Observations of FPS
#1
Posted 24 February 2020 - 05:05 PM
I had long understood that CPU utilization and FPS were inversely related. The other day I was running the monogame version and was a bit surprised by the numbers I saw.
Here's a screenshots I set up after noticing the numbers (click once on image to increase size for easier reading).
fps1.jpg (150.7K)
Number of downloads: 43
No trains were moving when I took that screenshot.
Out of curiosity I moved the came several tiles away to a location with no placed objects and aimed it off into the distance, also no placed objects. I figured this should be a lot better, but no such luck (again, click to enlarge):
fps2.jpg (153.15K)
Number of downloads: 37
The relevant specs:
Processor = Intel® Core i7-5820K CPU @ 3.30GHz (12 threads, 6 cores, 3.3 GHz) (CPU Internal L1 384 KB) (CPU Internal L2 1536 KB) (CPU Internal L3 15360 KB)
Memory = 15.9 GB
Video = NVIDIA GeForce GTX 760 (4.0 GB RAM) (nvlddmkm 26.21.14.3160)
Video = NVIDIA GeForce GTX 760 (4.0 GB RAM) (nvlddmkm 26.21.14.3160)
Display = \\.\DISPLAY1 (1920 x 1200, 32-bit, primary, 0 x 0)
and
Version = NewYear MG (unofficial)-U2020.02.16-0106 Rev. 53.1.
Build = 0.0.7344.17260 (2020-02-09 09:35:20Z)
Can you explain why the render prims went up by 3X in the second image? Or why the CPU utilization at 71% is unable to produce more than 33fps?
I'm asking for the purpose of understanding, not to complain... I think going to Monogame is a very good course to peruse and am thankful you guys are doing this work.
#2
Posted 24 February 2020 - 06:17 PM
Quote
CPU
Intel Core i7 3770 @ 3.40GHz 111 °F
Ivy Bridge 22nm Technology
RAM
32.0GB Dual-Channel DDR3 @ 799MHz (9-9-9-24)
Motherboard
Intel Corporation DH77EB (CPU 1) 90 °F
Graphics
VE248 (1920x1080@60Hz)
2047MB NVIDIA GeForce GTX 650 Ti (ZOTAC International) 74 °F
Storage
476GB Samsung SSD 850 PRO 512GB ATA Device (SATA (SSD)) 79 °F
931GB Western Digital WDC WD1003FZEX-00MK2A0 ATA Device (SATA ) 90 °F
931GB Western Digital WDC WD1002FAEX-00Z3A0 ATA Device (SATA ) 96 °F
931GB TOSHIBA External USB 3.0 USB Device (USB (SATA) )
1862GB Western Digital WD My Passport 25E1 USB Device (USB )
#3
Posted 24 February 2020 - 06:45 PM
#4
Posted 24 February 2020 - 09:18 PM
#5
Posted 25 February 2020 - 12:06 AM
can you check if you have the same result with the Unstable release (that includes MG too)? At what value have you set the multisampling slider in the experimental options?
I experienced a significant fall in FPS in the MG version passing from one MG .dll pack to a more recent one. This occurred a couple of years ago.
#6
Posted 25 February 2020 - 09:33 AM
fps3.jpg (72.76K)
Number of downloads: 4
The route I was running is my Goose Island route and so some of the values are not really relevant to this route (IMO it is long past due this stuff get moved to a file in the route's home directory). Note view distances. There are oh, 7, maybe 8 populated tiles surrounded by 6-10 tile sized ring of unpopulated tiles, mostly due to an ill defined scope when the terrain was projected. Last image above was within the outer ring looking further out.
#7
Posted 25 February 2020 - 09:02 PM
My suggestion is that if anybody has the full set of OR New Year MG releases that it get uploaded someplace so that testing for FPS performance drop be documented and FIXED. No amount of "re-factoring" is going to fix this, shall I say "mess". At this point, MSTS outperforms OR, which is unthinkable. We might have to go back to the "drawing board" and use what works, even if for the time being we are "stuck" with 32 bits.
There is no shame in acknowledging that this experiment did not pan out. I have been at this for almost 10 years now, I have never seen releases perform this poorly. In past years I would often test for this type of performance issue. But we seemed to be "demonically bent" on tossing out WinXP (32 Bit!), and supporting Win7/Win10 only. This created some problems for me, especially with regards to models for the sim, but over time some things have gotten resolved. I am writing this from a Win 7 machine.
Steve
#8
Posted 26 February 2020 - 02:08 AM
To ease comparisons with older revisions of OR NewYear MG, and also to OR MG revisions preceding OR NewYear MG, I have uploaded here http://www.interazio...usRevisions.zip a pack with 7 older revisions, starting from very old ones up to OR NewYear MG rev. 50.
If you want to publish comparisons, take screenshots in the same conditions (about 10 seconds after game start) and attach to each of them the related OpenRailsLog.txt file, because default settings might be different.
Take also into account the evolution of the Multisampling feature, which heavily impacts on FPS: in versions up to March 1st, 2019 "Enable Multisampling" option unchecked means Multisampling disabled, while option checked means Multisampling to max (which should be 32 if supported by hardware). In later versions that don't have yet the Multisampling slider "Enable Multisampling" option unchecked means Multisampling = 4, while option checked means again Multisampling to max. (So e.g. the attached revision 20190112_OR_NewYear_MG_4334 would perform better than 20190522_OR_NewYear_MG_22 in terms of FPS when "Enable Multisampling" option is disabled, because the first one would run with Multisampling disabled, while the second one would run with Multisampling = 4.) In later versions with slider you can select values from 0 (no multisampling) to 4 (default) to 32.
I remind also that, starting from my repository, any older version of OR NewYear MG may be regenerated rebuilding it.
#9
Posted 26 February 2020 - 05:04 AM
Yeah - while I have a pretty decent machine which powers through most issues - I've been noticing a hit on the frames as well...
Regards,
Scott
#10
Posted 26 February 2020 - 08:01 PM
Csantucci, on 26 February 2020 - 02:08 AM, said:
If you want to publish comparisons, take screenshots in the same conditions (about 10 seconds after game start) and attach to each of them the related OpenRailsLog.txt file, because default settings might be different.
Hi Carlo,
Tonight I am not writing this from my Win 7 machine. Thank you for the link to the older versions, I will download this from the Win 7 machine I hope by Friday. I forgot to add, that I also tested with multisampling off in the latest release, which resulted in FPS of about 42 FPS. This number is well below the "solid" 60 FPS for Rel. 35. I will pay close attention to the multisampling setting as you have mentioned. As you well know this type of testing is very tedious, especially in terms of documenting the exact point of failure. I realize that all these releases take up needed/precious disk space, but I will appreciate any effort to reproduce intermediate releases so we may pinpoint, I hope, exactly where performance began to drop off.
Sorry that this is frustrating, but I appreciate your cooperation in this matter. Hopefully, some others will help with the hard work of testing. In years past, this has often been a thankless task, with a lot of unwanted and rather useless rhetoric posted in response to results, not to mention denials that there was a problem/bug! I realized that thorough testing was not something that we were ready to deal with as far back as 2012! I have tailored my efforts since that time. Usually I find workarounds so as to not "bother" the team with my findings. In this case there does not seem to be a workaround, except to use a much "older" version.
Molte gracie,
Steve