Performance problem

[lagurus]

From firmware from Aug 21, 2014 (995a4c011dd57031bb9290a40286439c9a05c2ed) is no more
possible to achieve 1920x1080 with 30fps.

Tested with raspivid command:

Code: Select all

raspivid -v -w 1920 -h 1080 -fps 30 -t 60000 -b 1000000 -o /tmp/aa.h264

and also saved video file has aprox. 20fps.

Code: Select all

/opt/vc/bin/mmal_vc_diag mmal-stats

component		port		buffers		fps	delay
ril.video_encode    	0 [in ]rx	638       	20.1	67120
ril.video_encode    	0 [in ]tx	638       	20.1	67131
ril.video_encode    	0 [out]rx	641       	19.6	1080603
ril.video_encode    	0 [out]tx	640       	19.6	1078733
ril.video_render    	0 [in ]rx	640       	19.6	982425
ril.video_render    	0 [in ]tx	640       	19.6	982432
ril.camera          	0 [out]rx	660       	20.1	982426
ril.camera          	0 [out]tx	640       	19.6	982491
ril.camera          	1 [out]rx	658       	20.1	1075419
ril.camera          	1 [out]tx	638       	20.1	67128
ril.camera          	2 [out]rx	0         	 0.0	0
ril.camera          	2 [out]tx	0         	 0.0	0

With firmware from Aug 12, 2014 (3d99bfb69460105c91e688a6031971e641bff354) is it ok.

[6by9]

I'll have a quick check.
Two possibilities come to mind based on the commit text:
- setting the restart_dropped_frames value correctly may have issues if we're running behind (should only affect stills though).
- I've blundered in resizing the image pools for stereoscopic.
There was a followup patch for stereoscopic that I don't think has been released yet. I don't think that would affect this though.

[6by9]

Hmm, I'm not seeing it.

Code: Select all

pi@raspberrypi ~ $ /opt/vc/bin/mmal_vc_diag mmal-stats
component               port            buffers         fps     delay
ril.video_encode        0 [in ]rx       460             29.9    63563
ril.video_encode        0 [in ]tx       460             29.9    63568
ril.video_encode        0 [out]rx       468             28.5    1095226
ril.video_encode        0 [out]tx       468             28.5    1093140
ril.video_render        0 [in ]rx       462             28.2    989122
ril.video_render        0 [in ]tx       462             28.2    989121
ril.camera              0 [out]rx       482             29.3    989122
ril.camera              0 [out]tx       462             28.2    989100
ril.camera              1 [out]rx       481             29.3    1042443
ril.camera              1 [out]tx       461             29.9    63565
ril.camera              2 [out]rx       0                0.0    0
ril.camera              2 [out]tx       0                0.0    0

That was against 025eb5a58e62aca6a1a5f957d6706fe221c8faa4 which was released last night though.
Syncing back to 995a4c011dd57031bb9290a40286439c9a05c2ed also shows no issues on a quick glance.

Is anyone else seeing this? Time is getting very short so I can't spend too much time looking in to this if others haven't confirmed the issue. Please check with 025eb5a5 as the latest firmware.

[lagurus]

Now I have made next tests.

I have used official Raspbian (from 2014-06-20).
Then was made:

Code: Select all

apt-get update
apt-get upgrade
rpi-update

Personally I have 3 RaspiPi + 3 cameras (2 normal and 1 noir).

2 RaspPI have problems with last firmware, doesn't matter which camera was used.
1 RaspPI works fine with last firmware and with all cameras.

Always was used same power.

Note: Scene is quite dark.

Strange behaviour, maybe is still problem on my side, but I don't have any ideas.

Has anyone else same problem?

Thanks.

[AndrewS]

Maybe two of your Pis have slow SD cards which can't keep up with the write-rate?
Are all your Pis similar ages? The same Models (A/B/B+), same revisions?
Have you got the same GPU-memory-split on each Pi?

[lagurus]

Thank you for suggestions.

Maybe two of your Pis have slow SD cards which can't keep up with the write-rate?

I made tests with same SD card with same data,each PI separately. Same results if is used raspivid ... -o /dev/null

Are all your Pis similar ages? The same Models (A/B/B+), same revisions?

All PIs are model B with 512MB memory. Ordered and delivered from 6/2013 till 3/2014.

Have you got the same GPU-memory-split on each Pi?

Same SD card = same GPU split.

As I see, nobody has similar problems, so problem must be on my side.