TES Skyrim 0.195

[arsil]

[ADAPTIVEQUALITY] category of enblocal.ini is for reducing quality of heavy effects when frame rate is lower than desired.

Great job! I was going to suggest such a thing, but you anticipated me:)

I have got a suggestion: Could you pass something like "ADAPTIVEQUALITY_FACTOR" to external effects (shaders)? We could for example modify number of samples in DOF, or other things that affects performance.
It would be great if enb (and external effects) could maintain minimum frame rate.

[CruNcher]

Boris

It applies to every game when you are "short" on hardware (mainly because VRAM or RAM). With nearly every game, it's stuttering and/or FPS decrease.
But in Skyrim, whose engine is unstable, it's worse and can lead to CTDs and freezes too. But maybe this applied only before ENB 0.193

Would agree with KuroTenshi experienced such freezes on fast camera reactions before all the optimizations

current revision i experienced CTDs with ReduceSystemMemoryUsage=False the default setting for it changing this and i had less CTDs to none for a longer period.

[MEMORY]
ExpandSystemMemoryX64=true
ReduceSystemMemoryUsage=true
DisableDriverMemoryManager=true
EnableUnsafeMemoryHacks=false

the thing is i reached only the around 3gb mark though overall allocated GPU System Memory was very high despite of the Dedicated Memory

i was almost sure this would crash but it didn't




i also have only 1 GB of VRAM so it was close @ the edge but stable much much longer anyways

[ENBSeries]

arsil
Adaptive quality of effects like depth of field means depth of field only, because no other external effects yet made which require optimization. But i don't think that authors will bother with adaptive parameter, code mostly written to static computations and can be easily broken. Amount of passes is not under control, i can only make quality option for depth of field and then toggle different techniques.


Version updated again, dirty hack parameter separated from reducesystemmemoryusage

[arsil]

arsil
But i don't think that authors will bother with adaptive parameter, code mostly written to static computations and can be easily broken.

I would:) I saw a few DOF effect files and some of them have quality #defines, that translate directly to number of samples taken. It is very easy to change the code to take different number of samples dependent on desired performance.

[ENBSeries]

Show me that code

[CruNcher]

Boris whatever you have done know awesome that was the best latency result i ever experienced

i used the same

[MEMORY]
ExpandSystemMemoryX64=true
ReduceSystemMemoryUsage=true
DisableDriverMemoryManager=true
EnableUnsafeMemoryHacks=false

though i had a CTD but the overal latency was really the best i yet experienced especially in the problematic interior rendering parts of Helgen Keep and even with the counters pooling in the background amazing result

@ the almost exact same position you see above it ctd but i didn't reached that 3 GB (that was stable before) it was lower that last pooling before the CTD showed 2.76 GB then CTD and the last counter pooling result was 832 MB



unfortunately the GPU states die when the processed spyed on dies unexpectedly

i will try with

[MEMORY]
ExpandSystemMemoryX64=true
ReduceSystemMemoryUsage=false
DisableDriverMemoryManager=true
EnableUnsafeMemoryHacks=false

now

[Yersinia79]

Got one ctd too with that latest, looked like it did on first cell load after starting to move. but not 100% sure :>
Could not repeat it thou,atleast not yet :?

[ENBSeries]

Don't know what you guys trying to achieve, because i simply split ReduceSystemMemoryUsage in two vars: ReduceSystemMemoryUsage and EnableUnsafeMemoryHacks. When both active, code works the same as in 0.193.

[arsil]

Show me that code

You mean code that could easily benefit from dynamic performance adaptation?

For example K ENB DOF (have a bit older version at hand, tomorrow I will check the newer one):
#define DEPTH_OF_FIELD_QUALITY 3
and then
float discRadius=blurAmount * float(DEPTH_OF_FIELD_QUALITY);
...
int dofTaps=DEPTH_OF_FIELD_QUALITY * (DEPTH_OF_FIELD_QUALITY + 1) * POLYGON_NUM / 2.0;
...
sampleOffset*=sampleCycleCounter / float(DEPTH_OF_FIELD_QUALITY);

Changing number of samples taken have quite a bit impact on performance and this code can be made easily "dynamic".

[Yersinia79]

Boris
Just trying to help,maybe it just was something I had on (like FRAPS). Forget myself sometimes and have lots stuff running. Either way it still have not happen again :>