There are comparision shots. It’s clearly show that on rendering thread is about 5ms slower on average. Performance is also quite spiky. All that Slate time is coming from that stat command. We don’t use Slate.
on 419, is any entry slowing down with time? might take some time to notice.
I agree 100%. To add to the OMG… target fix on 4.21… geez
FDrawSceneCommand went from 22ms to 34ms after 15minutes. Everything else was about the same so and device was not hot so it wasn’t just lower clock rates.
Yeah… well, the truth is out there somewhere. I had not been able to pinpoint the exact change to blame yet.
Well… so far 4.19 has been an unpleasant experience indeed.
Could you possible repeat this test, but give minute by minute times for 20-30 minutes? I want to plot a graph and fit a curve to see if it’s a linear rate or an exponential rate. It would help rule out what kind of operations to look for in the source. I’m still convinced it’s an x+=1 error of some sort.
I can try but on mobile this will be quite problematic. Clock rates are constantly adjusted and overheating is also possibility. Maybe I can try to use profiling build that should disable some of these stuff.
I really don’t belive that it is something as simple as that. I haven’t managed to get simple reproducable project that also have this bug. My guess that it’s something to do with threading and it’s only show it self when there is enough load in first place.
Engine is leaking like a colander, **2MB per second ** in a clean blueprint project with some load.
(Open the project. Load BugMap level. Play in standalone).
4.18.1
4.19.2
Exactly how did you test this? I opened the project in 4.19.2 and the used physical went from 969.03 to 975.82 in five minutes.
There is a map in project. Not the default one.
Open it, PIE or play in standalone, or cook and play. That is pretty much it.
That’s pretty awful stuff, but personally, I haven’t experienced any sort of memory leaking. I wonder what’s causing yours.
I ran your test, opened the bug map with the RT cube of death, ran stat memory, saw no climb in ram usage. I also checked rhi.DumpMemory as well.
What command are you running to output that log? I’ll try that one as well.
memreport -full
Can equally check memory in task manager of a standalone game process.
Yeah I figured it out, forgot about that command. Here were my results with just memreport (didn’t use -full):
START:
Time Since Boot: 21.50 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 905.17 MB used, 912.59 MB peak
Process Virtual Memory: 1188.81 MB used, 1256.94 MB peak
Physical Memory: 4848.84 MB used, 7349.86 MB free, 12198.69 MB total
Virtual Memory: 6089.23 MB used, 7349.86 MB free, 134217728.00 MB total
END:
Time Since Boot: 320.98 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 891.34 MB used, 912.59 MB peak
Process Virtual Memory: 1176.10 MB used, 1256.94 MB peak
Physical Memory: 4841.64 MB used, 7357.05 MB free, 12198.69 MB total
Virtual Memory: 6065.16 MB used, 7357.05 MB free, 134217728.00 MB total
Ran the test project on my machine (16GB, i7 4790, Nvdia gtx970:
Time Since Boot: 15.80 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 885.30 MB used, 928.42 MB peak
Process Virtual Memory: 1165.01 MB used, 1225.09 MB peak
Physical Memory: 6323.74 MB used, 10008.88 MB free, 16332.62 MB total
Virtual Memory: 1924.24 MB used, 10008.88 MB free, 134217728.00 MB total
Time Since Boot: 144.25 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 1187.74 MB used, 1514.75 MB peak
Process Virtual Memory: 1467.54 MB used, 1796.19 MB peak
Physical Memory: 6625.25 MB used, 9707.37 MB free, 16332.62 MB total
Virtual Memory: 2205.23 MB used, 9707.37 MB free, 134217728.00 MB total
Tested on laptop (surface book i7 6600, 16GB):
Time Since Boot: 20.31 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 903.53 MB used, 926.16 MB peak
Process Virtual Memory: 1172.13 MB used, 1218.98 MB peak
Physical Memory: 6104.42 MB used, 10204.88 MB free, 16309.30 MB total
Virtual Memory: 6053.99 MB used, 10204.88 MB free, 134217728.00 MB total
Time Since Boot: 278.70 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 1079.44 MB used, 1259.39 MB peak
Process Virtual Memory: 1349.44 MB used, 1530.75 MB peak
Physical Memory: 6123.84 MB used, 10185.46 MB free, 16309.30 MB total
Virtual Memory: 6268.73 MB used, 10185.46 MB free, 134217728.00 MB total
I wanted to run it for longer to see if it was getting worse, but it was struggling at a staggering 7fps… so I gave it a break.
Not really sure what to think.
@IronicParadox
@spazfirem
Cheers for tests.
@spazfirem
BP_DummyLoad has a variable NumIterPerTick. You can lower it to something more reasonable, to get better fps.
To clarify, surface book, the one with Intel or Nvidia graphics card ?
This would be the one with the Nvidia graphics card. I’m not exactly sure which model was put in there, but its obviously nothing great.
Thanks for the tip on the variable. I’ll play around with that and see if anything presents itself.
Interesting test project. Left it run for a little longer.
Right after starting, game is running nicely:
Time Since Boot: 11.46 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 928.68 MB used, 952.60 MB peak
Process Virtual Memory: 1233.97 MB used, 1285.43 MB peak
Physical Memory: 9660.91 MB used, 22779.73 MB free, 32440.64 MB total
Virtual Memory: 6178.95 MB used, 22779.73 MB free, 134217728.00 MB total
25 minutes later, game runs at glorious 1 second per frame:
Time Since Boot: 1451.19 Seconds
Platform Memory Stats for Windows
Process Physical Memory: 2918.06 MB used, 4959.06 MB peak
Process Virtual Memory: 3226.55 MB used, 5272.55 MB peak
Physical Memory: 11556.80 MB used, 20883.84 MB free, 32440.64 MB total
Virtual Memory: 8182.88 MB used, 20883.84 MB free, 134217728.00 MB total
What kind of video cards and drivers are people running?
540m 385.69 here on this laptop and I don’t have this memory leaking problem.
1060 3gb on 390.65