Page 4 of 5
Re: 2019-03-28 snapshot
Posted: Tue Apr 09, 2019 2:46 am
by jlv
chaseb31 wrote:Hate to be the guy who says it, but first thanks to jlv for the help and ideas. But is there a way we can problem solve our self or can you find something that may be the issue, I hate asking this like this but we are still racing each week ad doesnt seem to stop the racing weather its happening to majority of the riders, and we paid to race its kinda sucky to have to race with this going on. Would love to get it fixed so we can finish out the last couple rounds and be amped on Outdoors to come.
You could try rolling back your drivers to see if an older version is better. I'll add some finer grained instrumentation code to the game so I can at least get a general idea of where it's hanging. I might be able to find a workaround for whatever's happening if I know where it is. It's hard to fix this stuff because it's usually in code I have no control over.
Re: 2019-03-28 snapshot
Posted: Tue Apr 09, 2019 3:02 am
by chaseb31
Awesome thanks for the idea jlv, I will give it a try. Glad to see the fast reply and thoughts on working on it and possibly finding a fix to the situation.
Re: 2019-03-28 snapshot
Posted: Thu Apr 11, 2019 8:04 pm
by Hi Im Skyqe
Some interesting findings by Seabolt. He already PM'd this information to JLV.
Everyone who is having these hard freezes and controllers intermittently freezing, can you confirm if this erode loading issue is happening for you as well so JLV can have something with substance to look for?
Run the 3-28 Snapshot in single player with 100 erode and 0 reshade delay. See if your erode does not load in behind you like in this clip.
https://gyazo.com/67b83f5ffb0a0730e76bc63540eaac2e
Run the 1-26 snapshot with the same settings as above and see if erode loads in behind you almost instantly.
https://gyazo.com/c6221b9d3e32ea0535ed84131a730bef
Windowing in and out of 3-28 Snapshot, erode loads instantly.
https://gyazo.com/1704dcff3d69e39090880a72e17a0387
Re: 2019-03-28 snapshot
Posted: Thu Apr 11, 2019 8:47 pm
by KaseoDip
[quote="Hi Im Skyqe"][/quote]
I'm confused by the last one. It's the same settings as the first two, the erode isn't loading, but then Seabolt tabs out and then back in and it loads immediately?
Re: 2019-03-28 snapshot
Posted: Thu Apr 11, 2019 8:47 pm
by KaseoDip
Re: 2019-03-28 snapshot
Posted: Thu Apr 11, 2019 9:01 pm
by Hi Im Skyqe
KaseoDip wrote:Hi Im Skyqe wrote:
I'm confused by the last one. It's the same settings as the first two, the erode isn't loading, but then Seabolt tabs out and then back in and it loads immediately?
Correct same settings on the 3-28 Snapshot. Erode is loading slowly, once you tab out of the game to desktop and back in erode will load immediately.
Re: 2019-03-28 snapshot
Posted: Thu Apr 11, 2019 11:06 pm
by Hi Im Skyqe
Can confirm this issue happens with a Ryzen 1600 and RX580 GPU too. This isn't isolated to just nVidia graphics cards.
Might have found a fix for AMD. Still doesn't really explain why it doesn't happen with the older snapshots but this seems to help.
Create a custom profile for MXS in the AMD Catalyst Panel and turn anti aliasing mode to super sampling or adaptive multi sampling. With this change erode loads immediately on the 3-28 Snapshot.
Re: 2019-03-28 snapshot
Posted: Fri Apr 12, 2019 1:44 am
by jlv
I'm not seeing any difference between 1-26 and 3-28 doing the single player erode test. Not surprising since they're running the same erode code. 1-26 was the last time it changed, adding more erode buffering. It'd make much more sense if the problem started at 1-26 since there was an actual change in that version. If it's different since 3-20 it'd have to be some kind of weird indirect effect from something else. I've asked Jeremi to send me the demo in question. Maybe that'll help.
Do the stutters go away if you set the terrain rendering method to "software"?
Re: 2019-03-28 snapshot
Posted: Fri Apr 12, 2019 2:10 am
by Hi Im Skyqe
jlv wrote:I'm not seeing any difference between 1-26 and 3-28 doing the single player erode test. Not surprising since they're running the same erode code. 1-26 was the last time it changed, adding more erode buffering. It'd make much more sense if the problem started at 1-26 since there was an actual change in that version. If it's different since 3-20 it'd have to be some kind of weird indirect effect from something else. I've asked Jeremi to send me the demo in question. Maybe that'll help.
Do the stutters go away if you set the terrain rendering method to "software"?
After a bit more testing, if he sets threaded optimization to auto 1-26 erode works, with 3-28 it does what is shown in the clip. If he sets threaded optimization to off for both 1-26 and 3-28 erode works as intended. With 3-28 and threaded optimization off, there are more frame drops with 3-28 than with 1-26.
Software made his game go to 25 fps, erode was still not loading in unless threaded optimization is set to off.
Re: 2019-03-28 snapshot
Posted: Mon Apr 15, 2019 5:16 am
by NoFear345
I just tested their finding of slow erode on 3-28 with the same result.
Re: 2019-03-28 snapshot
Posted: Tue Apr 16, 2019 12:48 am
by jlv
How big of a stutter are you guys getting? We're talking about a half second freeze right? Does it reliably happen in
this round 15 main demo? Does /fpshist show it? (Type /fpshist once after loading in to clear the counters and again at the end of the demo.)
NoFear345 wrote:I just tested their finding of slow erode on 3-28 with the same result.
The slow update in the editor thing is a red herring. All it's showing is how quickly you switched to the editor. If you look at the clips posted, in the "slow update" clip he doesn't even come to a complete stop before switching. The others he waits a second before switching. Stuff like this makes me seriously doubt it's a 2019-3-20 problem. I understand why people would want the problem to be in my code, I'd prefer that myself since it'd be easier for me to fix that way, but manipulating tests to make it look that way just confuses things and makes it even harder to find the real problem.
Re: 2019-03-28 snapshot
Posted: Tue Apr 16, 2019 1:51 am
by NoFear345
I didn’t test by going into editor, I turned erode up went down the start straight did a 180 and sat still. Watched erode slowly form for about a minute.
Re: 2019-03-28 snapshot
Posted: Tue Apr 16, 2019 2:52 am
by Hi Im Skyqe
jlv wrote:The slow update in the editor thing is a red herring. All it's showing is how quickly you switched to the editor. If you look at the clips posted, in the "slow update" clip he doesn't even come to a complete stop before switching. The others he waits a second before switching. Stuff like this makes me seriously doubt it's a 2019-3-20 problem. I understand why people would want the problem to be in my code, I'd prefer that myself since it'd be easier for me to fix that way, but manipulating tests to make it look that way just confuses things and makes it even harder to find the real problem.
There was no manipulation or ill intent with this, Seabolt just rode and I took the gyazo's through teamviewer to capture what is happening. It didn't matter if or when he stopped the erode was still loading slow. All settings were identical only the snapshot being the difference for accuracy.
We are just trying to give you information to help find the issue.
Re: 2019-03-28 snapshot
Posted: Tue Apr 16, 2019 3:02 am
by mx201
https://clips.twitch.tv/ScrumptiousToughTarsierFunRun
https://clips.twitch.tv/SleepyImpartialNikudonMcaT
Something we noticed tonight, not sure if this is normal but the erode would only appear for some and not others while also randomly showing for some before a new race even started. This clip was from a higher erode but this also happened with small erode aswell. Not sure if flowtech servers or if its a snapshot thing
Re: 2019-03-28 snapshot
Posted: Wed Apr 17, 2019 2:20 am
by jlv
NoFear345 wrote:I didn’t test by going into editor, I turned erode up went down the start straight did a 180 and sat still. Watched erode slowly form for about a minute.
It'll be the same in both versions. There were no erode changes in the versions in question. There will be some random variations due to how full the erode buffer is and where the shading is currently being updated but on average it will be the same.
Hi Im Skyqe wrote:There was no manipulation or ill intent with this, Seabolt just rode and I took the gyazo's through teamviewer to capture what is happening. It didn't matter if or when he stopped the erode was still loading slow. All settings were identical only the snapshot being the difference for accuracy.
I don't think there's any malicious intent. I'm just saying those tests are bogus. I was unable to reproduce it and then I noticed how the tests were actually conducted. He probably didn't do it on purpose but it's obvious what happened from looking at the videos.
Hi Im Skyqe wrote:We are just trying to give you information to help find the issue.
"How big of a stutter are you guys getting? We're talking about a half second freeze right? Does it reliably happen in this round 15 main demo? Does /fpshist show it? (Type /fpshist once after loading in to clear the counters and again at the end of the demo.)"
It would help to know how long the stutters are so I can write some instrumentation code that actually records the right info.
That's definitely a bug. I forgot to reset the erode buffer on restart. Will be fixed in the next update.