Intel QuickSync

bdg2

Member
Is your J:\ device a local hard drive, external or perhaps network?

I took a 1920x1080 BBC-3 video here, and recoded it on my I7-3770:

1) using QS @ 2.5 Mbps: 99 FPS output, low CPU usage.
2) using software only @ 2.5 Mbps: 45 fps, high CPU usage. This looks similar to your system.
3) Then did a deinterlace / resize to 720p (iphone 2 profile) using QS.: 91 fps, high CPU usage.

This just means we need to dig deeper. What version of the Intel Driver are you running? You can find it using the device manager. As an FYI, I'm looking at adding that information to our QSync information box. It's just takes a lot of work to get that one small piece of info.

I just saw something else. If you don't force a recode by changing the dimensions, cropping, etc, the smart encoder will be used, and the Encoder setting is ignored. Try changing the output mode to "force recode"
J: is partition on a local hard drive on a motherboard SATA port.

I will try 'force recode' soon, have to go out now.
I assumed that my choosing to reduce the bit rate to 3000kbps would effectively force recode but I will try explicitly stating it.
I may also try recoding to MPEG2 as an experiment.
 

Danr

Administrator
Staff member
Recoding to MPEG2 HD .TS output with QS was about the same, 92 FPS. Looking forward to your results.
 

SamuriHL

Member
Crashed the video driver. LMAO. Outputting to h.264 MKV using QS.

2014-09-18 13:49:55 Source video information:
File: Name : Under the Dome.S02E05.Reconciliation.TiVo
Size : 6.439 GB
Duration : 01:02:00.59
Mux type : TiVo TS
TS mux rate : 15.865 Mbps
Video: Encoding : MPEG2
VideoStreamID : x800
Frame rate : 59.94 fps
Encoding size : 1280 x 720
Aspect ratio : 16:9
Header bit rate : 20.000 Mbps
VBV buffer : 976 KBytes
Profile : Main@High
Progressive : Progressive Only
Chroma : 4:2:0
Bit rate : 12.925 Mbps
Captioning : EIA 608/708
Field order : Progressive
Audio Stream: 1 (Primary) Codec : AC3
Channels : 5.1
PID : x801
PES Stream Id : xBD
Bit rate : 384 Kbps
Sampling rate : 48000
Sample size : 16 bits

2014-09-18 13:49:55 Starting new Frame Accurate Output Segment: start:77160.422 (00:01:17.09), end:697882.211 (00:11:37.51)
2014-09-18 13:49:55 Preparing to send status to: 0 Audio volume changed
2014-09-18 13:49:55 Sending status: 'Audio volume changed' to module: 'MPEG2 decoder - 0', Type: Video frame
2014-09-18 13:49:55 Preparing to send status to: 0 Audio volume changed
2014-09-18 13:49:55 IntelVideoEncoder, Implementation: Hardware via 'D3D9' (514), Version: 1.10
2014-09-18 13:49:55 Sending status: 'Audio volume changed' to module: 'Audio divider:eek:utput - 0', Type: Audio frame
2014-09-18 13:49:55 Adding new graph range, Start: 77160 (00:01:17.09), End: 697882.21 (00:11:37.51)
2014-09-18 13:49:55 Output graph sets start time: 00:01:17.09
2014-09-18 13:50:00 Decode 1000, FPS: 202
2014-09-18 13:50:05 Decode 2000, FPS: 208
2014-09-18 13:50:09 Decode 3000, FPS: 208
2014-09-18 13:50:14 Decode 4000, FPS: 206
2014-09-18 13:50:19 Decode 5000, FPS: 207
2014-09-18 13:50:24 Decode 6000, FPS: 208
2014-09-18 13:50:29 Decode 7000, FPS: 207
2014-09-18 13:50:33 Decode 8000, FPS: 208
2014-09-18 13:50:38 Decode 9000, FPS: 208
2014-09-18 13:50:43 Decode 10000, FPS: 209
2014-09-18 13:50:48 Decode 11000, FPS: 205
2014-09-18 13:50:53 Decode 12000, FPS: 207
2014-09-18 13:50:58 Decode 13000, FPS: 207
2014-09-18 13:51:02 Decode 14000, FPS: 210
2014-09-18 13:51:07 Decode 15000, FPS: 208
2014-09-18 13:51:12 Decode 16000, FPS: 208
2014-09-18 13:51:17 Decode 17000, FPS: 209
2014-09-18 13:51:22 Decode 18000, FPS: 207
2014-09-18 13:51:26 Decode 19000, FPS: 208
2014-09-18 13:51:31 Decode 20000, FPS: 208
2014-09-18 13:51:36 Decode 21000, FPS: 205
2014-09-18 13:51:41 Decode 22000, FPS: 208
2014-09-18 13:51:46 Decode 23000, FPS: 206
2014-09-18 13:51:51 Decode 24000, FPS: 206
2014-09-18 13:51:55 Decode 25000, FPS: 208
2014-09-18 13:52:00 Decode 26000, FPS: 199
2014-09-18 13:52:05 Decode 27000, FPS: 205
2014-09-18 13:52:10 Decode 28000, FPS: 208
2014-09-18 13:52:15 Decode 29000, FPS: 206
2014-09-18 13:52:20 Decode 30000, FPS: 208
2014-09-18 13:52:25 Decode 31000, FPS: 212
2014-09-18 13:52:30 FFmpeg muxer, video write error, current frame, originalPTS: 606022.09, PTS: 200.00, DTS: 529462.06
previous frame, originalPTS: 606405.81, PTS: 529445.38, DTS: 529445.38
2014-09-18 13:52:30 Output graph aborted at module: Output muxer, error: FFmpeg muxer: error writing packet.

2014-09-18 13:52:30 Audio divider:eek:utput: Process thread complete.
2014-09-18 13:52:30 Graph, monitoring thread received terminate signal.
2014-09-18 13:52:30 Audio decoder:0 processing thread complete. In: 16550, Out: 16550
2014-09-18 13:52:30 Output muxer processing thread complete. Video in: 31726, Out: 31725, Buffer: 0
2014-09-18 13:52:30 MPEG2 decoder: Processing thread complete. In: 31749 Output:31743, Repeat: 0, Buffer: 3
2014-09-18 13:52:30 Graph monitoring thread finished.
2014-09-18 13:52:30 Output muxer processing thread complete. Audio stream 0 In: 16539, Out: 16539, Buffer: 0
2014-09-18 13:52:30 Muxer add / delete audio: stream: 0, add: 0 delete 0, sync: 0.02
 

Danr

Administrator
Staff member
@SamuriHL

1) Repeat with QS and see if it works?

2) If it does, try again.

3) Are you changing anything while saving such as video dimensions, frame rate, or audio volumes?

 

SamuriHL

Member
What's weird is that VRD just went back to the editing screen after it happened with no status screen. So I simply clicked save again, set the profile I've been using for about 50 or so encodes at this point (rock solid btw except for this little hiccup), and it worked fine. Didn't even have to reload VRD nor did I lose my cut points. I am not changing anything at all. Just cutting commercials and outputting to h.264 MKV from the mpeg-2 TiVo source files. I'm in the middle of doing 12 files and that's the only one that gave me any trouble so far. I have 4 left.
 

murrayt

New member
OK

Thanks to previous posters I have now got the hang of Intel Quicksync in V5
I hadn't registered that you need to engage the encoder in the profile options!!

I have also confirmed that for my ASUS Z87 Deluxe/Dual motherboard that the latest driver set from ASUS works, as do the latest from the Intel site. A poster in one of the threads (can't locate it at the minute) found that the newest Intel drivers didn't work on a similar board (Z87 Deluxe).

So the results for a video segment 1080p+DTS with a length of 6952 with recoding to 720p with Intel Quicksync gives me 122.9 fps whereas the default encoder achieves only 50.8 fps (Woo-hoo)
CPU usage in an i7 4770 processor sits at between 70-80% generally whereas for the default it sits at between 90-95%.

In V5 I have Hardware (any) selected in the options.

Hope that helps others
 

bdg2

Member
would the instructions on this website help anyone? It shows how to enable Quicksync without a 2nd monitor.
I can also state at least for my ASus motherboard that the latest Intel driver did not work but the latest Asus supplied
Quicksync VGA driver did indeed enable it perfectly.

https://mirillis.com/en/products/tutorials/action-tutorial-intel-quick-sync-setup_for_desktops.html

jon
Ah ha.

For me this step seems to be very important.
After following this, and experimenting a bit, QuickSync is working much better for me.

But this makes some things work really weirdly, some popup dialogues that used to come dead centre on my display (e.g. some prompts about elevating privilege) now come up right over at the right hand edge of my display. Also I now have some annoying extra things in the right click on the desktop menu which configure the display on an imaginary non existent monitor on my Intel HD 4600 Graphics, and I often lose my mouse pointer because it's somewhere on the imaginary display. This isn't really acceptable for me. :mad:

I'm still using the latest Intel driver. My motherboard is an Asus Z87A so maybe I should try drivers from the Asus site too.

I also need to choose 'Hardware (Device 2)' as the QuickSync implementation setting in VR5.
The other settings are nowhere near as fast.

I'm getting about 100fps and 30% CPU for recoding 1920x1080 H264.
(Using the profile I used before, I didn't need to explicitly choose re-encoding)

I can't seem to get re-encoding to 1920x1080 MPEG2 working at all.
 

Attachments

Last edited:

JMRDV

Member
I am using the Asus Z87 Deluxe - the latest Intel drivers are quicker. I will add that I got QS working and then something changed. After tearing out the last few strands of hair in head, I re-imaged Windows 8.1 from my install image and updated from there. All kinds of nonsense went away and of course Quick sync has been working fine ever since. At this stage I am still reinstalling a few last programs doing a restore before each just in case.
AS an aside I also see that DVD fab specifically mentions fixing problems recognizing Quick sync in their latest update. DVDFab 9.1.6.8 Updated!


Fix: A problem that DVDFab could not detect IQS (Intel Quick Sync) GPU with Intel(R) HD Graphics driver version 10.18.10.3907.

and this is the latest driver which I am using very nicely.

Good Luck
jon
 

SimonP

Member
I hadn't registered that you need to engage the encoder in the profile options!!
It took me a while to find it too. I wonder why it isn't under Tools > Options > Intel Quicksync as you'd expect it to be? I can't think of a situation where you'd want to disable it for one profile and it would save having to enable it for all of them manually. Better yet, wouldn't it be easier for the program to detect a QS compatible processor/graphics card and enable it automatically? Are there any down-sides to using QS?
 

Dan203

Senior Developer
Staff member
Having two options is confusing. That's why we removed global options for advanced TS, PS and MP4 features and moved them exclusively into the profiles. We've talked about adding an option to the installer that set QS set as the default encoder across all profiles but we haven't had time to do it yet.
 

Danr

Administrator
Staff member
I can't think of a situation where you'd want to disable it for one profile
Here are a few reasons:

1) QS doesn't offer dual pass encoding.

2) Software encoding can offer higher quality at lower bit rates.

3) We haven't tested QS at really high bit rates and/or 4K video.

That being said, Dan203 and I agree that having an option that sets the default encoder to QS would be the best way to go. Would certainly make the "out of the box" experience better for new users. In this case there would be 3 options in the encoder dropdown:

Default
Quick Sync
Software (Mainconcept).
 

SamuriHL

Member
For my workflows QS has been more than adequate and indistinguishable from the original. But yes, QS is not good in low bitrate environments...at least not as good as software. Not to overcomplicate things but would you consider a range default? Under a certain bitrate target use mainconcept, between x and y use QS, above y use....etc.
 

Danr

Administrator
Staff member
For my workflows QS has been more than adequate and indistinguishable from the original. But yes, QS is not good in low bitrate environments...at least not as good as software. Not to overcomplicate things but would you consider a range default? Under a certain bitrate target use mainconcept, between x and y use QS, above y use....etc.
That could be problematic. If you have a video with low complexity, i.e. fairly static scenes and not a lot of color variations (think of talk show, Charlie Rose comes to mind here in the USA), then a QS at low bit rate would be fine. I think for now, making "Default" be settable to QS or Software should be sufficient with an override on specific profiles.
 

SamuriHL

Member
That could be problematic. If you have a video with low complexity, i.e. fairly static scenes and not a lot of color variations (think of talk show, Charlie Rose comes to mind here in the USA), then a QS at low bit rate would be fine. I think for now, making "Default" be settable to QS or Software should be sufficient with an override on specific profiles.
Fair enough. Thanks!
 

glenpinn

Member
Hi, i know i may have raised this subject before, where i cited Handbrake as being much faster than the VRD Quick Sync alternative, but i thought i would throw it up again just to find out why this might be the case.

I just imported a 4min 1080/50p m2ts file and cut 1 minute from it, then output to the following, just to see what QS does compared to Software encoding.

VRD Smart Render to 1080/50p m2ts
Software = 31 seconds @ 24.0Mbps @ 630mb @ minimal cpu usage (this seems to take around 25 seconds just to start the actual output, once it starts, the process is completed in around 6 seconds)
QS = 6 seconds @ 23.9Mbps @ 629mb @ minimal cpu usage (this is 6 seconds flat, no 25 second delay at the start like there was with the software output)

VRD output to 720/50p mp4
Software = 180 seconds @ 12.2Mbps @ 312mb @ 90% average cpu usage
QS = 92 seconds @ 10.3Mbps @ 263mb @ 65% average cpu usage

VRD QS is only twice as fast as its software encoding, so i personally dont think this is enough of a time saver to warrant using QS.

Handbrake output to 720/50p mp4
QS with quality factor 20, High @ 4.1 = 33 seconds @ 12.5Mbps @ 320mb @ minimal cpu usage

Handbrake QS is 3 times faster than VRD QS with similar settings, and as i expected it was 5.5 times faster than VRD software encoding, which is about what i have experienced in the past.

What factors might be causing VRD QS to be that much slower than Handbrake (and other QS encoders i have used with similar output time differences)

Cheers
 

Danr

Administrator
Staff member
Handbrake uses more of the QS functions than VRD does and may have different internal quality settings as well. We will improve the QS performance after we get V5 released.

In the 707 build the QS settings accidentally enabled the V4 cutting engine, which is much faster than the V5 engine, but not nearly as safe. Our new V5 smart rendering engine is slow at cut points when there are no IDR frames, but very fast during the body of the video. When there are IDR frames, it's totally safe to switch to the V4 engine which is faster. However using Smart Cutting on short videos isn't a good test. A better test would be to cut a long video like a TV show, then the overhead of the smart cutting will less of % of the overall period. We will look at optimizing the V5 cutting engine as well, but for now we are concentrating on the release.
 
Top Bottom