or optimizing your timings are not worth it.
3200MHz RAM should be fine. But when I was buying the sweet spot was 3600MHz. Above that was really expensive and not worth it. But this depends on the time of buying and the country of purchase i guess.

[djinni999]

Hi guys. I did benchmark A. My system is Ryzen 3300X, a bit undervolted but @3.4 GHz all-core, 16 GB 3200CL14 RAM, 1080 TI, SSD. I got 3:12 and I'm a bit underwhelmed. I was expecting better. Is my result normal? Using the FM2020 Benchmark A settings on FM19 I got a really good score, beating all Ryzens and tying the 9700K for 5th place. 

[Brother Ben]

20 minutes ago, djinni999 said:

Hi guys. I did benchmark A. My system is Ryzen 3300X, a bit undervolted but @3.4 GHz all-core, 16 GB 3200CL14 RAM, 1080 TI, SSD. I got 3:12 and I'm a bit underwhelmed. I was expecting better. Is my result normal? Using the FM2020 Benchmark A settings on FM19 I got a really good score, beating all Ryzens and tying the 9700K for 5th place. 

Sounds about right for Ryzen on FM19, there's a 3900X on page 2 that gets 2:42.  bear in mind Benchmark A is different for FM20

[djinni999]

1 hour ago, Brother Ben said:

Sounds about right for Ryzen on FM19, there's a 3900X on page 2 that gets 2:42.  bear in mind Benchmark A is different for FM20

I did state that I used FM20 bench A settings for FM19. I equaled the 9700K at 1:00. Using FM19's Bench A, I got 3:12. I am aware that the 3900X gets 2:42, which is the very reason that I made the post. I was expecting it to be close. I'm confused as to the discrepancy. Could it be that loading 102 or so leagues totally taxes the 4-core, while the 12-core is able to handle it much better? FM20 Bench A has only 20 leagues loaded, so that could explain why I was able to beat the 3900X there. Also, it seems FM20 has benefited Ryzens more, as they seem to have closed the gap between the top Intels.

[Brother Ben]

2 hours ago, djinni999 said:

I did state that I used FM20 bench A settings for FM19. I equaled the 9700K at 1:00. Using FM19's Bench A, I got 3:12. I am aware that the 3900X gets 2:42, which is the very reason that I made the post. I was expecting it to be close. I'm confused as to the discrepancy. Could it be that loading 102 or so leagues totally taxes the 4-core, while the 12-core is able to handle it much better? FM20 Bench A has only 20 leagues loaded, so that could explain why I was able to beat the 3900X there. Also, it seems FM20 has benefited Ryzens more, as they seem to have closed the gap between the top Intels.

I think unless you use the same save file, which you can’t in this instance it’s hard to tell. Even if you attempt to replicate it as much as you can it won’t be perfect.

Your results for test A on the FM19 thread are exactly what I would expect.  Matching an overclocked 9700k is unlikely with your chip so all it proves is that it’s nearly impossible to replicate the conditions without using the same save file, unless you get FM20 it’s hard to be sure

I do think you’ve nailed it with your closing sentence though , SI have worked hard to accommodate more cores

Edited June 13, 2020 by Brother Ben

[djinni999]

I see. But I did set it up with almost the exact player count, database size. Anyway, so do you think the extra 8 cores is the difference between 2:42 and 3:12 times on the 102 league bench?

[Brother Ben]

Impossible to make exactly the same though unfortunately.  I'd say the difference is very slight considering the 3900x costs 4 times as much. I don't think its so much the cores as it is the higher single core frequency that FM likes, the cores will make more of a difference on benchmark C