[ATI] Flickering, Tearing in 1080p24, smooth as silk with rendering stats enabled. dshowhelper?? (1 Viewer)
- Thread starter te3hpurp
- Start date
Croatia
EVR uses DWM which is AERO. Aero switched of and EVR can/will revert to software instead of hardware render (gpu dependent) giving the realy poor performance and poor render time.
The link for the 1.2.3 dll is in the first page/post of the link for the 1.3 dll.
The thread is a monster read but the first post has the dll downloads.
Having tested the MP1.3 675 version it has given dramatic improvemnts to an HD2400pro pci and an Onboard HD4200 that i have for test perposes.
Your render time is inline with the gpu power you have, the above dll may help in improving this ( i had reductions from 13+ms down to 5.5 ms)
The STATS though can raise the ms and frame drops by just being active/displayed. For testing over long time period enable stats then hide them; view them again after 5-10 mnutes.
The link for the 1.2.3 dll is in the first page/post of the link for the 1.3 dll.
The thread is a monster read but the first post has the dll downloads.
Having tested the MP1.3 675 version it has given dramatic improvemnts to an HD2400pro pci and an Onboard HD4200 that i have for test perposes.
Your render time is inline with the gpu power you have, the above dll may help in improving this ( i had reductions from 13+ms down to 5.5 ms)
The STATS though can raise the ms and frame drops by just being active/displayed. For testing over long time period enable stats then hide them; view them again after 5-10 mnutes.
Last edited:
This reminded me of the DWM experience I've had. It seems to be common wisdom that the above is true, but I have tested a lot both ways and find that its not the case with my system. In Windows 7, it's easy to switch back and forth from an Aero theme while MP continues to run out to the HDMI attached TV.
I have tried both Aero and non-Aero several times in the last 6 months, and a non-Aero theme runs MP better. I see the CPU time for the DWM process running at 2-3% for Aero and 0% for non-Aero, while the TV shows dropped frames under Aero and not under non-Aero. This is using latest dshowhelper (675?) in MP1.3.0a and using GPU rendering in the LAV video setup.
So I guess it depends on the relative loading of the GPU and the CPU in a system whether Aero is better or not.
Unusual and counter to anything I have seen with any of the test setups I have.
I only see render time dramaticaly increase and droped frames also in non-aero.
I am intrigued if you have a commesurate increase in cpu usage in non-aero v aero.
Are you using DXVA native or CB in LAV and which ATI CATS ?
With your GPU and your powerful cpu maybe the cpu assists are better than the pure gpu when rendering. Interesting!
I only see render time dramaticaly increase and droped frames also in non-aero.
I am intrigued if you have a commesurate increase in cpu usage in non-aero v aero.
Are you using DXVA native or CB in LAV and which ATI CATS ?
With your GPU and your powerful cpu maybe the cpu assists are better than the pure gpu when rendering. Interesting!
DXVA2 Native
Catalyst version 12.6
I looked for CPU increase in any process -- none noticed. Nothing else is using much CPU and MP is 39% to 50%, usually upper 40's when watching TV.
One thing I forgot to mention is that I reduced the PC's monitor's color depth to 16 bits in non-Aero. This helped months ago, but I'm not sure it's needed now. I didn't do that in the most recent tests. I just put it back to 16 though, since the amount of GPU/CPU it saves is surprising.
I have been "fine tuning" this machine for months, so I can get smooth playback, GPU deinterlacing, and acceptable general purpose use. The dshowhelper's addition of buffering is what did the most good. Prior to that I was not able to achieve GPU deinterlacing -- I had to turn deinterlacing off altogether in LAV by checking "treat as progressive." This resulted in smooth 1080i/60 but added comb artifacts with fast motion.
I run Thuderbird always, IE10 several times a day, and some OpenOffice. These don't badly interfere with wife's TV watching, but Chrome is a real pig when it starts up, causing pauses and stutters, so I quit using it. I have comskp set to "play nice" in its .ini file, and a single instance is no problem, 2 instances causes some visible drops in TV, and 3 instances interferes a little with movie watching. I don't run any anti-virus at all. I run MP at High, TVserver at realtime, and MySQL at realtime.
I don't have any CPU usage numbers for those heavy scenarios -- running Task Manager causes drops all by itself.
EDIT: Deinterlacing set to Adaptive.
Catalyst version 12.6
I looked for CPU increase in any process -- none noticed. Nothing else is using much CPU and MP is 39% to 50%, usually upper 40's when watching TV.
One thing I forgot to mention is that I reduced the PC's monitor's color depth to 16 bits in non-Aero. This helped months ago, but I'm not sure it's needed now. I didn't do that in the most recent tests. I just put it back to 16 though, since the amount of GPU/CPU it saves is surprising.
I have been "fine tuning" this machine for months, so I can get smooth playback, GPU deinterlacing, and acceptable general purpose use. The dshowhelper's addition of buffering is what did the most good. Prior to that I was not able to achieve GPU deinterlacing -- I had to turn deinterlacing off altogether in LAV by checking "treat as progressive." This resulted in smooth 1080i/60 but added comb artifacts with fast motion.
I run Thuderbird always, IE10 several times a day, and some OpenOffice. These don't badly interfere with wife's TV watching, but Chrome is a real pig when it starts up, causing pauses and stutters, so I quit using it. I have comskp set to "play nice" in its .ini file, and a single instance is no problem, 2 instances causes some visible drops in TV, and 3 instances interferes a little with movie watching. I don't run any anti-virus at all. I run MP at High, TVserver at realtime, and MySQL at realtime.
I don't have any CPU usage numbers for those heavy scenarios -- running Task Manager causes drops all by itself.
EDIT: Deinterlacing set to Adaptive.
Last edited:
I have just done some tests on my HD4200 set up very interesting results.
Using 675 dll 1.3a latest and CATS 12.6. LAV DXVA native. 32 bit colour output
When Aero is ON the best results are with EnableDWMQueuedMode=1, no droped frames amd 5ms render; EnableDWMQueuedMode=0 and 17+ms and dropped frames.
When Aero is OFF the best results are with EnableDWMQueuedMode=0, small drop in frames 17ms render; EnableDWMQueuedMode=1 and 60+ms and serious dropped frames.
Actual visual experience I prefer Aero ON . Test material was BLU RAY. 32 bit colour output.
When I selected 16bit colour output in CATS Aero was automaticaly disabled and NOT available. EnableDWMQueuedMode=0 or =1 made no difference; smalll drop in frames 17+ms render.
Using 675 dll 1.3a latest and CATS 12.6. LAV DXVA native. 32 bit colour output
When Aero is ON the best results are with EnableDWMQueuedMode=1, no droped frames amd 5ms render; EnableDWMQueuedMode=0 and 17+ms and dropped frames.
When Aero is OFF the best results are with EnableDWMQueuedMode=0, small drop in frames 17ms render; EnableDWMQueuedMode=1 and 60+ms and serious dropped frames.
Actual visual experience I prefer Aero ON . Test material was BLU RAY. 32 bit colour output.
When I selected 16bit colour output in CATS Aero was automaticaly disabled and NOT available. EnableDWMQueuedMode=0 or =1 made no difference; smalll drop in frames 17+ms render.
Last edited:
Could you test with MPAR if the rendering time changes? This is mainly since MPAR will move the target v-sync. Rendering time just means the time that it took for the Present() call, which could be waiting the v-sync in addition to the actual drawing.
Here's what I see:
1080i signal
(Stall shown as 14ms for all cases, I think, but I forgot to check each case. I assume this is a fixed value for 1080i.)
Enabling MPAR caused a near collapse of MP. CPU up to 80%. Video stalled completely from time to time, with pixelation, sound problems, etc. Otherwise, the setup was the last one above.
I then tried a more standard MPAR setup: Aero, no buffering, 32-bit monitor (required).
17.5-19ms render, unstable, some drops reported, video OK but "off" somehow. Can't say what was wrong -- it wasn't jerky, but seemed wrong. Program being watched wasn't very active, so hard to tell.
That's all I checked for MPAR.
1080i signal
(Stall shown as 14ms for all cases, I think, but I forgot to check each case. I assume this is a fixed value for 1080i.)
Code:
Without MPAR:
Aero on, 32-bit PC monitor.
Buffering: off, 23-25ms unstable, many drops, video unacceptable.
Buffering: 7, 17.5-19.5ms unstable, many drops reported, video OK
Aero off, 16-bit PC monitor.
Buffering: off, 13.5ms flat, no drops
(This is a surprise result here. Some other change must be helping.)
Buffering: 7, 13.5ms flat, no drops
(This is how I have been running.)Enabling MPAR caused a near collapse of MP. CPU up to 80%. Video stalled completely from time to time, with pixelation, sound problems, etc. Otherwise, the setup was the last one above.
I then tried a more standard MPAR setup: Aero, no buffering, 32-bit monitor (required).
17.5-19ms render, unstable, some drops reported, video OK but "off" somehow. Can't say what was wrong -- it wasn't jerky, but seemed wrong. Program being watched wasn't very active, so hard to tell.
That's all I checked for MPAR.
Last edited:
I'd be interested in seeing the log for the Aero OFF + EnableDWMQueuedMode=1 situation - in case there is a bug in the DWM checking code. It shouldn't be enabling queued mode if Aero (DWM) is turned off.
Also you must restart MP whenever you change the EnableDWMQueuedMode registry value.
Tony