hardware accelerated H.264 decoding when using NVIDIA GPUs

From the change log of R14 b4:
"Added support for hardware accelerated H.264 decoding when using NVIDIA GPUs on Linux and Windows"

I have rerun the test I did here: viewtopic.php?f=21&t=58667#p337838 and I see no improvements at all in b4 compared to b1, b2 and b3 for h264 4k 50p and 60p

i7 7700HQ, 16 GB, GTX 1070 8 GBVram, M.2 SSD, Windows 10, Resolve v14 b4 free version

Is this HW decoding feature enabled in Studio only?
I read somewhere that apparently the perf of h264 is better in the Studio version, is it true?
Anybody is able to at least playback smoothly h264 4k at 50 or 60 fps within the media or edit tab?

thx a lot

Its on in the free version, I have a 1070, and I can definitely see the improvement.

Your cpu has onboard graphics, I bet DR is seeing that instead of the 1070.
https://ark.intel.com/products/97185/In ... o-3_80-GHz

Its on in the free version, I have a 1070, and I can definitely see the improvement.

Your cpu has onboard graphics, I bet DR is seeing that instead of the 1070.
https://ark.intel.com/products/97185/In ... o-3_80-GHz

It is disabled the iris gpu and I even forced on the preference settings CUDA and manually selected the 1070. Btw on the same machine I can preview MJPG material on a 4k 60 fps timeline with a few grading nodes and even optical flow slow motions in real-time. It would be impossible with the onboard intel iris GPU....

Did you try with 50 or 60 fps 4k h264 material?

Sorry, that's not what I mean. I'm a developer so I have a decent idea of how something like this could go awry. The new feature might not be fully integrated into the old code base. It could be a stand alone module/class that reviews the hardware and makes its own determination if decoding should be offloaded. Mistakes/oversights like this are easy to make when a team of people is substantially refactoring a code base, hence the reason for beta testing.



I've only tested with UHD 30p, but it's still a night and day difference. While on the edit tab, I would say its only slightly slower (10's of ms) than dnxhr, when clicking around the timeline. However it will still spitt and sputter if you start going crazy with the scrubber bar. For reference I'm running a 6850k overclocked to 4.2 GHz.

Sorry, that's not what I mean. I'm a developer so I have a decent idea of how something like this could go awry. The new feature might not be fully integrated into the old code base. It could be a stand alone module/class that reviews the hardware and makes its own determination if decoding should be offloaded. Mistakes/oversights like this are easy to make when a team of people is substantially refactoring a code base, hence the reason for beta testing.



I've only tested with UHD 30p, but it's still a night and day difference. While on the edit tab, I would say its only slightly slower (10's of ms) than dnxhr, when clicking around the timeline. However it will still spitt and sputter if you start going crazy with the scrubber bar. For reference I'm running a 6850k overclocked to 4.2 GHz.

Simply playing the material back on a 4k timeline in the edit tab with no grades how much CPU usage do you have?

Simply playing the material back on a 4k timeline in the edit tab with no grades how much CPU usage do you have?

I just did a quick test now that I'm home.

The cpu fluctuates between 25% and 35%, and spends most of its time in the 27% to 30% range. However all the cores aren't running at max frequency. Usually they all sat 3.1 or 3.6 GHz and occasionally one would drop down to 1.2 or 2.4, and occasionally one would jump up to 4.2GHz

For the gpu I was getting:
Load : 10% to 20%
Memory usage : steady at 29.7%
Memory controller load : 4%
video engine load : 0%
bus load : 3%

[Rohit Gupta]

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

Wow, this could finally be an end to threads where people complain that H.264 playback sucks on their machine.

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

If there were a way to buy the Studio version and not be stuck using a dongle, you would have my $299 this minute!

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

If there were a way to buy the Studio version and not be stuck using a dongle, you would have my $299 this minute!

You may already know this, but with 14 the intent is to move away from dongles and use licenses of some kind for activation. Exciting for dongle-haters like me.

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

Wow, if you're not off loading anything to the gpu in the free version you have made massive improvements to the cpu based decoder, because in 12.5 h.264 was unusable.

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

If there were a way to buy the Studio version and not be stuck using a dongle, you would have my $299 this minute!

You may already know this, but with 14 the intent is to move away from dongles and use licenses of some kind for activation. Exciting for dongle-haters like me.

What I want to do is to be able to have the Studio version in beta 14 without having to buy 12.5 Studio first with a dongle.

If it were possible to simply buy dongle-less Studio with access to beta 14, I would buy right now, this very minute. As it is, it appears that I will have to wait until 14 is released. Please, Blackmagic, make this possible and take my money now!

is the Nvidia h.264 hardware offloading limited to 8bit level+profiles or also for 10bit like from the GH5? I remember Nvidia having software issues with the MPEG profiles used.

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

Cool, thx for the answer it now all make sense.

As suggestion it would be helpful to better document that is a Studio only feature.
Even better would be that in the free version there would be a settings to enable/disable h264 HW decoding. Naturally when enabled it will display the watermark and point you to upgrade. Exactly like all the others Studio features. It makes the feature more discoverable and easier to test and at the end it doesn't confuse users.

+1 for the dongle free Studio version asap.

It is disabled the iris gpu and I even forced on the preference settings CUDA and manually selected the 1070. Btw on the same machine I can preview MJPG material on a 4k 60 fps timeline with a few grading nodes and even optical flow slow motions in real-time. It would be impossible with the onboard intel iris GPU....

H.264 AVC and H.265 HEVC 8bit and 10 bit decoding including HDR and Wide Color Gamut (Rec.2020) using the Intel Iris GPU should be even better than CUDA and is now being used by Edius Pro and Magix (who now own Sony Vegas).
https://software.intel.com/en-us/videos/how-magix-made-4k-360-video-editing-swift-and-easy
The Kaby Lake-U/Y GPU - Media Capabilities - Intel Launches 7th Generation Kaby Lake: 15W/28W with Iris
"Intel claims that Kaby Lake-U/Y can handle up to eight 4Kp30 AVC and HEVC decodes simultaneously. HEVC decode support is rated at 4Kp60 up to 120 Mbps"
"The major feature change in the Kaby Lake-U/Y media engine is the availability of full hardware acceleration for encode and decode of 4K HEVC Main10 profile videos."
http://www.anandtech.com/show/10959/intel-launches-7th-generation-kaby-lake-i7-7700k-i5-7600k-i3-7350k/6
H.264 4k editing with 4k realtime preview may be possible on a consumer level Kaby Lake Laptop/Desktop with no expensive discrete CUDA card - and H.265 4k is here in the GH5 and DJI cameras.

It is disabled the iris gpu and I even forced on the preference settings CUDA and manually selected the 1070. Btw on the same machine I can preview MJPG material on a 4k 60 fps timeline with a few grading nodes and even optical flow slow motions in real-time. It would be impossible with the onboard intel iris GPU....

H.264 AVC and H.265 HEVC 8bit and 10 bit decoding including HDR and Wide Color Gamut (Rec.2020) using the Intel Iris GPU should be even better than CUDA and is now being used by Edius Pro and Magix (who now own Sony Vegas).
https://software.intel.com/en-us/videos/how-magix-made-4k-360-video-editing-swift-and-easy
The Kaby Lake-U/Y GPU - Media Capabilities - Intel Launches 7th Generation Kaby Lake: 15W/28W with Iris
"Intel claims that Kaby Lake-U/Y can handle up to eight 4Kp30 AVC and HEVC decodes simultaneously. HEVC decode support is rated at 4Kp60 up to 120 Mbps"
"The major feature change in the Kaby Lake-U/Y media engine is the availability of full hardware acceleration for encode and decode of 4K HEVC Main10 profile videos."
http://www.anandtech.com/show/10959/intel-launches-7th-generation-kaby-lake-i7-7700k-i5-7600k-i3-7350k/6
H.264 4k editing with 4k realtime preview may be possible on a consumer level Kaby Lake Laptop/Desktop with no expensive discrete CUDA card - and H.265 4k is here in the GH5 and DJI cameras.

The impossible paragraph is in the context of multiple grades and optical flow on a MJPG source on 4k timeline. It was not meant to be about h264 or h265 decoding....

With the windows media player even on some Atom based CPU I can playback 4k 50p in real-time but if and how black magic will use this feature is all to be seen.

The good news is that they now support HW decoding on CUDA and considering that for real work Resolve really needs a good GPU I see no problem with this approach.

The good news is that they now support HW decoding on CUDA and considering that for real work Resolve really needs a good GPU I see no problem with this approach.

To my knowledge the HW decoding/encoding is done on a separate dedicated part of GPU, not CUDA processing pipeline. No that it matters much practically, although it will free the CUDA cores for other work.

With the windows media player even on some Atom based CPU I can playback 4k 50p in real-time but if and how black magic will use this feature is all to be seen.

Because it's done by dedicated chip which is there only for task of decoding h264/h265 etc. This chip will guarantee you some performance- e.g. 4K up to 80fps (it also depends on bitrate). Some files may struggle also, but typical ones should always work fine.
CPU itself in software mode would not be able to do it.

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

Now we understand one of the reasons, so the studio version is paid and what benefits it brings, virtually no limits.
thank you

decode one stream or multiple 4K25p 4:2:2 10bit video streams? Multicam editing has become the norm here.

Studio version only feature. Even ... the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

For those of us with a Studio dongle with 12.5, will the V14 Studio beta support GTX H.264 decoding when we use the dongle? If the dongle doesn't work -- how do we trial V14 beta Studio?



Question: how is h.264 decoding performed -- without a GTX -- differently under Studio than under Free? I assume the CPU is used in both cases. For example, is Free single threaded?

I think it uses different decoder (which is licensed from mainconcept)- free version does not support 10bit at all. Because you have to pay licensing fees if you give software for free it means you are actually loosing money. For 8bit you rely on "system" decoder- either provided by Win or OSX (which is free).

Studio version only feature. Even otherwise the H.264 performance on Studio version is a lot snappier than the free version. On Windows.

Does this feature depend on the generation of nvidia gpu? If so, which generations are supported?

I wanna know when AMD (RX580) is going to see the same love that NVIDIA is getting?

[Peter Chamberlain]

The Kepler generation and newer NVIDIA GPUs support this with the more recent releases having better performance. We don't have AMD acceleration at this time.

Cant find the page for Resolve vs Resolve Studio comparison on the black magic site. Previously it was on the main Resolve page (I think). Looks like in version 14 there will be more differences? Has the comparison page been removed?

Is there a difference between free and studio version on the Mac in this respect?

Old thread but important to me. Any updates; free vs studio? Nvidia cards that suoort the new feature

bob

[Peter Chamberlain]

You need the Studio version and a Kepler generation and newer NVIDIA GPU to support this feature. With the more recent GPU releases having better performance. This is not likely to change.

I believe I remember reading somewhere that Windows was required as well. Is that the case?

You need the Studio version and a Kepler generation and newer NVIDIA GPU to support this feature. With the more recent GPU releases having better performance. This is not likely to change.

Thanks, Peter

Where could I find a list of compatible cards? I doubt my 1070 will qualify.

bob

The 1070 does qualify.

The 1070 does qualify.

I'm on vacation and not at my desktop but rethinking here, I might have the 970, $150.00 card.

Not quite awake

thx

bob

I would have answered but as we do not know if it is on PC or MAC or LINUX: I abstain ...

I would have answered but as we do not know if it is on PC or MAC or LINUX: I abstain ...

If that's at me. Win 7 PRO, i7 6700 32 Ram

bob

Upgrade Windows. DR is not meant to be run under W7.

In this thread, it is stated that for getting hardware accelerated h.264 support, we need:
- Davinci Resolve Studio version (not free)
- an Nvidia GPU

Did this change since?
- Is that feature available on the Free version?
- Are the AMD GPUs also supported?

Thanks

Uhm I read somewhere that Cuda acceleration is on decoding 8bit only h264 4:4:4 or 4:2:0
Not 10 bit, not 4:2:2
I cannot give you more feedback be cause I ever convert all in cineform or prores to avoid many studdering, less time than process and optimize structure later


Inviato dal mio iPhone utilizzando Tapatalk

Bit depth and sub sampling capabilities when using hardware acceleration for encoding/decoding h.264/h.265 depends entirely on the hardware. Recent Nvidia cards do 10-bit just fine, but I don't think they do 4:2:2 (but 4:4:4).

AMD GPU support is enabled in v16 beta according to the release notes.

GPU decoding is only for 8bit h264, 8/10/12bit for h265 4:2:0. It's not working for any 4:2:2. If your files have very high bitrate it may also not work well.
Turning based cards (GTX2xxx, GTX1660) also support 4:4:4 (8-12bits) h265 decoding, but again no 4:2:2.

Only Turing cards support B frames in h265 encoding, so files from older cards use just I/P frames which is far from being good. Lossless encoding is also supported for h264 and h265 (not sure about decoding, but looks like it's not supported at all).

...and rendering will be 10x faster in the Studio version for h.264/265