MTM2 Beta Patch 0.47 - Improving from .46

[OG_RuDeE]

The game ran much more smoothly with no Sam, but I still noticed some small stutters. The camera was clean when turning and jumping though which was an improvement compared with having Sam on.

[Fila]

http://www.mtm2.com/~mtmg/misc/Beta0.47_Rudee_test5.zip

Sorry, I said it was the last AI test but found 1 more discrepancy that we need to re-test to make sure I'm reading it correctly. So do the whole aidiag thing again please

I also need you to do the hitch again, bobsled is enough. Here's some info the AI is trying to point out

2. The 25fps episodes — driver-side, with a clear test plan. The dominant pattern is 7–10 second stretches where every frame blocks ~30ms in present (3 vblanks at 75Hz) while our entire GPU submission is under 6ms. A modern GPU crawling on a 1998 workload, intermittently, recovering on its own — that's the classic AMD downclocking profile: the game is too light to hold boost clocks, the GPU sags to idle, frames start missing vblanks, vsync quantizes the misses. For the tester:
- Re-run with in-game V-Sync off ("Flicker Reduction" in 3D Hardware Options) — if the 30ms toggles vanish or become raw ~27ms, the quantization half is confirmed.
- Watch the GPU core clock in the Adrenalin overlay during an episode — if it's idling around 500MHz mid-race, that's the whole story.
- Give monster.exe an Adrenalin per-game profile: Radeon Chill off, Enhanced Sync off, FRTC off, power mode to maximum.

Please do the tests with SAM off for now. And let me know if you are running v-sync or not.


In other news, the flicker on quid pro quo is finally fixed.

[Fila]

The game ran much more smoothly with no Sam, but I still noticed some small stutters. The camera was clean when turning and jumping though which was an improvement compared with having Sam on.

So it sounds like Sam does have a big say. The hitch test is definitely shorter with nosam.

[Fila]

Sorry for dragging you through this, but you seem to be the only one so far that has this specific problem. (the AI thing is more of a bonus, lol)

[OG_RuDeE]

Fun to see the progress in real time and happy to help. I'll run the tests in the morning.

[Fila]

The new update for you does have some code to try and stop your game from throwing 30 instant fresh sound calls at the same time (most likely when the trucks start piling up). These calls should have been pre-cached but the code was spitting them out and having your computer re-do them live.

[Fila]

Also, in regards to the MTM1 drags when played on MTM2, I wrote this in my wishlist too.

Fix that god awful bug in MTM1's B.C Place and RCA Dome where hitting the second set of cars causes the truck to become stuck/snagged on the cars to the point of losing as a result. This doesn't really happen a whole lot in MTM1, so this may tie into the bridge bug as well. MTM1's crush car objects don't operate in the same manner as in the case of certain MTM2 replica tracks where crush cars are set to "No collide" with ground boxes used to create jumps upon impact.

There's nothing more in this game right now that makes me just as angry compared to this bug, which I'm only assuming it's also tied into the fact to how trucks landing on bridges causes them to get stuck/snagged upon impact.

Side note: Use soft suspension to stop the likelihood of getting stuck on the second set of cars for now, but too bad my OCD driven brain keeps seeing it as an utter inconvenience just to keep doing it every single time I drive on those tracks.

Next overall patch or hotfix is gonna have an attempt at fixing this. Not sure how well it works, but it's an attempt.

[OG_RuDeE]

Here are the tests: https://drive.google.com/file/d/1zl8B0e ... drive_link, https://drive.google.com/file/d/1SVXl0n ... drive_link, https://drive.google.com/file/d/18P-Nr_ ... drive_link

Definitely got some good hitches on my second test, often when jumping. This has always been an issue for me. Another place it seemed to happen was when an object very close to the screen left the screen, e.g. going under a checkpoint, around when the model would leave the screen I'd get a pause. On previous tests, I'd often get a pause just before hitting an AI, I guess from it calculating the collision before it happened.

I have vSync on because when I first tried these new patches, e.g. 0.46, with vSync off I would get over 100fps but not smooth. I was getting big drops in fps, probably from the stuff you are fixing now.

Edit: Here's a test with vsync off: https://drive.google.com/file/d/1XCCFdD ... drive_link

I drove to the same place on Bobsled. Much shorter hitch file. I was still getting plenty of slowdowns, but I guess less hitching?

[OG_RuDeE]

Vsync back on. Sam still off.

I tested Quid pro quo: https://drive.google.com/file/d/1Bkl8DO ... drive_link

Very long hitch file. I have 10 or 20 hitches one after another right at the start of the race. It smoothed out quite a bit after getting away from the AI. I can confirm the fence flickering is fixed on my end.

[Ford-chan]

Joakim Rehn's Mad Mix (MTM1) doesn't appear to be reacting too kindly to this patch (hotfix included). From lag spikes galore to the whole terrain becoming a graphical mess.



Turning off Advanced Shadows doesn't work at all.

[Ford-chan]

Oh, this is ridiculous!...



Now S-L-O Construction is weeping too! Something must contaged within this hotfix, cause I went back to the vanilla Beta.47 just to confirm if it's the hotfix doing this, turns out it true.

[OG_RuDeE]

Just tested Quid pro on my 5600 XT machine. Only one hitch I noticed during a jump, but the hitch file has plenty of entries still: https://drive.google.com/file/d/1Vr_x3p ... drive_link

Edit: Here's Bobsled's hitch from 5600: https://drive.google.com/file/d/1aDB_Ge ... drive_link

Also runs very smooth on this machine. Maybe some frame drops when turning, but not hitches.

One question that occurred to me: both these PCs have APUs and dedicated GPUs. Dx9 is choosing the GPU for rendering right (and not the APU)?

[Fila]

A lot to look over but I can't today at all, will look into all this stuff tomorrow.

Quick note: The AI behaving different on different CPUs has been fixed at least. Your AI in this latest test have behaved identical to how it behaved on my end. So that issue is closed (at least that one, lol)

[Fila]

One question that occurred to me: both these PCs have APUs and dedicated GPUs. Dx9 is choosing the GPU for rendering right (and not the APU)?

Good question! I guess that is something I should look into first before doing any other tests.

[Fila]

Now S-L-O Construction is weeping too! Something must contaged within this hotfix, cause I went back to the vanilla Beta.47 just to confirm if it's the hotfix doing this, turns out it true.

Good news is that whatever it was, I must have fixed it by accident because none of this shows up on my end (on neither of the tracks).
I'm probably going to release a hotfix tomorrow as a side to make sure these additional issues are gone while still releasing tests for Rudees machine.

[OG_RuDeE]

Sounds good! Thanks Fila. I won't send anything else until I hear from you.

[Fila]

Actually in regards to the question you had about choosing GPU, do you have that Adrenalin program to try and do the settings I pointed out? If you don't, no need to install it (I generally dislike extra unnecessary software so don't want to make others install them either).

[OG_RuDeE]

Yes I have it on both machines. I can have a look and see which GPU is getting used during the game.

Edit: Adrenaline shows that the dedicated gpu is the only one being used.

[OG_RuDeE]

PS. Is there a key combo I can enter to turn off the creation of these diagnostic logs so I can drive around a bit without these files being created?

[Fila]

Nope. But for tracking purposes they track if its their own fault or not for the slowdown. Doesn't help you visually

I'm curious if turning those settings off does anything and whether the GPU clock stays put.