Is vertex_cache_triangles showing more promise then the other candidates?

I was using vertex_cache_strips one but started encountering hangs on some meshes and mesh errors when testing the new skip list posted.

vertex_cache_triangles is the one being implemented in 2.1.6, it showed a slightly better performance. (vertex_cache_strips was worst)

There are some meshes in better cities (meshes\bettercitiesresources\3dmaps\v7*.nif) that takes a lot of time with 2.1.6 test builds

The test builds also optimize collision, and there are some meshes that fail a test with that. I think it is being looked into, but the FPS problem had priority.

I'm planning on setting up to redo PyFFI from scratch on everything, but that's going to take time. Main reason I want to do this is because of the possibility that some hairs and helmets might have been fubared because of the old skip lists (or in some cases, no skip lists) and I'd wager doing this in a properly cleaned setup would be more useful than simply redoing everything without first undoing any possible damage that's been done from older versions. Not going to use FRAPS though, because I don't want or need it for anything and don't want it introducing its own resource drain on things. The game's FPS counter has been a reliable enough measure so far.

That said, on those things from mods, vertex caching has made those areas perceptibly smoother while I'm in them. I'm not sure of the details of what it all means, but I've seen a lot of meshes go from something like an AVTR of 1.7xx to 1.000 and that seems to be the ideal - getting as close to 1.000 as possible. Every last one of the _far.nif files in RAEVWD for instance hit 1.000 which I'm hoping means those will perform better once I get them into the BSA repacks.

I don't think the collision optimization is doing squat because despite seeing a lot of that getting done, there was negligible improvement that could just be attributed to random chance. My guess is this only helps with meshes that don't already have MOPP type collision. A lot of mods have the wrong type though, so for those it's probably going to make a lot more of a difference.

Ok, so I was right about it. 2.1.5 make perfomance much worser for vanilla meshes. Tests on few different rigs clearly show huge degradation in perfomance.
2.1.6 seems improved in right direction, but I'm wondering for how much vram and main ram consumption is increased with it. From what I've seen optimized files are bigger in size than vanilla. Just 3 fps gain on hi-end system, on low-end it will be not visible or will be ~1 fps, with more vram consumption. That means more stuttering and doubtful overall benefit from using PyFFI on vanilla meshes.

Or, as before, you could be totally wrong and have no idea what you're claiming. Are you seriously going to sit there and try and say everyone who has seen substantial gains from this is wrong or lying in some way?

Look, I havent seen any test results from them or you, even one, and I've seen enough test results (which were always worse than vanilla with 2.1.5) already from other peoples on different rigs. How do you think science is working, they believe in someones words or base their judgement on objective tests?


No, thanks, I've seen enough to make my personal judgement. You should let others to decide as well, without saying that it works good to 99% of peoples, which is lie.

Oh, so you ARE calling us all liars. And of course it's our burden to provide you proof to a claim that's contradictory to everything that's been reported here up until you started making wild claims. Science works by you providing proof that the existing consensus is wrong, and you've not done that except on a computer so ancient I'm surprised you can even run the game at all.

Youre putting your words in my mouth, I havent said that, so youre lying again.


My computer running game fine. Pure vanilla game have 20-40 fps average on low-mid settings in 1024x768 resolution. Besides youve seen the tests of other peoples with low-end and high-end rigs.
Ok, I'm done with it. Don't see much sense in proving anything to close-minded person. Its just useless.

Except you JUST did it AGAIN.



Feeling's mutual really. I don't see the need to prove what's already been proven time and time again to someone who refuses to believe it.

:flamethrower: :gun:
...
...
...
:icecream:

I have pyffied the RAEVWD- mods:
- RAEVWD_Elsweyr_Anequina_V2-20053
- RAEVWD_SI_Edition-20053
- RAEVWD_v1_8-20053

with PyFFI-2.1.6.647f469.vertex_cache_triangles.win32.exe
settings:

found something

installed optimized meshes and loaded game
and loaded the save- file (standing in front of the door of the arcane university, looking to the white gold tower)
after the loading- bar is full screen goes black, music plays short and then ctd

I think the beta version destroies some meshes, with 2.1.5 and re- pyffied RAEVWD meshes I have no problem

You probably should have had --exclude=NiSourceTexture in there. I suspect it changed all of the texture paths so that they're pointing to something like textures/lowres/lowres/ instead of textures/lowres/ like they should be.

Provided of course there's not a bug in the bit that removes the parallax flag. The batch file I just used to fix up the _far.nifs looks like this:

c:\python26\python.exe c:\python26\scripts\niftoaster.py --jobs=1 --noninteractive --only=_far modify_makefarnif --exclude=NiSourceTexture C:\Oblivion\Fars > C:\Oblivion\Fars\log_far.txt

The only significant thing I see in yours is you used opt_cleanfarnif rather than modify_makefarnif. I don't think I've ever run one using opt_cleanfarnif.

BTW, which meshes did it say it was removing parallax flags from?

The difference between

opt_cleanfarnif (pyffi 2.1.6)



and

modify_makefarnif (pyffi 2.1.6)



is modify_makefarnif additionally changes the texture path (disabled with --exclude=NiSourceTexture) and optimizes the meshes


> BTW, which meshes did it say it was removing parallax flags from?
list follows later
- testcloudrulertemp_far.nif
- some skingrad meshes


Edit: PyFFI ver.

Huh? Where?

Tests are tests and words are still words. Anyway, its better for us to stop talking about whos wrong and whos right because it wont help. I prefer to look at current tests. They are quite objective.

@Baphometal: I suppose I'll find out soon enough then if something got fubared, because I haven't yet packed the BSA with the updated _far.nif meshes.

@Rancen: You do that. I'll stick by tests performed in the past by numerous users before you who all report more or less the same thing: Increased performance. I consider those tests to be just as objective even though they weren't tacked on to some sourceforge entry to wave around as proof of the deed.

That's probably because the command is a new one that has only been added to the v2.1.6 beta version of PyFFI.

Ah. That would explain it then.

Do you know if 2.1.6 (vertex cache triangles build) automatically applies that optimization when right-clicking on folders?

Thanks

What are the chances of incorporating "LOD' files into the optimization process? I have no clue about the development of PyFFI, so I don't know if it's even possible.

If you mean the landscape LOD files, they already get optimized if present, and the vanilla ones have quite a few verts dropped from them.

Rather lolsome really.

--------

Anyway, I'm not sure whether or not to PyFFi my stuff. I'm not sure I'd find the time. Plus now there are these claims of loss of performance as well. What's all this about? Is it something to do with 2.1.5, or just some wierd one offs? Should I get 2.1.5 or 2.1.6? Sorry about all the questions, but I know it's going to take a while and I don't want to have to do it just to find I used the wrong version or I lose FPS (although I don't understand how that could possibly happen. Something to do with vRAM?).

Weird one-offs. 2.1.5 is proven solid from God knows how many people who have used it and seen performance gains. 2.1.6 should improve on that but is still in beta, so if you want to go that route just be aware of that. There are also 4 versions of 2.1.6 currently available and it appears the one for vertex triangles works best.

You mean http://www.sendspace.com/file/w1km3g? I'm up for trying a beta.

A little off topic, but has anyone ever tried this on Morrowind? I'm wondering if I could get a few free FPS there.

The performance loss comes from the poor vertex caching in nifs after they are prffied with 2.1.5, which is more noticeable on complex objects like NPCs, however structural figures normally don't have so many vertices and therefore are affected less, most decent computers shouldn't see the performance loss in 2.1.5, until theres a good number of npc's on screen (varies with computer hardware & driver optimizations). Pyffi 2.1.6 (vertex cache triangles build) addresses those issues.

I still optimizing my meshes, (old computer :brokencomputer: ) so I don't any numbers for comparison.

Yes, that one. I JUST got done finally using it on my repack BSA, so I'll be able to prod it for performance numbers soon, though they won't be meticulously recorded in scientific format



Not that I'm aware, but I don't recall Morrowind being terribly resource demanding when I played it.

Indeed... Youre missed something there I guess...


They are even "weird" because they're reported by different peoples? Show your 2.1.5 testing results in wilderness, what are you afraid of.


Arguable.