666 replies to this topic

[jimmyfingers]

@ Toxie - I have an nVidia GTX Ti on my development system and a nVidia GTX 760 on my mini-cab.  The FPS number drops are consistent in their percentages from one execution to the next on the two tables tested so far as well as consistent when testing betweem both my development and mini-cab system as I normally do when assessing potential issues.  Both systems are Windows 7 32-bit. 

 

I can assure you that the performance drop is an actual thing and if it happens on my system(s) then it will happen on others.  Maybe / likely the difference between your guys ATI graphics if you really don't see a change in the FPS - it's very easy to reproduce for me and I also have a table that fuzzel and I are working on so he'll be able to compare with the exact same table / MOD that I am.  It does not appear to be much different at all in the loss of performance than what was experienced the first time the dynamic flippers were attempted. 

[bent98]

I'd like to add there are tables where I see large discrepancies between win xp and windows 7 from a fps perspective. I had a chance to test monster bash beta and it ran 300 fps on win xp but only runs 60 fps on win 7. Airborne was another table that win xp ran a lot high fps. Not sure if the fps is accurate or not but it was a substantial difference

[mukuste]

Ugh, these last few commits were a pain to merge... changed all the flipper render code that I had changed too, and the formatting changes didn't help.... I think I ended up throwing out most of the changes.... ugh ugh...

I'd like to add there are tables where I see large discrepancies between win xp and windows 7 from a fps perspective. I had a chance to test monster bash beta and it ran 300 fps on win xp but only runs 60 fps on win 7. Airborne was another table that win xp ran a lot high fps. Not sure if the fps is accurate or not but it was a substantial difference

 

I think that's a good indication that the bad DX7 support in modern drivers/OS versions throws us off, and makes me optimistic for the port.

[bent98]

I can't wait for this port

[jimmyfingers]

I'd like to add there are tables where I see large discrepancies between win xp and windows 7 from a fps perspective. I had a chance to test monster bash beta and it ran 300 fps on win xp but only runs 60 fps on win 7. Airborne was another table that win xp ran a lot high fps. Not sure if the fps is accurate or not but it was a substantial difference

I have a dual OS with Windows XP on my mini-cab that I occasionally boot into to assess various things (tables I'm releasing, performance, glitches, etc.). Windows XP has always been better on the FPS but the variance between the two has dropped for me over the last couple years and I don't get nearly the difference in FPS as what I used to nor close to the differences you report with the 300 and 60 example at least as what that percentage looks like from that particular example.

 

I had done quantitative tests and even created a spreadsheet of some tables a couple years back for comparing the two OSes and the gap in the numbers from back then are not nearly what I notice nowadays and with my Windows 7 system tuned well with things like affinity for core placement and a streamlined install with only a total of 35 processes running. I play with nVidia AA forced on and let Windows Desktop Composition stay enabled, the latter having hardly any effect anymore on performance if disabled with the latest VP builds / latest configuration of my system.

 

One way to make Windows 7 more like Windows XP is to change the compatibility options on the VP executable to "Disable Desktop Composition". This will make it closer to XP but you will no longer have the benefit of the tearing fix on things like flippers and some other objects (Windows XP doesn't even have this option). Like I say though, that setting used to be a bigger hit but just doesn't seem to be as much anymore whether from VP, the general demands of the tables now, or more work in the graphics card with using mostly AA when playing. I think switching to always playing (not testing or developing necessarily) with nVidia AA forced on is at least partially responsible with my numbers from what I used to get between the two OSes to now but I also think all the advancements and updates to the VP code levelled the playing field somewhat.  it also seemed that the difference with XP was a little bit like just off the top of the FPS and not necessarily totally a factor aspect (I used to see some tables that were only in the mid 100s in Windows 7 and on Windows XP would be high 200s but as things improved they still seemed to stay only a few hundred FPS away even though they were now both operating up in the several hundreds or even over a thousand FPS).

 

EDIT: Bent98, I'd be curious to see what numbers I get between the two OSes / configuration on my mini-cab for the Monster Bash Beta table you've tested with.  Maybe you or UW can forward me a link?

Edited by jimmyfingers, 26 January 2014 - 02:50 AM.

[toxie]

Ugh, these last few commits were a pain to merge... changed all the flipper render code that I had changed too, and the formatting changes didn't help.... I think I ended up throwing out most of the changes.... ugh ugh...

 

i think in the flipper case you could have mostly just taken the new code by fuzzel as-is, as it already uses vertex buffers.

 

i also just ported the changes in the fixes branch back to svn-trunk.

[fuzzel]

@JF: do you mean the flipper made out of 3d primitives are faster than the build in (now dynamic) standard flippers? If that's the case the requested model manager could be a solution to have both standard flippers or if you like the real 3d ones.

[mukuste]

 

Ugh, these last few commits were a pain to merge... changed all the flipper render code that I had changed too, and the formatting changes didn't help.... I think I ended up throwing out most of the changes.... ugh ugh...

 

i think in the flipper case you could have mostly just taken the new code by fuzzel as-is, as it already uses vertex buffers.

 

i also just ported the changes in the fixes branch back to svn-trunk.

 

 

Yes, I tried to take them, but I changed so much about the whole render system already that it would have been a lot of work.

 

Thanks about the fixes! I'll try to merge that and see if everything went ok.

[toxie]

@JF: do you mean the flipper made out of 3d primitives are faster than the build in (now dynamic) standard flippers? If that's the case the requested model manager could be a solution to have both standard flippers or if you like the real 3d ones.

 

but that wouldn't make any sense, right?

the only difference would be 4 vs 2 rendercalls, and if that matters that much then we are all doomed.. 

Edited by toxie, 26 January 2014 - 09:24 AM.

[fuzzel]

Well the behavior of DX7 and the legacy mode of different drivers are more than strange but I think rendering a 3D primitive is faster than the selfcreated build-in elements. I guess the 3D elements imported from a 3D app are more cache friendly. The standard elements aren't created with fast rendering in mind because everything is precalculated. I really like the way FP handles that, you csand use build-in objects or replace them with 3D models and that could be integrated into VP too. The only problem with 3D models is the collision detection. The current physics engine isn't prepared for such things and adding collision detection on 3D models would be a pain to implement and would mean to replace the physics engine with a new physics library like Bullit or Physx.

[unclewilly]

This is strange. The whole fps.
Bent is getting 300 fps on mb in co.
I have a hard time getting much above 60 in win 7 64.
But at 60 fps the gameplay is still smooth.

The other oddity, is that even if I redo the dynamic models at very low poly I really only gain 2 or 3 fps.

It's making me hesitant to release the table as I don't know if anyone will even be able to play it on a moderate system.

I can only hope that the dx9 port will speed things up.

Can't wait to get back to testing these new revisions but I've been tied up building mb
Thanks for all your hard work devs

[fuzzel]

I recommend to use other tools for measuring fps and not using VP for it (by pressing F11). E.g use fraps or bandicam.

[unclewilly]

I am using frapps.
I get 60 fps on mb.
The highest I've ever gotten on the table is 200. And that is when I made all the models except the bumpers and moving toys static.

I'd be glad to send you the table so you can test with it.
The table is basically finished except for fine tuning the flippers.

[fuzzel]

I'm waiting for it

Edited by fuzzel, 26 January 2014 - 12:43 PM.

[unclewilly]

Pm sent.

[fuzzel]

Thx again. As for the fps: yes this is odd and I scratch my head what is really going on in DX and the different driver implementations. The latest revision has dynamic flippers and everthing runs smooth here on my systems. If I test more and more I come to the conclusion that measuring the fps only is not a good way to test for playability. But we don't have an alternative yet..

[mukuste]

I merged the fixes from SVN trunk, went smoothly, no conflicts. Thanks. The only thing that didn't go through were a few changes to matrix math I did in pin3d.h, but those were only cosmetic, so not a big deal.

 

I too find it very strange that primitives are supposed to render faster than the new dynamic flippers. If that's true then there should be a way to speed up the flipper rendering. From my point of view, we shouldn't put too much effort into this while we have the old renderer. One of the things I'm excited about is that if and when DX9 works, we can use the NVidia PerfHud to really dig into the problems with the render calls.

 

As for the physics engine, I would be really wary of switching it out for a generic one. The physics are universally the one thing that VP is lauded for. FP uses an off-the-shelf physics library, and many people are not entirely happy with its physics.

I recommend to use other tools for measuring fps and not using VP for it (by pressing F11). E.g use fraps or bandicam.

 

Do you see discrepancies between these methods? I ask because I've often noticed that even when VP reports framerates of 60 or more, which should be absolutely sufficient for smooth gameplay, it often still seems to stutter. Is this a case of VP reporting a wrong framerate? Or maybe there are some quirks in its physics/timing code which make gameplay not smooth even at framerates at which it should be. I wonder if there is any insight on this.

[fuzzel]

If you measure the FPS with VP it has an impact on it because VP uses an overlay for the FPS text drawing. I would say that if you measure with VP the real FPS can vary around 10% or was it a difference of 10 FPS? I don't know anymore but there is a difference for sure.

[Pinball999]

If you need another FPS check on MB, don't hesitate to send

[kiwi]

If someone can give me an answer ,I am curious to know why when I press the F11 key the stutter disappears.
Before playing I have to press the F11 key and the activity of the GPU decreases by 20/30%, when stuttering is 99%.
Maybe something happened when I updated the video drivers, but I'm not sure.
I used the uninstall tool, tried using the restore point to install the previous drivers, but the stuttering remains.
To verify this I should format and install everything from scratch and I don't want to do yet.

 

Win XP SP3 and VaporX HD 5770

 

Thanks

 

Max