[`Black]

I have never below 30fps with my system... slowest setting is multiplex + info on & flow view... then it's 30fps with fast scrolling and cpu load goes to 90%. Without backdrops & fanart and flat view I have 40-45% cpu load with fast browsing... Idle fps in every view 60fps (more without vsync/beamsync) and idle load is about 10-25% depending on view.

What thumbnail size do you all use? I use 480x320 for my thumbs as it's enough... in list view & full info I use 1080px poster.jpg for maximum quality.

[Lukas]

I just enabled the Debug logging and played around with all the views a bit.

In Multiplex view my FPS never drops below 30 no matter what I do. CPU load goes up till 75% if I scroll through Movie Multiplex view as fast as possible. When doing nothing the CPU load used by XBMC is around 20%.

In Showcase my FPS is around 20, sometimes even lower. CPU load is about the same as in Multiplex view.

In List view FPS is around 30 when doing nothing, fast scrolling brings it down to 20.

Wall view is the worst. Idle FPS is around 20, fast scrolling between 10 and 15. CPU load goes up to 85%.

I use Black's mod and have Backdrops and Fanart on, all fanart 1920x1080, posters 1000x1500, MediaFlagging and DVD cases on as well so probably the most you can get from Aeon.

EDIT: Ok this is strange/funny: I turned off fanart and backdrops and the performance in Multiplex got a lot worse ?!?!

Ok, I don't wanna brag, but generally data is useless without frame of reference. So for comparison:

Windows, CPU load is 30% on idle, specs are in my signature.


Wall view: 60FPS idle, 57.9FPS scrolling fast (holding down the button), 59.9 scrolling slow (pressing the button repeatedly) (load: 33% - 39.5%)

Multiplex: 60FPS idle, 60FPS scrolling fast, 59.9FPS scrolling slow (load 40% - 52%)

List view: 58FPS idle, 32FPS scrolling fast, 56.8FPS scroling slow (load 39% - 55%)

Showcase: 58FPS idle, 58FPS scrolling fast, 59FPS scolling slow (load 44% - 51%)


I don't use Solid State Disks, no RAID0 setup, just plain SATA 7200RPM. I did defrag my XBMC folder yesterday though.
All my backdrops are 1920x1080, as for thumbs and covers I don't really know coz I download them strauight from tvdb (I don't use nfo files, and have no *.tbn/*.jpg files in the movie/tv-show folders. vsync always on.

[Lukas]

Any of you tried tweaking the nVidia/ATI settings? You could try to set everything to 'Performance' instead of 'Quality'... nVidia lets you specify settings per program as well:

[CapnBry]

It only swaps out textures if there is not enough texture memory for them.

For example I have 1.5gb video memory on my card and I do not think the total textures that aeon uses comes near that.

Oh let me clarify. I didn't mean swapping textures from VRAM to RAM and back. I meant swapping textures as in "the texture currently bound to the context in the 2D target". There is a latency with glBindTexture that comes from the fact that the verticies associated with previous texture need to make it to the rasterizer before the texture can be unbound. This can create pipeline stalls.

If reducing the size of the textures speeds up the rendering though, that generally means you've got a texture publish bottleneck, which should not be the case in your situation because of the large amount of VRAM.

Actually I just did some preliminary bottleneck checks. On my laptop, which has an incredibly shitty GPU with very little dedicated RAM, the Movie Showcase view with 0 items in the library takes ~55ms per frame. I then replaced all large texture data with a 2x2 white square. This reduced the VRAM usage from 25MB to 4MB. How long does it take to draw a frame? ~55ms. It does however load faster. This may be where some folks might be getting their claim that it is faster, when the rendering is still the same speed. Viewing from the files view, which also uses 30 textures weighing in at 4.4MB of memory takes ~33ms per frame.

So clearly this isn't a problem with the size of texture data, unless we're talking about the amount of time it takes to switch between windows. I'd continue the ole standard GPU bottleneck checks but it is almost time for beer!

[CapnBry]

Any of you tried tweaking the nVidia/ATI settings? You could try to set everything to 'Performance' instead of 'Quality'... nVidia lets you specify settings per program as well:

Except XBMC doesn't use AntiAliasing or MipMaps so everything on that page should not make a difference except for Texture Filtering Quality. Also, not pictured but moot would be Anisotropic filtering as the textures are presented isotropically.

[Hitcher]

EDIT: Ok this is strange/funny: I turned off fanart and backdrops and the performance in Multiplex got a lot worse ?!?!

Are you sure you don't have the curtain showing but fanart still on?

I noticed that when using curtains fanart was still being loading but hidden by the curtain and fixed the problem before.

[Montellese]

Are you sure you don't have the curtain showing but fanart still on?

I noticed that when using curtains fanart was still being loading but hidden by the curtain and fixed the problem before.

I first turned off Backdrops and noticed for a split of a second that the fanart is still loading behind the curtain (I'm not using your mod yet because I don't like the new font) and then I turned fanart off as well.

[needtotest]

Just to understand - why should it run 58+ FPS when just standing on the homescreen?
I would drop FPS way down if I should do-nothing-loops. A movie would only require 30 FPS so why should xbmc require more?

/needtotest

[needtotest]

Why are people using vsync? I disabled it on my system - only thing I noticed was a drop in cpu load - aprox from 20 -> 10%

/needtotest

[Lukas]

Except XBMC doesn't use AntiAliasing or MipMaps so everything on that page should not make a difference except for Texture Filtering Quality. Also, not pictured but moot would be Anisotropic filtering as the textures are presented isotropically.

Ok, learning.
Additionallt to Texture Filtering, all the OpenGL settings would still apply, right? Error reporting, extension limit which (by theory) could improve performance of an OpenGL application?

Grtz

Lukas

[Lukas]

Just thinking out loud...

Just to understand - why should it run 58+ FPS when just standing on the homescreen?
I would drop FPS way down if I should do-nothing-loops.

I have no idea about development, but that would mean that every first animation/interaction/navigation would be sluggish? Before XBMC would raise the fps from 0 back to 60?

A movie would only require 30 FPS so why should xbmc require more?

/needtotest

Even with a movie playing, I want smooth OSD animations... 30FPS would be too low?

[`Black]

Just thinking out loud...



I have no idea about development, but that would mean that every first animation/interaction/navigation would be sluggish? Before XBMC would raise the fps from 0 back to 60?

That's it because in the moment the animation start's XBMC would have to raise the fps and that would cause a lag... that's also the reason why people are using vsync... it can reduce big frame drops which can cause lags. In addition to that vsync can eliminate/reduce tearing.

[CapnBry]

Ok, learning.
Additionallt to Texture Filtering, all the OpenGL settings would still apply, right? Error reporting, extension limit which (by theory) could improve performance of an OpenGL application?

I'm not completely familiar with what those settings do, the only nVidia card I have these days is under Linux. Error reporting probably not, because XBMC only queries error state if GL_DEBUGGING is enabled, and it isn't. Extension limit, if it does what I think which is hide extensions, would actually hurt performance as XBMC tries to use the best extensions to increase performance when the card/driver support it.

To help clarify, what slowness is everyone talking about? Like sitting idle is your framerate low, or the amount of time it takes to go from screen to screen, or is everything fine 60fps and then you arrow to the next movie in the list and it chugs?

[CapnBry]

You know what XBMC needs? Friggin "Skin Profiling". Here's some interesting per-frame times for the Aeon's Multiplex view:
Objects_MediaMenuGroup: 4.2ms (yes, 4.2ms to draw a the menu on the top that's faded to 0%)
MultiplexInfoPaneMovies: 4.5ms
The parade of movies: 2.8ms
Furniture_MultiplexMediaFlags: 2.6ms (mostly doing string compares for studio names)

Those are the biggies. So if you turn off media flags and turn on kiosk mode you should see a noticeable jump in FPS if you're strapped due to being CPU bound.

[jmarshall]

CapnBry: I've thought of doing texture combination before, but never really found it a real impact on most skins. How much of the above times is actually due to the number of textures being bound? I'm not sure if combining textures will actually do all that much, due to the way info-rich screens are drawn - thumbs and other dynamic textures will mean that we have to switch from any combined texture to the dynamic texture and back again. I suspect we'll only get an improvement if the render order is changed. Not sure how easy that would be to accomplish. We can chat about this elsewhere (in the dev forums) if you have some further ideas.

Profiling is something else that could be added - it'd be reasonably easy to add a profile flag to the group control that logs on window exit, if skinners think it's something they'd use.

Cheers,
Jonathan