Evaluation of HEVC decoders (SW, Hybrid and HW)

[P.J]

TS doesn't work in Win10

[D2MAC]

Hi,

Does Microsoft plan to support .ts HEVC in WIndows 10 final ?

They claim that they support HEVC in WINDOWS 10.

But no, not full ....

Without .ts support is disapointing

.ts is a popular container in Broadcast.

Microsoft is aware about .ts / HEVC problem?

Same problem in VLC with HEVC + .ts

[huhn]

and they support HEVC up to main10.
there is simply no TS demuxxer with HEVC support. but this doesn't change that windows 10 fully supports HEVC.

[D2MAC]

and they support HEVC up to main10.
there is simply no TS demuxxer with HEVC support. but this doesn't change that windows 10 fully supports HEVC.

It's annoying.

there is a lot of .ts UHD 4K samples.
lav filters)
And broadcast use .ts.

they should add a .ts demuxer for HEVC like MPC-BE/MPC-HC

there a lot .ts HEVC samples :

example (found on the web) :

HEVC 10 BITS .ts container download link :

http://www.demo-uhd3d.com/files/uhd4k/Channel_4K.ts

http://www.demo-uhd3d.com/files/uhd4k/Roland_Garros_MonFils_2014.ts

http://www.demo-uhd3d.com/files/uhd4k/Samsung_Fashion_Show_Milan_UHD.ts


HEVC .mp4 container :

http://www.demo-uhd3d.com/files/uhd4k/LG_nuit_villes.mp4

HEVC .mkv container :

http://www.demo-uhd3d.com/files/uhd4k/IMAX_PreShow.mkv

[P.J]

Who's gonna use windows media player again?

[NikosD]

It's been 8 months since my first comments about Optimus technology and HEVC decoding here:
http://forum.doom9.org/showthread.php?p=1694992#post1694992

Unfortunately, not a lot have changed since then.

I tried on a clean Win 8.1 Pro x64 - Core i5 4200M - dGPU Nvidia 740M - iGPU HD 4600 with drivers 4061 for Haswell and 347.88 for Kepler with no better luck.

I did my benchmarks again with latest LAV x64 0.65.9 in DXVA native mode and DXVA Checker 3.4.1 in Decode Benchmark mode for all 8 HEVC clips 1080p & 2160p.

The tests were done in battery mode and balanced power profile.


1.ProRes-1080p@30fps-2Mbps


LAV CPU 113/165/206 CPU 89%

LAV DXVA 740M 105/147/192 CPU 26%

LAV DXVA HD4600 74/110/131 CPU 45%




2.Elephants Dream-1080p@24fps-1.7Mbps


LAV CPU 54/168/289 CPU 76%

LAV DXVA 740M 0/145/260 CPU 28%

LAV DXVA HD4600 50/113/160 CPU 42%




3.Big Buck Bunny-1080p@60fps-2Mbps


LAV CPU 88/204/288 CPU 89%

LAV DXVA 740M 0/157/276 CPU 23%

LAV DXVA HD4600 83/133/160 CPU 44%




4. Girls -1080p60 fps -11Mbps


LAV CPU 75/123/175 CPU 87%

LAV DXVA 740M 73/115/173 CPU 27%

LAV DXVA HD4600 59/95/126 CPU 48%




5.Beauty-2160p@30fps-12.3Mbps


LAV CPU 28/35/39 CPU 73%

LAV DXVA HD4600 22/27/30 CPU 33%

LAV DXVA 740M SW fall-back



6.Fitness-2160p@30fps-8Mbps


LAV CPU 35/45/70 CPU 90%

LAV DXVA HD4600 27/33/43 CPU 61%

LAV DXVA 740M SW fall-back



7.Ducks-2160p@50fps-4Mbps


LAV CPU 45/51/52 CPU 95%

LAV DXVA HD4600 31/36/38 CPU 35%

LAV DXVA 740M SW fall-back



8.Gravity - 2048x858@24fps-5Mbps-10bit

The hybrid decoders of Intel and Nvidia don't support 10 bit HEVC decoding, so all decoders work but in SW fall-back mode.

Intel is still crashing with P010/P016 output enabled, while Nvidia has fixed the bug.



Comments:

1) Nvidia decided to cut the HEVC decoder above 1080p, at least on Kepler, so everything above that resolution falls back to SW.

2) As you can see from the minimum figures, Nvidia still has problems decoding HEVC streams with its hybrid decoder.
The issue remains unresolved for almost a year now, so I wouldn't trust that decoder for HEVC decoding - I don't know if Maxwell drivers are any better than this.

3) Intel is still using 100% GPU Load but with lower than max clock. Only 800-850MHz of 1150MHz max.
Even in High Performance mode or with the laptop plugged-in, the clock doesn't go any higher.

4) Nvidia, on the other hand has a full clock of 1058MHz but with low GPU utilization in Balanced mode ~60-65%
In High Performance or in plugged-in mode the utilization goes up to 90-100% and the performance rises.

This is definitely an improvement compared with older drivers, but the incompatibility problems and the cut-off resolution of 1080p, are far more serious issues.

[NikosD]

Well, this time it looks like they did it right, we only have to wait for Nevcairiel to build the 10bit decoder and see the performance on that.

Based on the previous first generation 4K H.264 decoder (VP5), they didn't manage to even reach 4K30fps.
VP5 is a 4K26-28fps decoder.

So, we didn't know exactly what to expect from Nvidia.

Kepler (600, 700 series)'s VP5 can easily reach 4K35fps~4K40fps.
Only GT520/610(GF119) can't reach 4K30fps, even GT520M(GF119 as well) can reach 4K31fps, my friend tested before.

I did a test to check those claims about Kepler's VP5 performance regarding 4K H.264 compared to Fermi's VP5, especially GT610.

Well, there is a small difference of about 10%, not worth to mention I think.

In playback mode with scaling 1280x720, Fermi was a little faster, probably an incompatibility of Kepler with the specific file.

I used same drivers 347.88 for Kepler and Fermi, LAV x64 DXVA 0.65.9 in DXVA native and DXVA Checker 3.4.1 in both Decode Benchmark mode and Playback benchmark mode.

The laptop with Optimus technology was used in both battery - balanced profile mode and plugged-in / high profile mode with no difference in results.

Decode mode


1. Chimei-inn-3840x2160@30fpsRef3-50mbps

Kepler 740M VP5 31/34/36

Fermi GT610 VP5 28/31/32


2. RED-4096x2160@24fps-36Mbps


Kepler 740M VP5 30/36/41

Fermi GT610 VP5 28/32/36



Playback mode


1. Chimei-inn-3840x2160@30fpsRef3-50mbps
1280 x720


Fermi GT610 VP5 25/28/30

Kepler 740M VP5 22/26/29



2. RED-4096x2160@24fps-36Mbps
1024 x 768

Kepler 740M VP5 25/31/40

Fermi GT610 VP5 25/29/32

[NikosD]

Socketed Broadwell for desktop is out:
http://www.tomshardware.com/reviews/intel-core-i7-5775c-i5-5675c-broadwell,4169.html

Unlocked CPU with lower clocks than top Haswell, only 65W TDP in 14nm and amazing GPU performance over HD4600 (more than double) by Iris Pro 6200@1150MHz vs HD4600@1250MHz.

In games it has the speed of R7 250 with a 12Watt GPU (!)

Hybrid HEVC(8bit/10bit) and VP9 decoding using CPU/GPU and overall a lot faster transcoding than Haswell (in theory)

CPU performance of Core i7 5775c ~ Core i7 4790.

Two problems:

1) Skylake is too close for release, that's why Intel decided to cannibalize Broadwell for desktop with only 2 models (!)
Core i7 5775C and Core i5 5675C

But Skylake for desktop could be a disappointment because it will have half-speed of Braodwell in GPU (24EUs vs 48EUs) and I read that no AVX-512 version will be included in desktop Skylake (only Xeons)

2) No LGA1150 chipset 8-series support for Broadwell. Only 9-series.

ARE YOU SERIOUS INTEL ?

They should support all LGA1150 motherboards, because only 2 SKUs are out and one of them is a Core i5.

Nobody with a 4790K will "update" to Broadwell, with lower CPU performance besides serious overclockers and nobody with a 9-series chipset will put a Core i5 Broadwell on it.

[nevcairiel]

Broadwell is not very interesting. Hybrid decoding is a stop-gap measure without much merit. Skylake will be out in a couple month (August seems to be the target) and offer full hardware decoding, so its really wiser to just wait.
I certainly will update my HTPC with a Skylake CPU later this year, although probably not the very high-end, its a HTPC afterall.

[NikosD]

August will be presented, but about October will be sold (!) as I read somewhere.

Skylake is not very interesting too.

About same CPU/GPU performance with Haswell with no AVX512.

Only fixed - function HW will be interesting but I will not buy a new board and new RAM DDR4 only for that.

We need AMD back to competitive CPUs and we need AMD NOW !

[P.J]

Only up to 4Kp30...

[NikosD]

No, it's not accurate just fail safe.

My HD4600 can already decode much better than that, with low bit rates.

I guess it can reach 4K60fps but with much lower than BluRay UHD bitrates.

[NikosD]

AMD's Carrizo is announced with HEVC HW decoding, but no performance figures yet.
It's only for laptops, though.

The milestone is the HSA 1.0 compliance, something that could potentially change the whole CPU/GPU subject.

The new HandBrake will have VCE support and Carrizo is faster in transcoding 1080p H.264 than Broadwell Core i5 - 5200U, according to AMD.

http://arstechnica.com/information-technology/2015/06/sixth-time-lucky-amd-details-the-carrizo-apu/1/

[foxyshadis]

Nothing supports AVX-512 as it is, so exclusion of it isn't exactly very damning, while the 24 EU count isn't confirmed anywhere and can't be considered anything other than wild ass guessing, at this point. Best anyone can do is sit tight and wait a month for real announcements before getting their hopes up (or dashed).

AMD just released HEVC hardware decoding. I'm looking forward to the benches from that.

[huhn]

even if they stay at the same EU number they can be still a lot faster.

[NikosD]

You are both wrong, but I won't report my internal source from Intel to prove you wrong.

Skylake for desktop will be released with 24EUs.

About performance, it's clear that 48 EUs with 128MB eDRAM is at least 2x faster than 24EUs.

Comparing to Haswell will be at most 20% faster, which is not worth to mention it.

[foxyshadis]

I don't doubt they'll be launched with 24 EUs -- launching from the bottom seems to be Intel's strategy ever since Merom -- but I don't expect it to stay that way for long. Broadwell started at 24 and now has 48, after all. It really seems like Intel is deliberately drawing out its performance part releases to let yields and cost catch up, instead of releasing $2000 chips and hoping for a market.

[blaubart]

The good news is that as a separate side-project I've been working on reducing the CPU usage of Copy-Back, and assuming a direct decoder -> output chain (no software deint, no format conversion, clean nv12 decode and nv12 output, or p010 decode and p010 output), I could reduce the CPU usage on my NVIDIA system by up to 50%, and additionally the benchmarked performance is now 99% the same as DXVA2-Native benchmarks. These gains may be a bit smaller on Intel or AMD GPUs, but they will still be quite noticeable.

I may also add a P010 decode -> NV12 output path using these optimizations, as that may be a common use-case for now when viewing 10-bit content on many renderers.

May I say thanks to nevcairiel,
though my 4K testTV LG 42UB820V probably not 10-bit at all, anyway, I got per DVBViewer (live TV) the Astra UHD Channel (50fps 3840x2160 HEVC) to run smooth and fluently (CPU 7%, GPU 30%) (Intel i7-4790K, Nvidia GTX 960).

Your "GTX960 @ dxva2cb direct" improvements made this possible for the very 1st time (even with an active DVBSource). Thank you again!
If you speak German... (http://www.dvbviewer.tv/forum/topic/55897-quo-vadis-htpc-richtung-hevc/?p=428725)

[P.J]

http://www.techpowerup.com/213575/amd-fiji-silicon-lacks-hdmi-2-0-support.html

[huhn]

the nano had a chance to be a really good HTPC card but something is totally wrong with AMD how can they miss that a lot of people buy nvidia 9xx cards just for the HDMI 2.0 cnnector.

[iSunrise]

the nano had a chance to be a really good HTPC card but something is totally wrong with AMD how can they miss that a lot of people buy nvidia 9xx cards just for the HDMI 2.0 cnnector.
They buy it since it's perfect for a HTPC, has good performance/efficiency and it has a fixed-function decoder for both H.264 and H.265/HEVC. Last but not least, yes, it also has HDMI 2.0/HDCP 2.2. It's not just because of the HDMI 2.0 connector, but that is part of the equation to support all upcoming 4K content.

Fiji was primarily build for PC gaming (DP only 1:16) and AMD just didn't have their decoder/display block updated for some reasons that are currently unknown.

[huhn]

most people don't buy a 960 for an HTPC. madVR user do but let's get real this is still a gaming card not a amd 5450/6450 media card.

you can't use TVs as your UHD display now with the nano and that's a very huge issue. normal UHD monitors are pretty crappy and small UHD TV are pretty cheap and have a way better PQ and AMD totally missed that market.

[iSunrise]

most people don't buy a 960 for an HTPC. madVR user do but let's get real this is still a gaming card not a amd 5450/6450 media card.
Well, it was you who claimed that "a lot of people buy nvidia 9xx cards just for the HDMI 2.0 cnnector." Since HDMI doesn't make much sense for PC, but only for TV I was just pointing out that HDMI 2.0 does only make sense for HTPC use. For everything else DP is the perfect solution, since even the highest-end monitors all do DP perfectly fine (Eizo CG series, NEC, etc.).

you can't use TVs as your UHD display now with the nano and that's a very huge issue. normal UHD monitors are pretty crappy and small UHD TV are pretty cheap and have a way better PQ and AMD totally missed that market.
While that is true, it not totally unusable. With HDMI 1.4a you can playback movies with 4K and up tp 30 fps perfectly fine, you just cannot use specific UHD-content like Netflix or other content providers that require HDMI 2.0 and HDCP 2.2 as copy protection. But you can playback everything directly from your PC with 4K and <=30fps just fine. So the nano still has a place for the HTPC market, but it's limited.

4K is still in it's infancy, there isn't even a lot of content for it right now. The nano is limited here, but it's not like it's totally unusable. Yes, it would be perfect if it had HDMI 2.0/HDCP 2.2.

It's nothing new, for several years now, AMD is behind when implementing state-of-the-art display/encoder/decoder blocks and that is a very sad thing. In the past, AMD was always at the forefront, way outclassing NV in video, but that time has been long gone.

[nevcairiel]

I think his point is that many people actually use small TVs (around 32-40") as PC displays, and they would greatly appreciate HDMI 2.0 for 4k@60.

[iSunrise]

I think his point is that many people actually use small TVs (around 32-40") as PC displays, and they would greatly appreciate HDMI 2.0 for 4k@60.
Most of the displays I know don't even support HDMI 2.0, yet, including the Philips 40" 4K-TV (VA panel), which is often used and is very popular (for TV and PC use).

But yes, I certainly agree, the nano will not have a very long life in it, I think we can safely say that. We can only hope that AMD knows that this is a serious limitation for a HTPC card and will not take too long to come up with a better successor.

[huhn]

even the last years samsung support 4:4:4 at 60 hz.

and yes some TV still have old 10 gbps chip for example all sonys and most Phillips if not all Phillips.

here an example: http://www.hdtvtest.co.uk/news/ue48h6400-201411133951.htm

EDIT: Phillips was well know for a good PC modes but this totally changed with the UHD TVs same for sony.

let's see when the totally pointless TP vision bashing starts.

[Yups]

You are both wrong, but I won't report my internal source from Intel to prove you wrong.

Skylake for desktop will be released with 24EUs.

About performance, it's clear that 48 EUs with 128MB eDRAM is at least 2x faster than 24EUs.

Comparing to Haswell will be at most 20% faster, which is not worth to mention it.


BDW GT3e won't be 2x faster unless GT2 Gen9 can't outperform a GT2 Gen7.5. Also AVX512 is uninteresting for us. Look to AVX2. Barely a 5% improvement in x264 and beside this basically no other productive consumer apps are available who can really make use of AVX2. No AVX512 is no loss and given that AVX2 was a bigger issue regarding tenmperature check tools like Prime95 it might be better.

[NikosD]

BDW GT3e won't be 2x faster unless GT2 Gen9 can't outperform a GT2 Gen7.5.


I can't follow your math here.

GT2 Gen7.5 has 20 EUs and GT2 Gen9 has 24 EUs, so GT2 Skylake will be a little faster than GT2 Haswell.

But Broadwell desktop has 48 EUs with 128MB super fast eDRAM, so it will be 2x faster than Skylake GT2, assuming that internal architecture of Skylake will not cover the eDRAM additional performance.


Also AVX512 is uninteresting for us. Look to AVX2. Barely a 5% improvement in x264 and beside this basically no other productive consumer apps are available who can really make use of AVX2. No AVX512 is no loss and given that AVX2 was a bigger issue regarding tenmperature check tools like Prime95 it might be better.

x264 is yesterday, it has practically stopped developing any more.

x265 is today and tomorrow and has already improvements of around 20% using AVX2 - IIRC - and it will be further optimised in the near future.

I'm not sure though if AVX512 would help even more.

Anyway, AVX512 just like AVX2 has more uses than HTPC and I always consider a more general view regarding my CPU upgrades.

[Yups]

I can't follow your math here.


Just look into some of the BDW GT3e reviews. There is roughly a 2x gap to HSW GT2. Your thinking is flawed. 2x of the EUs doesn't result in 2x performance because of other limitation, otherwise we would have seen a 2x gap on HSW GT2>HSW GT3(e).....it wasn't the case.



GT2 Gen7.5 has 20 EUs and GT2 Gen9 has 24 EUs, so GT2 Skylake will be a little faster than GT2 Haswell.


You have to assume that beside the rather small EU increase nothing changed regarding efficiency or other aspects of the GPU. For example there is a 1.5x increase in texture sampler from Gen 7.5 to Gen8 (GT2). You also have to assume that Intel didn't improve anything performance relating from Gen8 to Gen9. I highly doubt HSW Gen7.5 GT2 can match a Skylake Gen9 GT2.

[NikosD]

Just look into some of the BDW GT3e reviews. There is roughly a 2x gap to HSW GT2. Your thinking is flawed. 2x of the EUs doesn't result in 2x performance because of other limitation, otherwise we would have seen a 2x gap on HSW GT2>HSW GT3(e).....it wasn't the case.



I think your thinking is flawed.

Almost all of the comparisons I have seen they compare a GPU of 1100MHz (5675c) or 1150MHz (5775c) with a Haswell GPU of 1200MHz or most of the times 1250MHz.

So, yes if you compare a 48EUs 1100 MHz vs a 20 EUs 1250 MHz GPU the
difference couldn't be exactly doubled or even more than doubled, but we are playing with figures here around very small percentages near to x2+ performance.


You have to assume that beside the rather small EU increase nothing changed regarding efficiency or other aspects of the GPU. For example there is a 1.5x increase in texture sampler from Gen 7.5 to Gen8 (GT2). You also have to assume that Intel didn't improve anything performance relating from Gen8 to Gen9. I highly doubt HSW Gen7.5 GT2 can match a Skylake Gen9 GT2.

Since sandybridge and the first iGPU of Intel, the most critical parameter regarding GPU performance was always the number of EUs rather than internal architecture.

So I expect a GT2 Skylake to be around 20% faster than GT2 Haswell.

[huhn]

the 750 ti is using 6 watt so i can easily assume they measured the GPU only.

i have an r9 270 which can't dynamically change the clock speed of it's Vram only max or min which is 150/1400(nearly all other r9 270 should be able to do that). and i get a whole system of about 40 watt at the power supply! this doesn't match at all with this chart.

i will grab my power consumption meter and will check again. but there results looks fishy and they could be easily a problem with the new driver.

[wanezhiling]

http://tpucdn.com/reviews/AMD/R9_Fury_X/images/power_bluray.gif

AMD still haven't fixed their ridiculous power consumption when decoding Blu-ray, 66W is just wasteful.
:p
66 is not a high number, indeed.

[huhn]

:p
66 is not a high number, indeed.

these are really high numbers for GPU only but this german test shows fury x beats the 980 ti in term of power efficient hardware decoding but lose to the 980 for FHD and wins at UHD.

the hardware decoder of this card looks like a boss.

http://www.computerbase.de/2015-06/amd-radeon-r9-fury-x-test/8/

so i take all these numbers just as "numbers" right now.

over 100 watt for DXVA decoding there base test system must be very wasteful which is ok to make sure the bottleneck is the GPU not the CPU in a test.

[P.J]

Mine uses only 8W :D

I believe Intel cpus use even less than Radeon graphic cards :p

[wanezhiling]

http://i4.tietuku.com/05398b87a064af75.jpg

r9 380 (r9 285), tonga already supports hevc?????:eek:

[P.J]

http://i4.tietuku.com/05398b87a064af75.jpg

r9 380 (r9 285), tonga already supports hevc?????:eek:

No 10-bit and It should be hybrid.

[P.J]

http://www.game-debate.com/news/?news=17350&graphics=GeForce%20GTX%20950%20Ti%202GB&title=Nvidia%20Looks%20To%20Be%20Readying%20A%20GeForce%20GTX%20950%20Ti%20Graphics%20Card

[NikosD]

Added a post to article's comments that they must use latest nightly build of LAV filters in order to use 4K H.264 DXVA decoder.

The problem with all these reviews is that they don't say a lot about HTPC use of such cards.

Anandtech.com used to dedicate a separate article for HTPC use.

They could probably do it for Nano card.

[NikosD]

I was already disappointed by Skylake.
That was the final hit.

So, it's only Nvidia then.
BAD thing.

Let's hope that by the time BluRay UHD and other HEVC 10bit content become popular, all three will have decent 10 bit HW decoders.

[wanezhiling]

first 4k h264 card -> ancient gt520
first 8bit h265 card -> gtx960
first 10bit h265 card -> gtx960

[P.J]

Intel http://forums.mydigitallife.info/vb4_style/smilies/poop.png

[wanezhiling]

http://img1.mydrivers.com/img/20150710/dcbf81f3611847fea6403a3838977b0f.png

i7-6700K

[NikosD]

Impressive, but is 10bit HEVC supported by hybrid or pure HW acceleration ?

To be 100% sure, we have to wait for a review.

[P.J]

It should be the same as Broadwell.

[huhn]

looks like MS MF HEVC decoding with windows 10240 is broken. did someone have the same problem?

[CruNcher]

With H.264 it wasn't as clear but this time with H.265/HEVC the decoder in lavfilter is blowing away every commercial solution so far in a combined 10 Bit Software 8 Bit Hardware workflow Performance test.

Especialy vs the x64 version Commercial solutiions currently see no real land :)

Elecards x86 Demo has no chance even the performance of their x86 code path is not up to the 32 bit part of the lavfilter used decoder crazy :)

http://www.elecard.com/en/products/end-user-software/playback/hevc-decoder-plugin.html

[NikosD]

Hi there.
It's been a long time.

Where have you been ?

To be honest, the situation is exactly the opposite.

It's the first time that LAV/FFmpeg HEVC decoder is awfully slow at x86 version and the HW DXVA performance of PDVD 15 is a lot better as a total expereience.

x86 H.264 SW decoder didn't have such problem.

PDVD HW DXVA is faster, smoother and with better error handling at least with Intel GPUs and their hybrid approach.

[CruNcher]

on WDDM 1.1 it looks way different here for Nvidia :)

and the Software x86 paths of Elecard and co are even slower and jitter more overall

Benchmark is something different then a Playback test where you CPU overhead is crucial for output and stable realtime render frametimes

also im testing on the lower end i5-2400 where every cycle counts especially for the 4k 10 bit Playback test :)

im currently using mainly 1 test scene done by Atemes H.265 Broadcast Software Encoder one of the Samsumg Demo Streams aired on Astras Hevc Demo Channel

Samsung_Fashion_Show_Milan_UHD.ts

http://demo-uhd3d.com/fiche.php?cat=uhd&id=44

i find the beauty sample a little strange

http://forum.doom9.org/showpost.php?p=1730251&postcount=19407

Here is a Real Life x64 playback test of the 10 bit stream it's also a compare of the Efficiency of the Render core of both players @ the same time

http://forum.doom9.org/showpost.php?p=1730299&postcount=2638


the results of PDVD15 are really not so good either with HAM or DXVA for this specific config, but Nvidia seems to still optimize on the Driver ;)

Powerdirector 13 Demo has the 64 bit decoder will try that next

Arcsoft seems to have left the Consumer Video PC Buisness to concentrate on Mobile and Pro Solutions it seems :(

So it looks like this for the current updated Cyberlink Decoder Release

32 bit = 7 june 2015
64 bit = 5 January 2015

The 32 Bit DXVA result on the GTX 970 awefull playback experience :(

but still better then the Software Part of Cyberlinks decoder alone :)

http://abload.de/thumb/brakeups09j2u.png (http://abload.de/image.php?img=brakeups09j2u.png)

32 Bit Software Decoding 64 bit is virtually the same result

http://abload.de/thumb/softwaredecodetrbsl3.png (http://abload.de/image.php?img=softwaredecodetrbsl3.png)

Maybe there are some problems with MPC-HC and its handling of colorspace though between Cyberlinks Decoder output and the Nvidia Driver as Lav outputs in NV12 by default

Lav Filter x86 Software Decoding basicly the same MT under utilization result of Cyberlinks x86 as well as x64 code path

http://abload.de/thumb/lavvideo32tnbq8i.png (http://abload.de/image.php?img=lavvideo32tnbq8i.png)


And now Lav Filter x64 10 bit Software Decoding (Amazing) no MT brakeups everything absolute jitter free :)

http://abload.de/thumb/supercleani6kmy.png (http://abload.de/image.php?img=supercleani6kmy.png)

@NikosD

What is the Decoding result with Intels Hybrid mode of that 4K@25Fps 10 Bit 50 mbps clip when you further restrict it to 4 cores ?


Intel's hybrid decoder doesn't support 10 bit HEVC decoding, so all decoders work but in SW fall-back mode with the exception of Lentoid OCL which crashed.

ahh this is still the case ?

[NikosD]

From your post you seem to forget two very interesting decoders.

Lentoid x86/x64 and Microsoft's x86/x64 HEVC decoder.

Lentoid x86 is the fastest x86 HEVC decoder, with a serious limitation of no 10bit decoding.

MSFT depending on the CPU architecture is also very fast and for x86 generally faster than LAV x86.

For DXVA decoding, PDVD 15 is generally smoother than LAV in real life playback.

When I use the word faster, I mean benchmarks which is translated to lower CPU usage in playback normal mode.

When I use the word smoother I always mean real time playback.

Regarding OCL (HAM) PDVD 15 and other OpenCL HEVC decoders are not as good as DXVA decoders.

So, as I wrote above I use PDVD DXVA mode for hybrid real time playback experience.

About your question, generally speaking using a hybrid DXVA mode for decoding isn't affected by the number of cores when I use a CPU like core i7-4790.

Hyperthreading has a real impact in CPU decoding, not hybrid.

Your question is not valid because you are asking about hybrid decoding of a 10 bit clip, which means SW decoding, because there is no support of 10bit decoding in HW or shaders for Haswell

[CruNcher]

im only considering 10 bit supporting decoders that can decode that test clip if they can't they useless ;)

so lentoid and uhdecode are not part of any testing because 8 bit Hardware Decoding (doesn't really matter if Copyback,Native or CUVID only some exceptions currently as with the beauty clip) is absolutely efficient with Lav Filter alone

PowerDVD 15 main Application is currently broken for me (Maxwell V2) i can't test that get no output currently most probably a Nvidia Driver issue with their renderer

The good old Arcsoft Theatre 6 still works without any issues

MSFT Decoder is no option because WDDM 1.1 testing only most probably it's based on Mainconcepts HEVC SDK anyway ;)

yeah i saw your post about intels hybrid decoder for haswell has no 10 bit support so you cant test that.