[chief1983]

Well, how much work would it be to a) record everything that happens during a mission for future playback and b) code the support for playing it back?  I have a feeling that a) is easier.

[Cobra]

I remember WMCoolmon edited the main hall with LUA hooks (i'm sure it's the wrong term, but I'm not a coder. ) to play some form of a demo. but that wouldn't affect FPS in the main hall would it?

[taylor]

Well, how much work would it be to a) record everything that happens during a mission for future playback and b) code the support for playing it back?  I have a feeling that a) is easier.

Both already exist, was my point.  originally had code in the source which did that.  The problem is that they never actually finished it.  The demo files were/are massive, easily exceeding 60meg each.  There was no UI for loading them, but it does have the ability to load and play one of the demo files as a default.  The current code also doesn't fully record everything for the demo.  It uses placeholders for things, and doesn't record many other things, just the results of them (for instance: no missiles, just the missile hit).

That is what I had fixed at one point, fully working playback code and mostly working record code.  I did all of this years ago when I was initially running the game through Valgrind, a memory leak detector.  Since running the game through Valgrind is very slow, coding on the demo system was my saving grace.  I was able to record large, and long, demos while playing normally.  Then, while running the game through Valgrind, I would playback one of the demos.  What originally took 30 minutes worth of gameplay normally, would take several hours to playback with Valgrind.  This let me record a complicated mission play, then play it back through the memory leak detector so that I didn't have to sit in front of my computer that entire time.  And that's how I found approximately 80meg worth of memory leaks in 's original code.

The problem is that after I got done with the Valgrind tests I do something with the code changes to the demo system.  I spooled off the code tree to a network drive which I have in storage somewhere, or I simply erased it all.  I plan to redo the code again at some point, but I just don't care to make the time to do it.  Plus, I think that most people would rather have me spending my time on other coding tasks for the foreseeable future.

[Huggybaby]

Plus, I think that most people would rather have me spending my time on other coding tasks for the foreseeable future.

Indeed. If only there were someone to whom you could delegate...

[CP5670]

I think the closest we could get to a benchmark right now is an action cutscene, but it would have to involve larger warships only since anything with fighters will proceed quite differently every time it's run.

And the reason for that is pretty simple.  AF has more of a hit on general processing performance of the GPU, so if you have a fast GPU then you can get away with a really high AF setting.  AA on the other hand will have a significant hit on memory usage.

You actually don't even need a particularly fast GPU. Pretty much everything from the 9700 line and onward takes roughly a 2-4% performance hit from 16x AF. (leaving aside the 7 series cards on HQ mode, but that's a whole separate issue)

By the way, are there any known performance issues with the beams, on the game engine side? To add to what I mentioned earlier, there is one point in one of my missions where there is a substantial and reproducible performance drop when some beams fire for the first time, but they are only using the retail graphics. I can PM you with more information if needed.

[taylor]

By the way, are there any known performance issues with the beams, on the game engine side? To add to what I mentioned earlier, there is one point in one of my missions where there is a substantial and reproducible performance drop when some beams fire for the first time, but they are only using the retail graphics. I can PM you with more information if needed.

There are known performance issues with numerous parts of the code, and it varies as to what those issues are.  Texture handling, collision detection, and things like that have a pretty severe performance hit.

If you just want to test whether it's a memory/loading issue though, I can get you a test build of my current code tree for you try out.  It's not 100% yet, but I have pretty much all of the performance issues in my test cases worked out already.  Currently, it can cut the memory requirements of a mission by about half, and is a bit smarter with several aspects of texture usage.  Just need to get the new code a lot more ALT-TAB resistant before it's actually safe to use by the masses.

If you are interested then let me know and I'll get you a build link.  The changes should show up in 3.6.10, so it won't really be a long wait before everyone gets to benefit from the new code.

[CP5670]

Cool, I'll give it a try and see if it does anything here.

[Herra Tohtori]

I think the closest we could get to a benchmark right now is an action cutscene, but it would have to involve larger warships only since anything with fighters will proceed quite differently every time it's run.

That's why I suggested that one mission from JAD2. It has an action cutscene, and only one fighter that is moving - the player. And, it offers loads of those things that tend to produce the most of the FPS drops in the game: Particle effects and shockwaves in explosions.


Just keep the throttle to zero in the mission beginning and the mission will proceed quite similarly practically every time. Alpha One will turn to face the Disco Installation with dancing CommNodes, then Colossus 3000 will emerge from subspace causing everything to go pop. That will generate perhaps the most graphics-intensive conditions in a FS2 mission that I have ever seen. When the whole screen is filled with particles and/or shockwaves, it becames a good test for effects.

Unfortunately that mission is not that good for testing how much many high-poly HTL ships in view will affect the frame rate. Perhaps someone should make an action cutscene missione of an HTL Aquitaine, HTL Lucifer, HTL Whatever and some CommNodes exploding. However, explosions are as far as I know the "bottleneck" for most setups, so measuring the FPS during intensive explosions would give a rather good idea about system capabilities.

[Bob-san]

Any chance I can try the test build? I am currently having a few performance and stupidity issues; Intel Celeron 2.0ghz, 512mb DDR memory, GF FX5200... my computer is a snail, yet somehow I still manage to run adveffects. If that can boost FPS on my computer... I'd be quite confident it can do that on any computer.

[Ravage]

I'm running a P4 2.53, 768 RAM, and a PNY 6200 AGP 4x with 256 on board.  I don't use the adveffects vps, because I've heard of tremendous slowdowns, but the thing which seem to affect my performance the most are missile lighting effects.  I keep missile lighting effects off and my fps ranges from 18 at the lowest and 60 at the high end.  It usually hovers at the V-Sync of 30 or so though during dogfights I think.

I wonder if there wouldn't be some way to apply a performance value to each effect so that the launcher could provide a total value which could be compared against a composite score of your system.

[taylor]

Any chance I can try the test build?

No.   

The performance improvements aren't done yet, so the code has some inherent instability and so you have to be pretty mindful of what you are doing with it.  The changes will be completed and start showing up in public builds soon enough though, so you're better off waiting for that.

[Huggybaby]

I don't use the adveffects vps, because I've heard of tremendous slowdowns, but the thing which seem to affect my performance the most are missile lighting effects.

Thanks for the observation. I've noticed slowdowns with missile lighting too. There is one scene in particular where, at a long distance, 20 or 30 missiles are launched at some ship and my system slows way down. I think missile lighting is a good effect to turn off, considering you probably can't even tell if it's on most of the time.

You should try adveffects though if you haven't. You can unzip this file in
"C:\Games\FreeSpace2\mediavps\data\effects\" to override the big crawl inducing shockwaves.

[fsi.scsi]

I've noticed that (at least on my system) enabling -specular decreases the framerate an average of 10 fps.   

Alas, no shiny ships...

[castor]

No wonder, R100 starts to be a rather old chip nowadays.

[Taristin]

indeed, I replaced my tnt2 back in... '02...

[Ravage]

Thanks Huggybaby, I'll give it a try.

Before I do though, could you explain to me how this one file is enough to override the adv effects shockwaves?  From reading other posts, I can see that there are varying levels of shockwaves, and medium seems to be the most efficient...

Also, the file path you mentioned would put the file in the default freespace 2 data directory, not the mediavps...shouldn't it be in the media vps?  That's where my table mods are...

Forgive me I'm new to this.

[Turey]

that file should go in "C:\Games\FreeSpace2\mediavps\data\effects\" to override the ones in the mediavps.

[Huggybaby]

Ravage,

Thanks for asking...

Put the shockwave where Turey said to, I'll edit my post.

BTW, that's the same shockwave as in the normal mv_effects file, I just exported it for you. I'm assuming it's medium sized, I don't know of any others.

[Ravage]

Thank you Huggybaby and Turey.

I'm still a little confused though.  How does putting this file into the data directory of the media vps compare to the stuff in this thread?  http://www.hard-light.net/forums/index.php/topic,34618.0.html

Is this single graphics file change the super high end shockwaves to low end, mid, high?

Thanks for the help!

[Huggybaby]

That thread is from 2005!

I ignore those old posts for the sake of my sanity.