Good Performance Hunting

[Bealdwine]

@z929669 - A comparisson with what you have found using DDSopt would be really useful.

 

I tried the NIF tab in SMCO and had no files processed, not sure why yet, as you say the documentation is poor for both that and TO which is a shame.

I'll review the relevant forum pages to see if there is more info available there.

[stoppingby4now]

Been following this thread, but I'm more interested in what kind of quality loss there is. I have enough VRAM, so in no real rush to need to compress textures just yet.

[z929669]

What I am currently testing is optimization effects on FPS/VRAM rather than reduction (using vanilla textures as previously). I will look at reduction effects on FPS/VRAM once I install full hi/lo STEP 2.1

[z929669]

EDIT: The missing contents of this post are being added to the DDSopt Guide now on the Wiki .

---

I found an interesting result with regard to fir trees and foliage in general. More specifically, mid-long range foliage in general: DDSopt corrects mip maps that are not scaling properly in the vanilla textures. Notice that in the Riverwood images (above) and in the following, the mod-long-distance trees are bushier and softer in the standard textures and more crisp and bristly in the DDSopt'ed versions. I am guessing that this is a correction, because the tree in the foreground of the following images is identical, but only in the DDSopt'ed version do the mid-long firs match the close up. Similarly, I like the DDSopt'ed stonework in Windhelm better than the standard (see above).

 

 

• Trees
• Trees-DDSopt

In general, DDSopt'ed textures look comparable, and in many cases better than standard. Additionally, there is a respectable performance boost in terms of VRAM using these settings (which are what I used in the processes described above):

---

Performance Change Summary (Standard Vanilla Textures --> DDSopt'ed Vanilla Textures)

 

 

• 3.3% VRAM increase (Hi rig) UPDATE--> Adding Skyrim HD1k to the mix boosts this up to 5.0% VRAM increase
• 5.0% VRAM increase (Lo rig)
• 0% net FPS boost (Hi & Lo rigs)
• 0.8% GPU Load decrease (Hi rig)
• 1.2% GPU Load increase (Lo rig)

I venture to guess that these performance boosts will accumulate with application to other less-than-optimized texture mods used in STEP.

[stoppingby4now]

The settings at the end, are those the settings you used for your results? I'm seeing on average, 2% improvement to GPU load, and ~3.6% change in memory usage. Is that noticeable in game play? It is also interesting in regards to the textures in Windhelm, where some details appear to pop out in the rocky areas, and the base textures look like those areas are smudged.

 

Are you planning on running any tests on texture packs? Considering that a lot of Bethesda's textures will be over-ridden, it would make for an interesting test case.

[z929669]

Actually, with respect to GPU Load on HD textures, DDSopt'ed textures cost just a tiny fraction more than non-DDSopt'ed. (less that 1%), and this likely does not translate in to any detectable FPS loss even on a low-mid system. The real benefit is in terms of VRAM savings (about 2.8% on average for HD textures) and mip-map correction as well as texture optimization. (this I think relates to the "smudge" correction in Windhelm and fir trees in Riverwood).

 

I will definitely test again on all STEP 2.1 once installed. Just wanted to get a baseline using vanilla and a very clean setup. Would love to see how these differences perform on a lower-end system. I have a P4 with ATI 4770-1Gb card and 2 Gb RAM that I could test on, but that could take some time to set up ;)

 

EDIT: forgot to answer s4n's questions explicitly....

 

1. I noticed no diff in performance during game play, but I also have a pretty juiced system, and if you look at my FPS, I am likely running well withing my FPS cap. Would need to use a low-mid end system to test for sure, but VRAM is also within max, so this would really only increase the ceiling of texture load for 1 Gb cards.

 

2. Settings I used are same as I show at bottom. Added clarification to that post.

[stoppingby4now]

Great work! Can't wait to see more results.

[z929669]

OK, I set up my old P4 box. Here are the REAL specs:

 

Intel P4 Extreme 3.2 GHz (Dell XPS-Gen2)

2 Gb RAM

ATI 4650-1Gb (AGP) - 1024 X 768

RAID0 (fairly slow ex, since drives are 9+ years old [WD Raptors @7200)

 

Needless to say that Bealwine's and Monty's boxes are beasts in comparison.... Only getting about 15 FPS in pure vanilla with low-med preset config.

 

Will optimize config and test diffs using this old lady of a rig.

[MontyMM]

Excellent work - thanks for the effort. Apart from anything else, it will be worthwhile to establish the best practices for this. I have seen a problem in the WATER mod thread, where textures mangled by 'Optimizer Textures' appear to cause unexpected artifacts.

[z929669]

  Quote

Excellent work - thanks for the effort. Apart from anything else, it will be worthwhile to establish the best practices for this. I have seen a problem in the WATER mod thread, where textures mangled by 'Optimizer Textures' appear to cause unexpected artifacts.

 

 

Water textures use normal tangent-space maps (I think), and these must be treated specially. DDSopt has provisions for this, but my guess is that OT and STMC do not.

[Bealdwine]

  Quote

Excellent work - thanks for the effort.

I heartily second that zMan, especially now you are looking at the effects on lower end rigs as well :) (even though you really have no personal need to, that's at least as impressive as the 'selfless' work done by modders). I'm looking forward to implimenting optimizations based on your dedicated work once STEP 2.1 goes live.

 

I tried out a d3d9.dll from 'Gigantic Skyrim FPS Performance Patch' today and may have clawed back 2-3 fps but it is so hard to tell :P Unfortunately it seems to be an nVidia specific improvement for older cards, but responces on the forum are extremely mixed and confusing.

 

Obviously it has the usual problem in being superficially incompatible with ENB series & FXAA mods - nothing a rename and file [proxy] wouldn't cure.

[MadWizard25]

  Quote

OK, I set up my old P4 box. Here are the REAL specs:

 

Intel P4 Extreme 3.2 GHz (Dell XPS-Gen2)

2 Gb RAM

ATI 4650-1Gb (AGP) - 1024 X 768

RAID0 (fairly slow ex, since drives are 9+ years old [WD Raptors @7200)

 

Needless to say that Bealwine's and Monty's boxes are beats in comparison.... Only getting about 15 FPS in pure vanilla with low-med preset config.

 

Will optimize config and test diffs using this old lady of a rig.

 

 

Sounds like an epic undertaking, good luck!

[z929669]

The worst part is waiting for saves to load and installing/uninstalling the compressed Wrye Bash optimized vanilla texture packs. Literally takes about 5x-10x longer on the older rig than my current one

[frihyland]

  Quote

The worst part is waiting for saves to load and installing/uninstalling the compressed Wrye Bash optimized vanilla texture packs. Literally takes about 5x-10x longer on the older rig than my current one

 

 

It will go much faster if you use a non-solid compression flag, you can set that up in bash and your favorite compression program.

[z929669]

  Quote

 

  Quote

The worst part is waiting for saves to load and installing/uninstalling the compressed Wrye Bash optimized vanilla texture packs. Literally takes about 5x-10x longer on the older rig than my current one

 

 

It will go much faster if you use a non-solid compression flag, you can set that up in bash and your favorite compression program.

 

Do tell. I know about setting this in 7z gui, but not in bash and the 7z exe that it uses.