It did, it was the same setup and saves as vanilla meshes. FPS was measured with console open to render a static scene.

Right, but I think this is the issue that has been discussed the most - NPCs seem to suffer. What I've been getting annoyed about is the blanket claim that because a few NPCs spoiled the whole thing that the entire process was worthless and bad. There's no disputing the many people who have reported, and continue to report, respectable gains from optimizing the game's meshes.

Looks like you just dont want to see it at all. Look down a bit, on second test WITHOUT npcs. Its SLOWER on 3.422 fps than with vanilla meshes.

* Vanilla (save 2: Market District, 0 NPCs on screen): - from the link.
29.917, 29.950, 29.917

Pyffi 2.1.5:
save 2 (0 NPC in scene)
26.433, 26.567, 26.517 (-3.422 fps)

The tests I have done show that optimizing vanilla with 2.1.5 does decrease performance with (-7.722 fps) or without (-3.422 fps) NPCs in the scene. (when I say without I only mean visible, but the game probably still access the mesh for obscured NPCs)

I have not done any tests for non-vanilla meshes, but have the impression that 2.1.5. increase their performance.

I'd be more inclined to think you don't want to see it or that you think your results on an ancient PC should be blanket applicable to everyone.

3 fps is within the statistical range for random game burps to have caused that.

7 is an indication something isn't right, but only provides statistical proof that NPCs might not be getting optimized as well as they could.

You're not going to convince me with invalid results that what *I* and *OTHERS* have seen improve is nothing but placebo, because that's simply not true.

Bottom line: If you don't like what it did, don't use it. That's pretty much the mantra of mod users, yes? Why try and rain on everyone else's parade because your particular results on an old PC didn't come out as planned?

Anyone who knows jack all about testing should have a pretty good handle on the fact that it would take more than one test, by more than one tester, on more than one rig.

It seems you have provided the data that your tests on your rig are not working well with PyFFI. I'm personally not convinced. But I just want to assure you that it is well understood what data you are presenting. It is just not matched by experience and the data of others.

I did not test with numbers - I tested with just how the game plays and of the many many 'fixes' for this game this is one of 4 that really made a huge difference. The game that was sluggish and slow at times became snappy and quick. I feel no need to test it because it was never a doubt in my mind that it worked. You seemed to doubt it and with good reason. Results on an older system though are not going to be that much of a concern for those of us with newer hardware.

My vanilla meshes I think were done with 2.15 but I could be wrong it could have been 2.14

I agree, unfortunately when amorilia wanted help in testing the cache optimization, almost nobody reported results.

They arent rendered by videocard anyway.


Mine result, H2Odk result (will be great to see test results on mid-end and hi-end pcs as well), guy with hi-end pc from tracker's link. Only my particular result? And without NPCs fps drops too so its not a NPC meshes only issue with 2.1.5. 2-3 fps is enough for test to show thats something wrong with optimization when with vanilla its always higher on those 2-3 fps with same testing conditions. It can work without problems for some, but totally ignore that it degrades performance for others isnt nice.

I just hope that it will be fixed in 2.1.6.

I would have been happy to, but the timing of when he wanted the testing done was bad for me. The vast majority of my stuff is already optimized with 2.1.4, I have yet to even get to hitting everything with 2.1.5, and have simply decided to skip ahead to 2.1.6 after some quick testing on my rig produced a small gain, perhaps within the margin of error, but hey. I'll take anything I can get, and it looks like the triangle caching method eeks out just a hair more than the others. I'll be able to comment more once I get around to doing my custom repacked BSA with the bulk of my install in it.

As an FYI - your updated regex for the skiplists seems to be doing a much better job at spotting things it shouldn't be touching.

(This was @H2Odk btw)

@Rancen: Sorry, but I'm standing firm on my assertion that testing on such a low end system has any validity in general. Especially since you seem to still be trying to make it into a blanket proof that the process isn't worth it. That simply is not the case for the grand majority of people who will be using this to get better performance. The bottom line here is that the game is a bag of inefficiencies, and PyFFI removes one of those inefficiencies.

I just want to clarify that I do not see my tests as proof of anything. They are just results that I got on that PC with those saves.

I really wouldn't feel comfortable making a recommendation whether or not to optimize vanilla meshes with 2.1.5., people really should test it for themselves.

I think its just better to wait till I get new PC and check results again with it. Just conversion on this dinosaur rig takes really long time.


Can you test 2.1.5 with your other gear as well? Like you done it in tracker. I will appreciate this.

I don't understand the discussion here, are you guys saying that using optimized meshes on a low end system doesn't do much? I find that very hard to believe, if anything the low end people such as myself should benefit the most from it, way more then the high end people. How would the high end people even notice a difference?

Either way, I am in the process of pyffing my vanilla meshes, been going since 9PM last night and is now 12:32 PM the next day :brokencomputer: about half done I'd say. I will personally see myself if it was worth it or not.

Because the high end computer would waste less resources per cycle and push out more performance margin then the low end system, its all relative.

It's also entirely possible that the methods of optimization that work for 99% of people are things that ancient hardware isn't equipped to deal with properly and ends up making things worse because of it.

Highly unlikely. There was a perfomance drop even on hi-end system. I think all the problem is in vertex optimization which in vanilla meshes is slightly better than Pyffi 2.1.5 can do.

Sorry, but that I'll dispute until the end of time. Folks on medium and high end systems alike consistently report good results, better frame rates, etc when using PyFFI. My own experience directly contradicts any claim of there being a drop on a high end rig.

Question: I haven't read through all this, but are you guys repacking the meshes into a bsa? I understand that to see any performance improvements you have to repack the meshes into an uncompressed bsa once you're done...

Will you deny the words of PyFFi creator as well?

http://sourceforge.net/tracker/index.php?func=detail&aid=2994654&group_id=199269&atid=968813

And not all gear, hi-end or not, work exactly the same. Maybe Radeon cards are weaker to such optimizations than Geforce. I have Radeon, H2Odk tested it with Radeon card, guy with hi-end pc has Radeon also...


Yes, should be packed into bsa without compression.

I will, because the claim lacks merit when the overwhelming number of people who have done this report gains, not losses. Being a programmer does not make one a god.



Then I guess you'll twist yourself into a fit when I remind you *I* use a Radeon and all testing I've done with PyFFI has been on a Radeon. PyFFI didn't become well known until long after I'd stopped using my old nVidia 8800GTX.

I didn't tried Pyffi since may 2010. At that time, I also encountered fps hit (15-20) with my system (core i7 860, ati 5850) on the same save spot in the initial dungeon.

But I am not sure I can relate it to Pyffi. I had much trouble with this 5850 and I probably did something wrong with my pyffi (I am not an expert computer user).

I did the tests on my game rig.
Core i7-930 @2.93GHz
6 GB ram
ATi Radeon HD 5850 1GB @1920x1200

Opened console right after the save loaded.
Used Fraps 60sec avgfps.
5 runs for each test
Restarted oblivion each time.

2.1.5: did NOT optimize collision
test build: http://http://www.sendspace.com/file/w1km3g

scene 1, market district, 12 NPC             run 1      run 2      run 3      run 4      run 5vanilla      50.800     50.933     50.367     50.767     50.9002.1.5        31.317     31.683     31.700     31.483     31.550test build   52.667     53.250     52.983     53.017     52.600scene 2, market district, 0 NPC             run 1      run 2      run 3      run 4      run 5vanilla     109.250    109.783    109.767    109.467    109.9672.1.5        78.967     78.500     79.050     79.167     78.250test build  110.283    111.350    110.250    110.517    111.450

I would like for someone who has a increase with 2.1.5 to do the same tests. While I don't know much about meshes, my concern is that people who has increased performance with 2.1.5. might have decreased performance with the test build (2.1.6.).

As of last night I finished pyffing all vanilla, DLC and SI meshes. I got rid of Operation Optimization in favor of using the pyffied meshes. But id like to know whether Operation Optimization overwrites any UOP meshes so i can redownload that and add them back.

Yes it does. It overwrites several. Mostly daedric shrines.

I dont think id see much of an improvement on my computer from using pyffi.


------------------
System Information
------------------
Time of this report: 10/17/2010, 12:11:10
Machine name: HOMEUSER-PC
Operating System: Windows 7 Home Premium 64-bit (6.1, Build 7600) (7600.win7_gdr.100618-1621)
Language: English (Regional Setting: English)
System Manufacturer: Gateway
System Model: Unknow
BIOS: )Phoenix - Award WorkstationBIOS v6.00PG
Processor: AMD Athlon™ 64 X2 Dual Core Processor 4000+ (2 CPUs), ~2.1GHz
Memory: 2048MB RAM
Available OS Memory: 2046MB RAM
Page File: 1194MB used, 2898MB available
Windows Dir: C:\Windows
DirectX Version: DirectX 11


Card name: Radeon X1600/X1650 Series
512mb Ati X1600 pro
1920 x 1080(p) (60.000Hz)

Bah, just great