TES Skyrim 0.245

[Oyama]

Boris
Some report from me, for what its worth :

About IL types, it's incredible how the difference can impact the whole thing, would almost deserve a complete re-balancing to fit. Great

About performance, I'm sorry to conclude the same... It's heavier on it.
I'll do some tests in different locations to provide you with solid statistics, but performance is not good with that version, I'm losing 5 FPS average everywhere, whatever the settings (supersampling on or off, etc...), and whatever the AO types / IL types.
I'll check the stats carefully for a much longer long time now.

[Jafin16]

So, am I the only one that is noticing a performance increase with the latest update then? I know I'm not just imagining it, I'm definitely up by 3-5 fps everywhere that's not already at 60. I guess I could go back to the previous and compare if Boris wants me to double check, but I am certain I've gotten improved performance. And I have double-checked that I have supersampling enabled on AO in case anyone is wondering. Maybe the Hawaii GPU handles this type of calculation better than others? I don't know, I'd better not guess why for risk of looking stupid...

[Oyama]

Jafin16
What are your actual [SSAO-SSIL] settings ?

[ENBSeries]

Gabber
Have you tried latest version with old amd drivers then? Because i'm not sure should i continue looking in to problem by removing part of code and give such builds or not. Try also to disable cloud shadows and skylighting in versions which crashing.

Arkngt
Ssao filtering stats jumps in all 3xx drivers, i remember only forceware 196(or was that 192) where the last with stable results. I'm inable to do anything, it's driver issue, there is no other way to test performance, except to force videocard to do the job. Of course i will not write code which just execute 1000 frames of ssao filtering only to get smooth average statistics. Regarding "16.2 FPS" for ssao, do you ever tried this to compare without profiler enabled? It broke all driver optimization, when turned on. Btw, GUI also cost a lot of performance, if someone wish to measure fps, use integrated fps display feature.


About performance in previous and current versions on amd cards, this may depends from driver and especially from videocard model, probably sli. I just found some strange behavior, when videocard execute slightly modified code much faster or slower, so tried by lots of experiments to find perfect one. What is happening inside gpu in such case, i don't know absolutely, there is no information available for such things and i can only assume similar things as for cpu optimization, but at same time all my previous tests with any dx10-11 cards both amd and nvidia gave no results, no parallel execution happening, nothing at all. I don't know how to make ideal optimization for all gpus then. May be to run several versions of the shader at the beginning at cost of several seconds? But this not precise result, because game have many threads and file system is also stressed at game startup.

[Gabber]

Boris

I had old drivers installed last night, because I wanted to test them, I only reinstalled 13.12 a little bit before you sent me the crash test pack.
The first report I made with the latest .245 was tested with old drivers.
So the flickering issue is appearing on old driver, as well as new one. I don't have cloud shadows enabled, but skylighting was on.

[Arkngt]

Gabber
Arkngt
Ssao filtering stats jumps in all 3xx drivers, i remember only forceware 196(or was that 192) where the last with stable results. I'm inable to do anything, it's driver issue, there is no other way to test performance, except to force videocard to do the job. Of course i will not write code which just execute 1000 frames of ssao filtering only to get smooth average statistics. Regarding "16.2 FPS" for ssao, do you ever tried this to compare without profiler enabled? It broke all driver optimization, when turned on. Btw, GUI also cost a lot of performance, if someone wish to measure fps, use integrated fps display feature.


About performance in previous and current versions on amd cards, this may depends from driver and especially from videocard model, probably sli. I just found some strange behavior, when videocard execute slightly modified code much faster or slower, so tried by lots of experiments to find perfect one. What is happening inside gpu in such case, i don't know absolutely, there is no information available for such things and i can only assume similar things as for cpu optimization, but at same time all my previous tests with any dx10-11 cards both amd and nvidia gave no results, no parallel execution happening, nothing at all. I don't know how to make ideal optimization for all gpus then. May be to run several versions of the shader at the beginning at cost of several seconds? But this not precise result, because game have many threads and file system is also stressed at game startup.

Something must have been up with the particular save load when I got such low fps. When I redid it from a save in the same location there was almost no difference between the latest and the previous version of .245 (c. 23.5 with the previous, c. 23 with the latest - using the FPS counter in both cases). No idea why I got such a discrepancy before as it's exactly the same ini's/versions I've compared (and I'd guess the GUI would take appr. the same amount of resources in both cases), but there you go.

[Oyama]

Boris
I can guess also that SLI doesn't do so much good.... I need to disable it for other tests.
Also, for something I can't explain, SSAO profiler doesn't give me any measure actually, only triangle counts.
I'll do noob talking here, but something's definitely wrong on my side with the interaction between latest optimizations / Nvidia latest WHQL / maybe my specific 770 GPUs / and SLI.

It happens I do have some triangulation artifacts, wich are typical of high temps on GPUs, which I precisely don't have. They're going to 75° on max load, and no other game brings them there, even downsampled to the max with full effects on.
They happen only in exteriors, independantly of any effect enabled or disabled one per one.

This is definitely NOT on your side, I'll speak about it as kind of a bad planetary alignment
And fact is you can only test things on your rig, and can't envision all possible scenarii.

[ENBSeries]

Oyama
Try older drivers, if they not fail with triangles and last version.

Gabber
I don't see any reason for 0.240_3 to crash, but 0.240_2 not. But lets try different code changes to test, files attached

EDIT: file deleted

[Canlocu]

Is the link for the Non-test v0.245 broke? I don't see it on the download page.

[Oyama]

Boris
Got it, will revert.