Evaluation of HEVC decoders (SW, Hybrid and HW)

[huhn]

DXVA doesn't work for me too.

[MisterXDTV]

Just bought my GTX1050ti but MPC-HC can't hardware decode files with HVC1 codec ID works fine with HEVC ID

What am I doing wrong??? what's the difference between HVC1 and HEVC???

[sneaker_ger]

Which decoder are you using? Are you using LAV Video? Use "DXVA2 (copy-back)". If it still doesn't work post MediaInfo and sample.

[MisterXDTV]

Which decoder are you using? Are you using LAV Video? Use "DXVA2 (copy-back)". If it still doesn't work post MediaInfo and sample.

I resolved remuxing the mp4 file to mkv.... :o:o

The codec ID of the mp4 file was not recognized by LAV Video

[NikosD]

The codec ID of the mp4 file was not recognized by LAV Video

Why don't you post the sample to LAV's video thread in order for the developer to check it out ?

[CruNcher]

@MisterXDTV

im not yet stumbled above that fourcc at all but

fourcc.c

search

A("h265"),

and on the end add

A("hvc1"),

this should then let ffmpeg recognize it as supported and let it through.

@All

Lentoid partly still performs better then lav video, especially when it gets really rough on the CPU Multithreading Edge on 4 Cores really nice for a 2015 Decoder with only 1 major decoding issue :)

http://i1.sendpic.org/t/2X/2XpDnqIyRYx5DxDUN3WsIhqE6IJ.jpg (http://sendpic.org/view/1/i/vJFQd9s4pmgvfSsiFxgbFSWytKN.png)


currently torturing 160W (250W 3D Engine) stable out of my I5-2400 system (complete i/O everything for video playback) with all 4 cores under maximum load while rendering on EVR CP so i should be pretty maxed out at the 95W tdp :D

the LG Chess Demo should be added.


Idle state (after caching, cycle overhead mostly Firefox related, though causes no visible CPU breakdowns)

http://i1.sendpic.org/t/uM/uM6JWTFOj44dkDAiZ1BfUjzzMsA.jpg (http://sendpic.org/view/1/i/y5f4FiShIwdZkvcyh1RSKmPyZGe.png)

Playback Result (Decoding/Rendering)

http://i1.sendpic.org/t/uy/uy2ONC28q4NfaBANa2r82hw4SyV.jpg (http://sendpic.org/view/1/i/kbOlJXnoiYyTezTj0OXLKzykIhp.png)

@Yups

How does this look on your Kaby-Lake currently with Lav Video (avcodec) and compared to the Quicksync Decoder (Via DXVA and via Quicksync) ?

[MisterXDTV]

Why don't you post the sample to LAV's video thread in order for the developer to check it out ?

It's the Sony 4K HDR Demo here:

https://mega.nz/#!0VMWhRoD!6hR2-lRoYKqO1Dpf7ylHNna6z1MIuNO4ByySxUIp-E4

Check it out



General
Complete name : ok_Sony_4K_HDR_Camp.mp4
Format : MPEG-4
Format profile : Base Media
Codec ID : isom (isom)
File size : 1.12 GiB
Duration : 2 min 7 s
Overall bit rate mode : Variable
Overall bit rate : 75.8 Mb/s
Encoded date : UTC 2016-02-03 08:01:30
Tagged date : UTC 2016-02-03 08:01:30

Video
ID : 1
Format : HEVC
Format/Info : High Efficiency Video Coding
Format profile : Main 10@L5.1@High
Codec ID : hvc1
Codec ID/Info : High Efficiency Video Coding
Duration : 2 min 7 s
Bit rate : 75.6 Mb/s
Maximum bit rate : 123 Mb/s
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 59.940 (60000/1001) FPS
Color space : YUV
Chroma subsampling : 4:2:0 (Type 2)
Bit depth : 10 bits
Bits/(Pixel*Frame) : 0.152
Stream size : 1.12 GiB (100%)
Encoded date : UTC 2016-02-03 07:59:49
Tagged date : UTC 2016-02-03 08:01:32
Color range : Limited
Color primaries : BT.2020
Transfer characteristics : SMPTE ST 2084
Matrix coefficients : BT.2020 non-constant
Mastering display color primar : R: x=1.000000 y=1.000000, G: x=1.000000 y=1.000000, B: x=1.000000 y=1.000000, White point: x=1.000000 y=1.000000
Mastering display luminance : min: 0.1000 cd/m2, max: 0.5000 cd/m2

Audio
ID : 2
Format : AAC
Format/Info : Advanced Audio Codec
Format profile : LC
Codec ID : 40
Duration : 2 min 7 s
Source duration : 2 min 7 s
Bit rate mode : Variable
Bit rate : 192 kb/s
Maximum bit rate : 197 kb/s
Channel(s) : 2 channels
Channel positions : Front: L R
Sampling rate : 48.0 kHz
Frame rate : 46.875 FPS (1024 spf)
Compression mode : Lossy
Stream size : 2.91 MiB (0%)
Source stream size : 2.91 MiB (0%)
Language : English
Encoded date : UTC 2016-02-03 07:59:49
Tagged date : UTC 2016-02-03 08:01:32

[nevcairiel]

That file plays just fine here, hardware decoding and all (in both native and cb modes), must be something on your end.

[CruNcher]

Ahhh so it was already added jesus actually mostly the Sony are hvc1 i wonder these is the encoder Lentoid 2015 has problems with it seems
if Channel 4K matches that it gets interesting.

Codec-ID : 36

huh


Sonys colors Demo has the same problematic with Lentoid all use Slices it seems

Codec-ID : hvc1

hmm which encoder is this ?

Now i wonder if all these Codec ID 36 where written on the stream capturing part from some hardware device.

But yeah Lav Videos Parser/Decoder doesn't really care about these ids and other parser let them through as well so the ID seems not the only issue.

also older Intel Platforms have Problems with these sliced streams

https://communities.intel.com/message/433628



What i find really crazy

http://xhevc.com/en/hevc/decoder/download.jsp

they say no HEVC 10 bit decoding but lot of HEVC streams shown as 10 bit work only these sliced Sony Demos and other Japanese Test streams make real problems compared to lav video (openHEVC)

[MisterXDTV]

That file plays just fine here, hardware decoding and all (in both native and cb modes), must be something on your end.

As i said, remuxing to MKV container changed the ID to V_MPEGH/ISO/HEVC and that alone fixed the problem. It took me 5 seconds..

The LG Chess was an MP4 with the proper hevc ID and worked fine directly without remuxing

[CruNcher]

No its no .mp4 lot of these demos have wrong file extensions but in general that shouldn't be really a major problem
Most of these Demos are Satellite Captures or coming from Device Demos that where on the Devices themselves for Showroom Demo purposes.
Some are actual x265 retranscodes from uknown source.

Some could be also legit x265 tests in cases of LG but LG widely uses ATEMEs Titan Encoder same as Samsung

[MisterXDTV]

No its no .mp4 lot of these demos have wrong file extensions but in general that shouldn't be really a major problem
Most of these Demos are Satellite Captures or coming from Device Demos that where on the Devices themselves for Showroom Demo purposes.
Some are actual x265 retranscodes from uknown source.

Some could be also legit x265 tests

Yeah, I couldn't believe it at first, the remuxing was a wild guess by me but I didn't believe it could work. Usually when a file doesn't play right remuxing to another container does nothing to fix the problem. In this case it did but I don't know why.... :D

[CruNcher]

Im not surprised by anything currently looking at my own oddities in behavior outside of lav video ;)

but nevcairiel is on Nvidia and it would be really crazy if his older VPX has not these issues and the one on the Driver (Firmware) side of the 1050 Ti has those, nothing impossible but highly unlikely.

did you tried a direct playback chain (Graphedit) maybe something gets in the Dshow Chain and interferes somehow ?

and is the Channel 4K stream http://demo-uhd3d.com/fiche.php?cat=uhd&id=54 with Codec ID: 36 also failing for you on the 1050 TI not remuxed ?


Found some new camera Hardware Encoder sample with a very interesting resolution signaling :)

https://www.youtube.com/watch?v=AqlTk4uAcTI

[MisterXDTV]

Im not surprised by anything currently looking at my own oddities in behavior outside of lav video ;)

but nevcairiel is on Nvidia and it would be really crazy if his older VPX has not these issues and the one on the Driver (Firmware) side of the 1050 Ti has those, nothing impossible but highly unlikely.

did you tried a direct playback chain (Graphedit) maybe something gets in the Dshow Chain and interferes somehow ?

and is the Channel 4K stream http://demo-uhd3d.com/fiche.php?cat=uhd&id=54 with Codec ID: 36 also failing for you on the 1050 TI not remuxed ?


Found some new camera Hardware Encoder sample with a very interesting resolution signaling :)

https://www.youtube.com/watch?v=AqlTk4uAcTI

I will download that stream but that server is really slow, it will take a while

this is my graph edit for that Sony file, looks fine to me

https://abload.de/img/grapheditssq8e.jpg

EDIT: I have the same problem with this Exodus demo clip (hev1 ID)

http://files.hdrsamples.com/downloads/hdr/Exodus_UHD_HDR_Exodus_draft.mp4

Again remuxing to mkv fixes the problem and DXVA2 copy-back works perfectly

[CruNcher]

I will download that stream but that server is really slow, it will take a while

this is my graph edit for that Sony file, looks fine to me

https://abload.de/img/grapheditssq8e.jpg

EDIT: I have the same problem with this Exodus demo clip (hev1 ID)

http://files.hdrsamples.com/downloads/hdr/Exodus_UHD_HDR_Exodus_draft.mp4

Again remuxing to mkv fixes the problem and DXVA2 copy-back works perfectly

this is really interesting slowly i guess we can pinpoint your issue it to certain HDR10 bitstreams i wonder if the fourcc actually in the standard for HDR10 is actually hev1/hvc1.

that maybe can explain why Codec ID: 36 maybe gets not correctly detected by some logic and handled differently then it should and in your case maybe it's the same by using the normal Main10 fourcc something gets ignored by the Hardware parser and then it works

Which in the end might be meaning though that you will lose HDR on the Renderer side and need to manually set it up to get the correct wide color gamut and luminance response ?.

Though in this case you should really refer to MadVR instead especialy on pre Windows 10 Aniversary ;)


https://www.youtube.com/watch?v=fHa10EbyKTk

Apollo Lake playing 4K 120Mbps HEVC Main10 video.

Looks great so far but in reality way to expensive if your target is just efficient avc/hevc/vp9 4K 60 decoding at lowest power possible ;)

For that price you get a Entry 40 4K TV with well performing MediaTek Platform with Fake HDR ;)

Though at 250 entry point it becomes very interesting and the following HDMI sticks as well :D

[famasfilms]

I have an i5-6600 and GTX 970.

I use MPC-HC, lavfilters and madVR, but trying to play 10bit HEVC files results in dropped frames.

I thought this PC would handle everything 4k but apparently the 970 series doesn't have hardware HEVC decoding.

Is there alternative software I can use or is the only answer upgrading my gfx card (pointless as I am not a PC gamer) or downgrading to a 960?

[CruNcher]

I have an i5-6600 and GTX 970.

I use MPC-HC, lavfilters and madVR, but trying to play 10bit HEVC files results in dropped frames.

I thought this PC would handle everything 4k but apparently the 970 series doesn't have hardware HEVC decoding.

Is there alternative software I can use or is the only answer upgrading my gfx card (pointless as I am not a PC gamer) or downgrading to a 960?

On Nvidias side you have 3 choices

960/1050TI/1060

neither is really optimal and you should chose wisely based on your heaviest workflows.

The GTX 970 CUDA HEVC Decoder is limited it copes with Medium 8 bit Complexity rather well but if the complexity is 1 step above it falls apart fast and there is the problem that the PWM works not so stable at all over Windows DXVA and you better refer to CUVID instead.

a good balance is 1060 if you could survive VP9 10 Bit Decoding maybe not becoming available or being as in your current situation for H.265 on GTX 970 Cuda Driven and so overall limited to the Shader array performance.

the 1060 is virtually exactly the highest Performing GTX 970 but updated towards lower power consumption and no additional latency impact due to to low ram with a more punchy VPX heart overall the GTX 970 was Designed for 4K 30 FPS 3D so is the GTX 1060 both can drive 60 FPS for Video without any problem and low processing 3D like the Windows GUI has itself no issues running 60 FPS stable also optimized 3D results have no Problem running 4K 60 stable as long as they not heavily use the shader array for Compute tasks at the same time for example, the GTX 1060 is a bit more efficient here then the GTX 970 which can help on latency when both use cases crash into each other heavily. :)

If you mix Compute, 3D and Video Decoding/Encoding at 4K 60 at the same time you need some punch for Realtime results and i advise to think about the 1070 in that case with more headroom in the Consumer space, don't invest though then in the most expensive that's total nonsense get the cheapest with balanced results.

All my tests on GM204 also showed that still the efficiency of Nvidias Mpeg-2/4 (ASP) Decoder Core is not good @ all and it can be better for efficiency purposes to not use them @ all but instead use Lav Video (avcodec) at least for Playback Decoding (DVD,Broadcast and Blu-Ray, Xvid, DivX) i would advise to prefer avcodec, this could save a lot of Energy for most System configurations especialy when you use a overclocked GTX 970.

Happy new Year everybody.

[CruNcher]

You can already see from the size just for 4K playback it is inefficient

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc7i5bnk.html

But maybe it can press some usefull results out of MadVRs processing chain

That thing becomes though a interesting contender to Nvidias Shield TV slowly :)

Maximum System output seems to be around 65W that's a lot if you can't use it efficiently

i would go more for that Apollo lake @ cheap :D


N4200
J4205
J3455

but you should keep an eye on the USB3 transfer speeds and forget USB2 devices at least 1 USB3 port ;)



I hate Youtube reviewers these bunch of corrupted idiots.

Advising people to get overpriced Chinese Mini PCs for so much more then a complete Notebook from a well known Brand with the same hardware how dumb must you be.

[NikosD]

We need RyZen ASAP to escape from another Intel's fraud, the most miserable and useless "upgrade" ever for the desktop PC (besides media capabilities)

The shameless guys of Intel released a new CPU with the same IPC, with exact same performance clock for clock with the previous generation.

AMD, please help us!

[JohnLai]

We need RyZen ASAP to escape from another Intel's fraud, the most miserable and useless "upgrade" ever for the desktop PC (besides media capabilities)

The shameless guys of Intel released a new CPU with the same IPC, with exact same performance clock for clock with the previous generation.

AMD, please help us!

Down with INTEL CPU, thumb up with intel IGPU features.....

[CruNcher]

Sorry NikosD but you dont seem to understand Chip Design and Balancing.

And that you have to make decisions where to invest your shrink advantage.

Or you endup with Chips overheating at maximum gate load ;)

Apollo lake is very impressive for what it packs inside and to the outside for a non RISC Chip.

https://www.youtube.com/watch?v=HjXbAPXf_Vo

And as i told you ARM feared only Intel would reach their GFX capabilities fast and by now they overrun them the ARM CEOs worst nightmare became true with Apollo lake ;)


http://i1.sendpic.org/t/lh/lhE3x7ro7Q3UuLmx2hRMoX49m1S.jpg (http://sendpic.org/view/1/i/bBBAEXHivMwmT3OcNwJ0QFMUnpO.png)

[NikosD]

What I understand for sure is that Intel is so afraid of RyZen that was forced to update Skylake with an overclocked Skylake naming it Kabylake, just to have something to compete with the architecturally more advanced RyZen - in terms of 8C/16T design with less than 100W TDP very close to i7 7700K TDP actually.

So, AMD managed to produce a CPU for desktop that is competing with HEDT CPUs of Intel which sells them in the price range of more than 1000$ and a TDP of 140W.

I'm waiting for the official release of VEGA and RyZen to start real celebration for AMD products climbing up to the roof of both CPU and GPU performance

[CruNcher]

AMD still lacks a lot on UVD and VCE

Quicksync is much better overall integrated and reaches astonishing results since Haswell already :)

If AMD cant integrate as efficient using GCN as the foundation it will look bad and they will fail at low power again.

The real advances happen at low power not where Rayzen currently is targeting for.

6W SDP 15W TDP thats where AMD always had problems maintaining Performance and if you lose their you lose everywhere ;)

Also AMD never wanted to catch Intel @ all they knew they never could they just wanted to position as 2nd and survive.

With RyZen they will at least manage that again and hopefully the successor to Carizzo will also show that.

[NikosD]

What happened to your arguments regarding Chip Design and Balancing ?

Went back to UVD and VCE ?

Tell me more about the fraud of HEDT chips of Intel.

[CruNcher]

What happened to your arguments regarding Chip Design and Balancing ?

Went back to UVD and VCE ?

Tell me more about the fraud of HEDT chips of Intel.

Never had a problem with those issues always made sane decisions on intels side with the chip capabilities and tried to find a balance so i found the prices off intel on the i-5 always ok compared to AMDs Power Hungry HEDT with less performance ;)

i7 plays no role for me its unbalanced ;)

and AMD has not to win against i-7 but i-5 if they really want the crown with RyZen you should understand what ZEN really stands for, it doesn't stand for Gamer owning as a marketing word but initially it stands for "balance" ;)

comparing vs i-7 is absolutely not what ZEN stands for and i found it a big marketing gag ;)

in my perspective i-7 lineup could die tomorrow wouldn't miss it ;)

[oddball]

Someone suggested I post decoding results of LG's HDR Chess demo here.

First is MPC-BE with LAV filters using DX Copyback and EVR renderer.

http://i.imgur.com/IPjPrs7.jpg

Second is with MadVR. I cannot seem to get it under control GPU usage wise using MadVR.

http://i.imgur.com/4IqPHJB.jpg

This is using an AMD Radeon RX 480 overclocked to 1400/2200Mhz and a Intel i7 4770K @ 4.2Ghz with 16GB DDR3 @ 2200Mhz.

[CruNcher]

4% or lower with DXVA Native ?

was this EVR Custom or EVR Native and what playback application is this ?


The MadVR overhead is really crazy double the consumption something is going very wrong somewhere.

Are you sure there is none of the OpenCL/Directcompute Processing running ?

Very strange that extreme overhead if here only nearest neighbor is active.

What is shown in the Rendering Statistics CTRL+J ?

[tahaa7]

I've been following this thread, and it's been a fascinating and very informative read.

A quick question (from a relative noob): I'm buying a new laptop soon, and let's say I buy one with a Kaby Lake CPU and without a dedicated GPU. Would I be better off buying a CPU with Iris Plus or, say, 7700HQ (with Intel HD 630), in terms of multimedia performance (HEVC 4K 10-bit decoding, VP9 decoding etc.)? I see many of these U-series CPUs have higher prices than quad-core H-series ones, so I wonder why exactly that is. Is it because of graphics performance, or something else?

[MisterXDTV]

Do you think it's theoretically possible that Pascal cards could get their HDMI port upgraded to 2.1 via firmware upgrade? They decode 8K after all...

[huhn]

no not everything the connecter can't do 48 mbit.

some of the HDMI 2.1 stuff could be added. they did this with kepler before.

[aufkrawall]

I'm fiddling around with Linux a bit atm, so I could give VDPAU decoding a try.
Unlike DXVA2 native, it doesn't suffer by extremely blurry chroma, quality seems to be the same as software decoding with mpv and opengl-hq profile (has deband active, didn't pay attention to banding).
Unfortunately, Nvidia doesn't yet support VP9 and HEVC 10 bit decoding via VDPAU, which is a real bummer.

[CruNcher]

You should look into Wayland it's awesome for Video Desktop Performance, sadly not much supports it yet natively :)

[aufkrawall]

Dunno if it is possible yet with Nvidia.
I'm without any compositor since everyone for X11 doesn't seem to work perfectly.

[CruNcher]

It is most supporting Distro currently seems Red Hats Fedora (Workstation), some really nice results there on latency and usability i hope it wont fail it's a big step since compiz and beryl to a real nice user experience on every device :)

https://www.youtube.com/watch?v=F6PFjoYuml0
https://www.youtube.com/watch?v=Ux-WCpNvRFM
https://www.youtube.com/watch?v=s8XiEyGVKcI
https://www.youtube.com/watch?v=ETSsex3M_Oc

https://vimeo.com/3220996

[aufkrawall]

But only with open-source drivers, and you can totally forget about Nouveau.
And I also hate Gnome, waiting for KWin Wayland session that works with proprietary Nvidia driver...

[NikosD]

I've been following this thread, and it's been a fascinating and very informative read.

A quick question (from a relative noob): I'm buying a new laptop soon, and let's say I buy one with a Kaby Lake CPU and without a dedicated GPU. Would I be better off buying a CPU with Iris Plus or, say, 7700HQ (with Intel HD 630), in terms of multimedia performance (HEVC 4K 10-bit decoding, VP9 decoding etc.)? I see many of these U-series CPUs have higher prices than quad-core H-series ones, so I wonder why exactly that is. Is it because of graphics performance, or something else?

The performance of HW decoding alone, is only dependant from the clock for the same generation chips e.g Kabylake.

But for scaling, the low number of EUs could be a problem e.g when you try to decode a 4K clip on 720p resolution.

The weak Intel iGPUs have such problems.

If you are interested in HW encoding too, then more EUs means faster HW encoding.

And of course more EUs means faster 3D and GPGPU (computing) performance.

Products like Iris (Pro or Plus) have embedded memory which helps not only the GPU but the CPU too and for certain tasks can be very helpful but not that much in general CPU tasks.

For example Broadwell Core i7 5775C is faster than any desktop Core i7 even Kabylake in some compression tests.

Personally I wouldn't mind that much for embedded ram and I would wait some months for RyZen for sure.

We don't know the multimedia features of VEGA that RyZen APUs will have inside.

[CruNcher]

The performance of HW decoding alone, is only dependant from the clock for the same generation chips e.g Kabylake.

But for scaling, the low number of EUs could be a problem e.g when you try to decode a 4K clip on 720p resolution.

The weak Intel iGPUs have such problems.

If you are interested in HW encoding too, then more EUs means faster HW encoding.

And of course more EUs means faster 3D and GPGPU (computing) performance.

Products like Iris (Pro or Plus) have embedded memory which helps not only the GPU but the CPU too and for certain tasks can be very helpful but not that much in general CPU tasks.

For example Broadwell Core i7 5775C is faster than any desktop Core i7 even Kabylake in some compression tests.

Personally I wouldn't mind that much for embedded ram and I would wait some months for RyZen for sure.

We don't know the multimedia features of VEGA that RyZen APUs will have inside.

Waiiiiit
What
Intel HD Graphics has the best Hardware Scaler since GEN2 way more advanced then AMD and Nvidia back then and partly even still today.

Intels Hardware Scaler is Lanczos4 fully Hardware implemented, since GEN2 when Nvidia and AMD where still pushing around with Bicubic/Billinear by default.

And i think even today not much changed their and Intel still has the better Hardware Scaling and the Performance effects of it are almost non existent natively.

Nvidias and AMDs approach go more over the shaders if you want higher quality then Billinear/Bicubic.

Intel doesn't neeed to touch the EUs for Scaling really at all and that overhead doesn't exist with better quality because Intels Hardware scaler is also Adaptive.

[NikosD]

It was about time to reveal one more secret of Intel iGPUs, although I have posted that info in Doom9 a few times.

Almost all of the problems mentioned in Intel forums and even by PJ (IIRC) there with the low EUs APUs of Intel trying to playback 4K files, I have concluded that come from the scaler.

I had myself a Pentium Haswell G3420 and although it had the same QuickSync ASIC like the bigger models, it had definitely a lot of problems using playback mode of 4K clips scaled down to lower resolutions.

If you try to decode those files and scale them in lower resolutions you will see that GPU usage goes to almost 100%.

With all pre/post filtering off, there is no other excuse for the huge GPU utilisation of pure HW decoding.

The number of EUs play a significant role in scaled playback of 4K files even when they are decoded in HW.

My experience from Pentium G3420 says so.

[CruNcher]

It was about time to reveal one more secret of Intel iGPUs, although I have posted that info in Doom9 a few times.

Almost all of the problems mentioned in Intel forums and even by PJ (IIRC) there with the low EUs APUs of Intel trying to playback 4K files, I have concluded that come from the scaler.

I had myself a Pentium Haswell G3420 and although it had the same QuickSync ASIC like the bigger models, it had definitely a lot of problems using playback mode of 4K clips scaled down to lower resolutions.

If you try to decode those files and scale them in lower resolutions you will see that GPU usage goes to almost 100%.

With all pre/post filtering off, there is no other excuse for the huge GPU utilisation of pure HW decoding.

The number of EUs play a significant role in scaled playback of 4K files even when they are decoded in HW.

My experience from Pentium G3420 says so.

I have a Bay Trail system and i test a lot on 4K AVC Decoding efficiency on it and it's just awesome that it can easily beat very advanced ARM Soc IP that is not able to playback 4K 60 FPS at all ;)

Scaling doesn't play a part here Zero copy efficiency does more then anything else and certain things can be problematic under Windows in that regards, this shows up at that 4W target extremely.

The Problem with Sonys Color Demo has different reasons.

Overhead of EVR-CP should also not be underestimated as well especially not for MPC-HC ;)

And i can't really wait to get my fingers on N4200 to be true :D

And well see if it's really impossible getting it efficient working without official Windows 7/8/8.1 support

[NikosD]

Try your internal iGPU of Sandybridge

[CruNcher]

That doesn't support 4K and i currently put it in sleep mode and optimized everything that way not sure what will happen if i wake it up to be honest, but i gonna do that some day again.

Else you would see in a lot of my Screenshots the HD Graphics Framebuffer as well ;)

[MisterXDTV]

no not everything the connecter can't do 48 mbit.

some of the HDMI 2.1 stuff could be added. they did this with kepler before.

I think almost everything HDMI 2.1 has to offer can be added except new resolutions/frame rates. (as 8K/120fps)

So dynamic metadata HDR, eARC and Variable refresh rate DON'T require the new 48G cable. They only need the current 18G cable....

Basically current HDMI 2.0b can become "2.0c" or something like that.....

[tahaa7]

OK, so it seems you guys are suggesting that Iris iGPU would be a better choice. However, in the case of mobile Kaby Lake processors, there is a trade-off in terms of CPU cores: you have a choice of either getting a dual-core CPU with Iris graphics, or a quad-core CPU with Intel HD 630 graphics. How much does that influence final performance (in terms of HEVC 10 playback in MPC-HC, madVR, etc.)?

(5775C chip you mentioned @NikosD, is also a quad-core chip).

[CruNcher]

They will be no major differences only difference might be overall cpu utilization changing but all will playback up to 4K 60 at least without problems.

So it doesn't really matter only in terms of overall efficiency their are big differences with N4200 the most efficient it seems so far thus used in entry laptops :)

Then you have those hardembeded ITX barebones then comes the NUC and then Desktop.

All have very different targets and so the chips are laid out for these :)

But all of them will pretty certainly reach the 4K 60 10 bit Quicksync Decoding goal.

So if only Decoding is your goal obviously you go with the most power efficient ;)

that currently is N4200

http://ark.intel.com/products/95592/Intel-Pentium-Processor-N4200-2M-Cache-up-to-2_5-GHz


You surely will also see a lot of cheap China Mini PCs based on it but most of the time their wont be worth it vs a Laptop platform which is exactly the same with additional display ;)

Though the price is overall heavy as such in one or another way you get highly optimized ARM based OctaCore HDR solutions much much cheaper that can also do 4K 60 and if you have a UHDTV of the newest Generation you need nothing overall they all come pre included with such High Efficient Playback and on some you can directly install other Playback solutions that will work out of the Box ;)

And they will still overun the Intel Kaby Lake Barebones in Price/Value/Efficiency and in terms of Price vs N4200 Kaby Lake is no match and much to expensive just to be used for that usage scenario.

Though if you want to use some very Specific things like Netflix HD/4K you better of with the Intel solution.

The Barebones are the only semi efficient solution Price/Value/Efficiency wise but still Efficiency lacks.

https://www.youtube.com/watch?v=F_eEJk6lWlc

is pretty efficient :)

[tahaa7]

They will be no major differences only difference might be overall cpu utilization changing but all will playback up to 4K 60 at least without problems.

So it doesn't really matter only in terms of overall efficiency their are big differences with N4200 the most efficient it seems so far thus used in entry laptops :)

Then you have those hardembeded ITX barebones then comes the NUC and then Desktop.

All have very different targets and so the chips are laid out for these :)

But all of them will pretty certainly reach the 4K 60 10 bit Quicksync Decoding goal.

So if only Decoding is your goal obviously you go with the most power efficient ;)

that currently is N4200

http://ark.intel.com/products/95592/Intel-Pentium-Processor-N4200-2M-Cache-up-to-2_5-GHz


You surely will also see a lot of cheap China Mini PCs based on it but most of the time their wont be worth it vs a Laptop platform which is exactly the same with additional display ;)

Though the price is overall heavy as such in one or another way you get highly optimized ARM based OctaCore HDR solutions much much cheaper that can also do 4K 60 and if you have a UHDTV of the newest Generation you need nothing overall they all come pre included with such High Efficient Playback and on some you can directly install other Playback solutions that will work out of the Box ;)

And they will still overun the Intel Kaby Lake Barebones in Price/Value/Efficiency and in terms of Price vs N4200 Kaby Lake is no match and much to expensive just to be used for that usage scenario.

Though if you want to use some very Specific things like Netflix HD/4K you better of with the Intel solution.

The Barebones are the only semi efficient solution Price/Value/Efficiency wise but still Efficiency lacks.

https://www.youtube.com/watch?v=F_eEJk6lWlc

is pretty efficient :)

OK, thank you very much. Well, I will be buying a laptop that will be my main computer, but also occasionally be used as a HTPC. If the prices are right, I will certainly be looking to get one with a discrete GPU, and/or a quad-core CPU.

[Yups]

If you try to decode those files and scale them in lower resolutions you will see that GPU usage goes to almost 100%.

With all pre/post filtering off, there is no other excuse for the huge GPU utilisation of pure HW decoding.


I have never seen such a big GPU load on a Skylake and Kabylake GT2 nor do I noticed shader performance issues on a pure hardware decode, except for Hybrid decoding. As for encoding it is very important because of the Hybrid mode, even more for HEVC. But even with the huge shader performance advantage AMD has no chance regarding speed and quality with their current hardware. They have to catch Nvidia @Pascal as well. Personally I'm not interested in RyZen because of the lack of iGPU. Desktop APU seems to come only in H1 2018, which is probably a 4C SKU. I'm more interested in Coffe lake 6C in this case.

[CruNcher]

That is strange

https://youtu.be/y5EqYQ-6cEI?t=929

The Sony Camp HDR sample definitely fails

doesn't fit to what we saw in Tralaks jellyfish test at all

Interestingly it gets color graded as well not more into the slog range displayed it was recorded in looks like somekind of HDR to SDR being done :)

https://youtu.be/RxPxW8WaPks?t=489

[nevcairiel]

The Sony clip lacks proper HDR metadata, different players handle it differently. Some notice that its metadata is broken and assume defaults, some use the broken data and everything plays back all super saturated.

[CruNcher]

Hmm it looks more like the metadata get interpreted from the Windows 10 Parser/Decoder or even Quicksync side and a conversion takes place thus the colors stay not dull but get saturated like some lut applied

And the Kodi result on S912 is the pure one that resambles more the native slog space of the camera without any conversion going on or lut applied which is virtualy the most case players do it based on ffmpeg i saw so far what gets outputed by the Windows 10 Decoder (Player) here looks very different to anything else.

But the Performance is unusable in that N4200 result anyways no matter what goes on on the HDR side here .

Far away from what we saw here

https://www.youtube.com/watch?v=fHa10EbyKTk

Maybe a Driver issue

All these reviewers concentrate to much on jelyfish it's ridiculous

https://youtu.be/Fk5i3fvmIJc?t=354

Ok much better then this one this is really nahh 24 2000 kbps unbelievable that's a real challenge geez a Youtuber tests 24 @ 2000 kbps 4K :D

https://youtu.be/WvWX0MTo7WA?t=685


Redskull at least makes usable reviews

https://www.youtube.com/watch?v=y5EqYQ-6cEI

Also ok (but limited to jellyfish)

https://www.youtube.com/watch?v=Fk5i3fvmIJc

seems Chrome still runs the libvpx Decoder

Limits himself to DX11 tests on a Windows 10 system with a GPU Designed for DX12 ouch ;)

Cinebench 15 seems to be the most interesting Score to compare the CPU Efficiency with virtually every reviewer uses it in one way.

it indicates that my i-5 2400 is ~2x more efficient then the N4200 in that Laptop configuration which is when we compare the maximum output difference very impressive :)

from 2nd 22nm max 95W ->14nm 7th Generation max 6W it seems CPU Performance in direction of Tegra X1 i guess good comparable

i guess no bad compare point on the 3D side ;)

https://www.youtube.com/watch?v=qm16p4qYjF4

of course P1 will kill it inside the Nintendo NX but it's allways nice to see what Intel runs for ;)

Anyway cant wait to get my N4200 Laptop and take that thing apart in all Details :)

Also interesting because AVX is missing in N4200 :D

oha Nvidia has allready prepared his answer ;)

https://www.youtube.com/watch?v=AK9hpNb68xY

a Shrinx of X1 with i guess the newest VPX thats also in GP107

looks super sleek and sexy :D

Intels preparation of full module updates at least on the SOC side ;)

https://www.youtube.com/watch?v=VWtSBSRSg3s


This will make Video Hardware Codec upgrading very easy for TVs very useful, i saw this future coming :)

Hoho now they prepare to attack MediaTek :D

New Video Codec comes out your TV doesn't support no problem just replace the Module in the back (others surely will follow the idea including MediaTek).

This is so great so much Waste will be reduced and you don't have to buy a TV each Video Codec cycle but just replace the Module, that's the future :)

You can also upgrade obviously advancements in Post Processing this way too (more advanced Motion interpolation ect, instead of X amount of models that only differentiate by that) :)

So instead you need to buy a new TV for say 800$ you upgrade it for 200$

https://www.youtube.com/watch?v=aMeVl4PLKaQ

http://ark.intel.com/products/95449/Intel-Core-m3-7Y30-Processor-4M-Cache-2_60-GHz-
http://ark.intel.com/products/95592/Intel-Pentium-Processor-N4200-2M-Cache-up-to-2_5-GHz

very efficient and includes AVX compared to N4200


They Balancing Instructions vs Cores and GPU here to get higher efficiency :D

And strangely the Resolution is better specified then with N4200

http://www.cpubenchmark.net/compare.php?cmp%5B%5D=2864&cmp%5B%5D=2886&cmp%5B%5D=2875


Nice :)

Compared vs my Desktop

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-2400+%40+3.10GHz&id=793


I wonder if that Sony Clip Performance result of the N4200 could have todo with what YUPS said about the Clock Domain and how it works in combination

https://forum.doom9.org/showpost.php?p=1790164&postcount=1622

Also maybe the missing AVX plays into it especialy in that form factor but ok it's actively cooled and shouldn't really fluctuate on the clocks so much at all but unfortunately we never gonna see such indepth tests from any Reviewer in the World ;)


AMD is on the right track

http://www.cpubenchmark.net/compare.php?cmp%5B%5D=2861&cmp%5B%5D=793&cmp%5B%5D=2889

RyZen will be a exciting Value opportunity for End Users of all target groups and paired with Vega it could even do Hybrid Workflows more efficient then we ever seen before :)

https://www.youtube.com/watch?v=lGuFVc8AFXk

https://youtu.be/FwcUMZLvjYw?t=1458

[Yups]

That is strange

https://youtu.be/y5EqYQ-6cEI?t=929

The Sony Camp HDR sample definitely fails

doesn't fit to what we saw in Tralaks jellyfish test at all

Interestingly it gets color graded as well not more into the slog range displayed it was recorded in looks like somekind of HDR to SDR being done :)

https://youtu.be/RxPxW8WaPks?t=489


Why is this strange? His player clearly doesn't support HW decode.