BMD adds HEVC 422 hardware decoding for Intel (+482 update)

Comments

lenard wrote on 3/17/2021, 7:01 PM

Support for 10-bit video is crucial as this unlocks meaningful SLOG grading workflow and HDR.

Yes ofcourse, and it makes their editor incompatible with the most popular phone manufacturer in the world, it's an extremely embarrassing situation for Magix, and they are in no hurry to fix it. We've seen countless users change editors because of it. IT'S CRAZY and self destructive

 

Ps, I tried the 422 Otter Import >4444 Prores conversion and timeline playback is ~15fps with 20% nVidia & 50% CPU usage on my machine. Won't be using that.

Did Lenard tell you to do that?

Lenard said to use Shutter Encoder to create Prores422, Lenard then went on to show you how good it works on a lowly $200 cpu in comparison to AVC. Has Lenard steered you wrong the in the past?

RogerS wrote on 3/17/2021, 7:36 PM

HEVC is superior as an acquisition format but not an editing format. Try 10 bit AVC which works today.

Why use ProRes 444 instead of 422? File sizes get massive and performance suffers.

JN- wrote on 3/17/2021, 8:25 PM

@JN- both those cameras are not on my radar, but support is crucial. And if one pro editor supports it, Vegas should too.

@Reyfox I agree for sure. In some other thread somewhere, I saw MagixDerek quoted as that they didn’t get it into this update, but working on it, so I'd guess it’ll be in the next one.

Everyone has their own ideas of workflow for sure, personally if I have a need to use the current build 482 I use my own ffmpeg based util to convert hevc to Prores 422 HQ. It then plays really smoothly in VP18.

Also, the conversion rate on my laptop for a 5 second test sample 4K hevc clog 422 10 bit file was ~ 1.4x, a lot lot faster than when I previously converted to mp4. It was nearly 20x clip duration. My PC should chew through it. So until the fix is in, plenty of options, including VP16.

Last changed by JN- on 3/17/2021, 8:48 PM, changed a total of 3 times.

---------------------------------------------

VFR2CFR, Variable frame rate to Constant frame rate link to zip here.

Copies Video Converts Audio to AAC, link to zip here.

Convert 2 Lossless, link to ZIP here.

Convert Odd 2 Even (frame size), link to ZIP here

Benchmarking Continued thread + link to zip here

Codec Render Quality tables zip

---------------------------------------------

PC ... Corsair case, own build ...

CPU .. i9 9900K, iGpu UHD 630

Memory .. 32GB DDR4

Graphics card .. MSI RTX 2080 ti

Graphics driver .. latest studio

PSU .. Corsair 850i

Mboard .. Asus Z390 Code

 

Laptop… XMG

i9-11900k, iGpu n/a

Memory 64GB DDR4

Graphics card … Laptop RTX 3080

lenard wrote on 3/17/2021, 8:50 PM

HEVC is superior as an acquisition format but not an editing format. Try 10 bit AVC which works today.

 

Because 10bit AVC is not GPU decoded by anything (consumer based) you need a lot of CPU, 4K60 it's really not useable on my ryzan3600 6 core, 4K30/24, it's more useable, but not pleasant to edit with. I am happy to see Vegas maxing out my CPU though, unlike AVC that uses GPU decoding, where it can drop frames, but not be using all my cpu or gpu. With an 8core 4K10bit AVC likely works very well, but if you have to transcode I don't see any advantage to 4k10bit 420 AVC over 4k10bit 422 prores. Prores file size is larger, but that's really only a problem for acquisition . Speed of transcode and playback performance should favour prores422

 

AVsupport wrote on 3/17/2021, 9:50 PM

Why use ProRes 444 instead of 422? File sizes get massive and performance suffers.

Indeed that was the wrong test to make, I wanted to explore 'uprezzing' to HEVC 444 as this is in theory NVidia supported today by Turing chipsets. Mind you, I cannot test nor create. Estimate residual file size not being much bigger than 60% more?

@lenard , ProRes isn't gonna help me much as it's CPU decode only on my machine, playback rate for the 10bitHEVC4K422>Prores422 clip is 15 fps on an 8bit/1080 timeline @422Mbps. That's not much better news than my XAVCS-Intra 10Bit422 @470MBps which plays back with 9fps.

Personally, I don't get much about the hype with ProRes; for me this has been primarily an Apple format, and VPs strength was always PC based. I'm not keen on changing systems and dropping $3000 on an Apple Afterburner card that only works in certain imacs.

Last changed by AVsupport on 3/17/2021, 10:45 PM, changed a total of 2 times.

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.

lenard wrote on 3/17/2021, 11:38 PM

 

@lenard , ProRes isn't gonna help me much as it's CPU decode only on my machine, playback rate for the 10bitHEVC4K422>Prores422 clip is 15 fps on an 8bit/1080 timeline @422Mbps. That's not much better news than my XAVCS-Intra 10Bit422 @470MBps which plays back with 9fps.

I can replicate your 15fps by turning on legacy AVC, which is how prores used to play on vegas. Ofcourse the improvements are to so4 reader, not legacy. I have dynamic ram set to 1000, but I don't think that's relevant

fr0sty wrote on 3/17/2021, 11:47 PM

Personally, I don't get much about the hype with ProRes; for me this has been primarily an Apple format, and VPs strength was always PC based. I'm not keen on changing systems and dropping $3000 on an Apple Afterburner card that only works in certain imacs.

I'm able to edit 4K ProRes 422 multicam in VEGAS 18 without proxies with a full resolution QFHD preview at full frame rate, that's what's so good about it. No HEVC format can touch that, even with hardware decoding in VEGAS. Not to mention, ProRes is higher quality than HEVC.

Last changed by fr0sty on 3/17/2021, 11:47 PM, changed a total of 1 times.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

RogerS wrote on 3/18/2021, 12:18 AM

I find the same as Frosty, very good performance with 422 without GPU decoding and acceptable file sizes. No specialty hardware needed. I also use MagicYUV which plays back well though file sizes can be high. I am using so4compound to decode.

Marco. wrote on 3/18/2021, 3:00 AM

"Because 10bit AVC is not GPU decoded by anything (consumer based)"

This seems to depend on various properties, though your generalized claim seems not to be true.

For HD video, on my system GPU usage of decoding of 10 bit 4:2:0 AVC is between 45 % and 70 % and decoding of 10 bit 4:2:2 XAVC-I (which is a subset of AVC) is between 25 % and 50 %.

For 4k video, on my system GPU usage of decoding of 10 bit 4:2:0 AVC is between 0 % and 10 % and decoding of 10 bit 4:2:2 XAVC-I is between 20 % and 35 %.

AVsupport wrote on 3/18/2021, 3:46 AM

@fr0sty @RogerS this is 10-Bit ja? is it that you see any iGPU activity that I don't have?

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.

lenard wrote on 3/18/2021, 5:57 AM

"Because 10bit AVC is not GPU decoded by anything (consumer based)"

This seems to depend on various properties, though your generalized claim seems not to be true.

For HD video, on my system GPU usage of decoding of 10 bit 4:2:0 AVC is between 45 % and 70 % and decoding of 10 bit 4:2:2 XAVC-I (which is a subset of AVC) is between 25 % and 50 %.

For 4k video, on my system GPU usage of decoding of 10 bit 4:2:0 AVC is between 0 % and 10 % and decoding of 10 bit 4:2:2 XAVC-I is between 20 % and 35 %.

What this discussion is about is hardware decoding, aka ASIC decoding, not GPGPU computing , there is ASIC decoding for 10bit 420 HEVC, but not for 10bit 420 AVC, this is true for Amd/Intel/Nvidia. 1 of @AVsupport's complaints are that although his Nvidia GPU has HEVC 10bit 420 decoding, it doesn't work on Vegas Pro, the fault is a Magix one.

This seems to depend on various properties, though your generalized claim seems not to be true.

No, it's binary, ASIC decoder is in operation, or it's not. I did an extreme test, 4K120p played back at 120fps. 8bit AVC vs 10bit AVC. Where there is hardware decode, there is low cpu and high fps, and in this extreme example where there is no hardware decode there is a black screen. I then re position the playhead on the 10bit AVC and allow it to cache, it then plays back at 100% CPU at half the frame rate

The reason for even talking about this is because @RogerS keeps recommending 10bit AVC, but this demonstration could just as well be a 10bit vs 8bit 420 HEVC, with VP18b452 and that should not be the case

Marco. wrote on 3/18/2021, 6:10 AM

That doesn't make your generalized claim any better or more true.

lenard wrote on 3/18/2021, 6:35 AM

Ok, what this thread is about is this and what codecs it works for

The title of this thread BMD adds HEVC 422 hardware decoding for Intel (+482 update), is about using the new intel asic that decodes 422 HEVC. This is not a generalised conversation, but a specific one. Yes the 10bit 420 AVC is aided by GPU compute, just as the 8bit is, nobody said otherwise

 

RogerS wrote on 3/18/2021, 7:12 AM

@fr0sty @RogerS this is 10-Bit ja? is it that you see any iGPU activity that I don't have?

No, no iGPU activity for ProRes. 32-bit ACES now has some GPU acceleration, apparently. Frosty does HDR work (I don't). I can run tests later.

Update: just tested a Z-Cam 4K ProRes 4:2:2 file in VP 18 8-bit full and 32-bit video and full modes. It decoded at a full 30fps with no drops in each mode.

Last changed by RogerS on 3/18/2021, 8:08 AM, changed a total of 2 times.

Custom PC (2022) Intel i5-13600K with UHD 770 iGPU with latest driver, MSI z690 Tomahawk motherboard, 64GB Corsair DDR5 5200 ram, NVIDIA 2080 Super (8GB) with latest studio driver, 2TB Hynix P41 SSD and 2TB Samsung 980 Pro cache drive, Windows 11 Pro 64 bit https://pcpartpicker.com/b/rZ9NnQ

ASUS Zenbook Pro 14 Intel i9-13900H with Intel graphics iGPU with latest ASUS driver, NVIDIA 4060 (8GB) with latest studio driver, 48GB system ram, Windows 11 Home, 1TB Samsung SSD.

VEGAS Pro 21.208
VEGAS Pro 22.239

Try the
VEGAS 4K "sample project" benchmark (works with VP 16+): https://forms.gle/ypyrrbUghEiaf2aC7
VEGAS Pro 20 "Ad" benchmark (works with VP 20+): https://forms.gle/eErJTR87K2bbJc4Q7

Howard-Vigorita wrote on 3/18/2021, 11:26 AM

I'm able to edit 4K ProRes 422 multicam in VEGAS 18 without proxies with a full resolution QFHD preview at full frame rate, that's what's so good about it. No HEVC format can touch that, even with hardware decoding in VEGAS. Not to mention, ProRes is higher quality than HEVC.

@fr0sty The last time I compared 4K ProRes 422 in Vegas the file sizes were dramatically larger than hevc 420 and the visual quality and playability was dramatically lower. My impression using 4k 10-bit hevc 420 as my default format of choice since then is that Vegas 18 runs it even better... I have the original clips with me at the moment but not the desktop... will rerun this on the same hardware when I get home again:

fr0sty wrote on 3/18/2021, 12:06 PM

The exact opposite has been my experience. HEVC playback sucks on my system without hardware acceleration, even with it the playback isn't as smooth as prores.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

JN- wrote on 3/18/2021, 12:21 PM

Just to add to my comment's, on a previous page, when using a stand in hevc clog clip.

Tested using my PC.

With the OP's original 26s clip, (thanks for the link wwaag) converted to Prores 422 HQ the file size is 1.05Gb. The hevc mp4 file size is 353 MB. The data rates are 346,278 kbps Prores, and 98,758 kbps for the hevc file.

The FPS playback rates with no proxy assistance are ...

hevc between 1.6 and 2, but mostly less than 2 FPS using VP16.

Prores 50 FPS using VP18 b482.

I know which I would prefer to use.

Time taken to convert the 26s clip to Prores 422 HQ was ~ 20s.

Last changed by JN- on 3/18/2021, 12:43 PM, changed a total of 2 times.

---------------------------------------------

VFR2CFR, Variable frame rate to Constant frame rate link to zip here.

Copies Video Converts Audio to AAC, link to zip here.

Convert 2 Lossless, link to ZIP here.

Convert Odd 2 Even (frame size), link to ZIP here

Benchmarking Continued thread + link to zip here

Codec Render Quality tables zip

---------------------------------------------

PC ... Corsair case, own build ...

CPU .. i9 9900K, iGpu UHD 630

Memory .. 32GB DDR4

Graphics card .. MSI RTX 2080 ti

Graphics driver .. latest studio

PSU .. Corsair 850i

Mboard .. Asus Z390 Code

 

Laptop… XMG

i9-11900k, iGpu n/a

Memory 64GB DDR4

Graphics card … Laptop RTX 3080

fr0sty wrote on 3/18/2021, 1:28 PM

Not only that, but it maintains 10 bit, 4:2:2, while also having performance on par with GPU accelerated HEVC. ProRes didn't use to decode this well in VEGAS, it does now.

Last changed by fr0sty on 3/18/2021, 1:36 PM, changed a total of 1 times.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

Howard-Vigorita wrote on 3/18/2021, 1:34 PM

The exact opposite has been my experience. HEVC playback sucks on my system without hardware acceleration, even with it the playback isn't as smooth as prores.

@fr0sty I think it's all about decoding which hevc performance benefits from... looks like all your hw decoding is on the same gpu as everything else. My Xeon system has no igpu either so I threw in a 2nd gpu (Nvidia 1660) assigned to decoding and it helped... but maybe not as much as an onboard Intel would. I suspect an amd g-series cpu would have a similar performance advantage... if anyone could actually get one. Btw, I just benched v17 and v18 on a Nuc rendering 30 sec hevc and ProRes clips. Wow! Deserves it's own thread.

AVsupport wrote on 3/18/2021, 5:06 PM

@RogerS it would make sense in this discussion to compare 10-bit source material, if you can.

As for Prores vs HEVC, It's a little difficult to compare effective performance because there is no decent VP18 nvidia HW acceleration so this is hypothetical. I would expect that at similar data rates, perhaps a quick HEVC transcode to HEVC Intra , with a decent GPU, the HEVC would fare better. Transcoding HEVC to Prores would add a de-generation through the reencoding process, apart from the workflow step.

Prores is intra also, which would explain one apect of its fast processing. But there's no hardware acceleration for it, so you are solely relying on and burning CPU cycles

Last changed by AVsupport on 3/18/2021, 5:08 PM, changed a total of 2 times.

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.

lenard wrote on 3/18/2021, 10:54 PM

@RogerS it would make sense in this discussion to compare 10-bit source material, if you can.

As for Prores vs HEVC, It's a little difficult to compare effective performance because there is no decent VP18 nvidia HW acceleration so this is hypothetical. I would expect that at similar data rates, perhaps a quick HEVC transcode to HEVC Intra , with a decent GPU, the HEVC would fare better.

 

Prores and HEVC(INTRA) + HW acceleration perform the same, without HW acceleration Prores is much more efficient. These clips are 420 8bit to get a general idea of how vegas handles HEVC, I would guess the HEVC420(INTRA) used should perform the same as 444 with a turing card if vegas ever supported it. You could test that theory with other players/editors

clip1 - prores, clip2 HEVC(intra) , clip3 HEVC (LongGOP)

It's also interesting to see how poor the vegas HW decoder works at 4K with LongGOP, but works fine with INTRA

 

 

 

AVsupport wrote on 3/19/2021, 12:14 AM

Since we love going around in circles: NVidia claimed the ability to decode 157 HEVC 1080/30 clips simultaneously with their latest chipset.

worst case this is 8 bit, hence 20%less streams than 10-bit = 125 clips.

if these are indeed only 420 and not 422 then this would again reduce the amount to another 20% less = 100 clips.

lets say you want 4K which is 4x the data of 1080 = 25 clips.

not sure if you could do that with Prores.

Last changed by AVsupport on 3/19/2021, 12:23 AM, changed a total of 1 times.

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.

fr0sty wrote on 3/19/2021, 2:36 AM

But you can't do any clips in 10 bit 422 on Nvidia, so...

Last changed by fr0sty on 3/19/2021, 2:37 AM, changed a total of 1 times.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

AVsupport wrote on 3/19/2021, 2:52 AM

You’re absolutely correct @fr0sty but one could uprez to 444 and still get impressive results, maybe 15 clips??

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.