Quick look at GTX1070 Encode Performance with PD14

JL_JL

Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline

Solved by

Jun 14, 2016 23:07

Presented is an initial comparison of GeForce GTX1070 with a few other popular PD14 editing GPU’s. Many forum contributors have made claims as to how good it should be but no one has posted any relevant data to compliment claims.

One of the main advantages for the 1070 is a new NVENC ASIC IP encoding block which is supposed to significantly improve encoding performance for 4K H.264 & HEVC when compared to previous generation NVENC. Some limited history of NVENC versions here: https://en.wikipedia.org/wiki/Nvidia_NVENC As reported in many other performance assesments, until timeline scrubbing is changed within PD, from what I have seen high end cards provide little relief for fluid timeline playback of compressed codecs for either complex timelines or multicam playback.

The main benefit of such cards comes from those who wish to use the GPU hardware encoder vs the CPU software encoder with PD14 for significant improvement in encode times. As many have seen, real performance varies dependent on PD internal architecture for which end users can only evaluate given hardware and see how it performs against some common timeline edits. Several have identified bottlenecks that consistently throttle overall throughput. The attached table shows some performance ratios for basic NVENC VE encoding and OpenCL loads. Pretty much as expected. When the GPU is loaded substantially for basic pure encoding, the GTX1070 is very beneficial, ~1.3x faster than the 960/970 (same basic generation of NVENC).

Jeff


Filename	PD14_GTX1070a.png	Download
Description
Filesize	37 Kbytes
Downloaded:	671 time(s)

tomasc

Senior Contributor Joined: Aug 25, 2011 12:33 Messages: 6464 Offline

Jun 15, 2016 01:19

It is interesting that your chart show that the GTX970 better than the 960 in h.264 encoding. The new GTX1070 does not surpass the 970 in every test. Maybe the GTX1070 will surpass both earlier cards in the hevc tests.

This message was edited 1 time. Last update was at Jun 15. 2016 01:32

ynotfish

Senior Contributor Location: N.S.W. Australia Joined: May 08, 2009 02:06 Messages: 9977 Offline

Jun 15, 2016 03:13

Thank you for posting your observations Jeff.

It's good to have real PD-related information in front of "one" - something you don't get when you read the reviews & blurbs.

Cheers - Tony
Visit PDtoots. PowerDirector Tutorials, tips, free resources & more. Subscribe!
Full linked Tutorial Catalog PDtoots happily supports fellow PowerDirector users!

SoNic67

Senior Contributor Joined: Sep 27, 2014 14:14 Messages: 1308 Offline

Jun 15, 2016 05:21

Thank you Jeff! Nobody posted because maybe nobody bought one yet

Although I don't understand why the 970 encoding is slightly faster than the 960 (was the 970 an OC variant?), I do appreciate the effort and certainly is useful to draw a cost/benefit conclusion from them.

PS: Of course that conclusion will be strongly dependent of that cost represent to each one of us (as proportion of disposable income).

This message was edited 2 times. Last update was at Jun 15. 2016 05:23

JL_JL

Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline

Jun 15, 2016 21:45

Quote: It is interesting that your chart show that the GTX970 better than the 960 in h.264 encoding. The new GTX1070 does not surpass the 970 in every test. Maybe the GTX1070 will surpass both earlier cards in the hevc tests.

Not sure exactly what specifics you are referring to with the 1070 does not surpass the 970. If the ratio difference between 17.80 and 17.88 for the same input, that's a .4% difference, if the 1.61 to 1.63, that's a 1.2% difference. No testing on a WIN PC is any tighter than that, basically they are the same, no discernable difference.
I did look at HEVC as well, (1920x1080/30p 11Mbps and 4096x2160/30p 37Mbps MP4 profiles) and about the same overall conclusion, for basic simple encoding, the 1070 is about 1.3x faster with some variability.

Tony, a little data for the product you intend to use is always good. Hard to benchmark effectively but some basic tests give some guidance.

Quote: Thank you Jeff! Nobody posted because maybe nobody bought one yet

Although I don't understand why the 970 encoding is slightly faster than the 960 (was the 970 an OC variant?), I do appreciate the effort and certainly is useful to draw a cost/benefit conclusion from them.

PS: Of course that conclusion will be strongly dependent of that cost represent to each one of us (as proportion of disposable income).

The 960 and 970 are both SC versions, both EVGA.

I stumbled upon access to the 1070 so thought I'd post a few results, surely not advocating anyone go buy one. To each their own on what to spend and how. Heck some have 3GPU's, and you know what benefit that gives PD!

Jeff

Andrew - Wales, UK

Contributor Location: Wales, UK Joined: Jan 27, 2009 19:16 Messages: 545 Offline

Jun 22, 2016 09:36

Hi all,

I'm now lucky enough to have the MSI GTX1070. I haven't done lots of exhaustive tests but one I have done relates to a 1 hour project I recently completed. Nothing too fancy, just family footage with occasional text overlays and a few fades.

The footage is 1920x1080 50p 28mbps m2ts rendered in 'Produce' to 1920x1080 25p 16mbps MP4 for streaming at home.

Using my old Vapor-X Radeon R9 290X 8GB the project took 22 mins to render, using the MSI GTX 1070 took 14 mins. Both tests had hardware acceleration enabled.

Please note that in 'Preferences' -> 'Hardware Acceleration' I only have the bottom box ticked as I find this speeds up rendering on my system.

That's all I have done so far!!

Cheers,

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

tomasc

Senior Contributor Joined: Aug 25, 2011 12:33 Messages: 6464 Offline

Jun 23, 2016 01:05

Andrew – Thank you for sharing that h.264 GTX1070 Hardware encoding information. That is 4x real time where your Radeon R9 290x encoding is 2.7x real time.

You have supplied valuable information sought by two recent users on getting a possibly a better video card for their new not yet purchased pc’s. See this link: http://forum.cyberlink.com/forum/posts/list/48799.page .

Andrew - Wales, UK

Contributor Location: Wales, UK Joined: Jan 27, 2009 19:16 Messages: 545 Offline

Jun 23, 2016 03:11

No problem! This is worth a look:

http://www.videocardbenchmark.net/high_end_gpus.html

It appears from this and from other reviews I've read that the GTX1070 outperforms the GTX980Ti along with the TitanX. The GTX900 series prices are dropping. A look on Amazon UK suggests that, looking at the MSI Twin Frozr Gaming Models:

GTX1080 £629
GTX1070 £429
GTX980Ti £433 (down from £565 2 weeks ago)
GTX980 £344 (down from £429 2 weeks ago)

Cheers,

Andrew
Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

SoNic67

Senior Contributor Joined: Sep 27, 2014 14:14 Messages: 1308 Offline

Jun 25, 2016 10:04

Andrew, most of the reviews (98%) are oriented only to gaming performance. Means very little to actual PowerDirector performance.
The test that Jeff did is relevant to PD because it tests the encoding part of the video cards, that is not usually tested in reviews.
For that scope we have a maximum 30% improvement (4K->4K conversion), but it can be as little as 2%.
When you add effects, the differences become insignificant, almost to the point of statistical irrelevance.

Also, % sometimes are tricky, especially when the result might have a constant component (doesn't scale perfectly). If from 1 minute (1070) vs 1.66 minute (960) you expect that a bigger job of 10 minutes (on 1070) will mean 16.6 minutes (on 960)... you might find out that the result is different. It could be any value from that to 10min vs 10.6min.

PS: Jeff uses a relatively small video (boats), repeated. I suspect that the compression coefficients are calculated simpler (reused?) than if the file was complex, with more diverse scenes. I don't know how that would affect the scaling process.

This message was edited 1 time. Last update was at Jun 25. 2016 10:05

Andrew - Wales, UK

Contributor Location: Wales, UK Joined: Jan 27, 2009 19:16 Messages: 545 Offline

Jun 25, 2016 10:18

Which is why I included a PD test I'd done - the link I included for GPU benchmarks was for information only.

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

Notonotonurb

Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline

Aug 07, 2016 05:39

My comparaison between MSI GTX 960 4GB and Palit GTX 1070 8GB.

Codec: HEVC, max bit rate (128mb/s)

Resolution UHD

30fps

Clip length: 4 minutes (<a>https://www.youtube.com/watch?v=Me_vPONzy4I)</a>

This clip has no color grading.

NVidia driver: 368.81, Windows 10, i7 6700, 32GB ram

Results:

Without hardware acceleration: 116 mins

GTX 960: 12 mins 8 secs

GTX 1070: 12 mins 14 secs

Conclusion: in this case, the encoding speed is the same for both cards

Bruno Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1

Ray Mc

Member Joined: Dec 23, 2014 19:01 Messages: 61 Offline

Sep 02, 2016 22:36

I started delving into this just to get smooth editing in HD (1920 x 1080 mp4... Since I am shooting 4k and downsizing the money is worth it to go as high as possible. Is smooth editing and especialy hd possible with the following hardware ?????

Win 10 / I7 5930 intel cpu / 16 gb ram / ssd main and video drives. The 970 nividia does not give smooth playback and the opinion i have formed is this particular card has a bottleneck on the backside of its processing bus.

I will be installing a fresh load and retesting this hardware on a new system drive to try and rule out any other problems.

Thanks

JL_JL

Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline

Sep 02, 2016 23:23

Quote: I started delving into this just to get smooth editing in HD (1920 x 1080 mp4... Since I am shooting 4k and downsizing the money is worth it to go as high as possible. Is smooth editing and especialy hd possible with the following hardware ?????

Short answer, no.

I don't know the degree of non smooth you are talking about or if you've applied other video corrections, but if symptom is jump skips in multicam with 4 video streams and you want full HD preview in PD14, my answer is no. Not with a 960, 970 or a 1070, I have all of them and a editing platform that is 35+% faster than your referenced platform. I'll also extrapolate to the 1080 since I don't have one handy but from a editing point of view with PD14, it's a 1070 capable GPU. Simply use shadowfiles, manual proxy files or intermediate codec conversion to uncompress your mp4 are currently about the only option for fluid playback I'm aware of.

Please explain, "card has a bottleneck on the backside of its processing bus" and how it affects PD14 video editing? Never heard of it. Nvidia did settle a class action lawsuit for improperly stating video RAM speeds on the 970, makes no difference for PD14 as you will not come close to 3.5GB VRAM anyhow when the speed throttled.

Jeff

Ray Mc

Member Joined: Dec 23, 2014 19:01 Messages: 61 Offline

Sep 03, 2016 01:27

The hd files are down converted and video enhanced from 4k uhd to HD 1920 x 1080p 30fps h264 codec.

In Multicam the audio is lagging something awful. I can get by in music video's but interview show dosen't work so well. My projects are multi camera and I have dropped the quality preview quality down as needed to try and make it go

Doing straight edits on the timeline is smooth with multiple cuts from a multicam pass.

This is a explanation from the forum and I read it elsewhere: 970 is not good for video editing in 4K, because it has an older generation ASIC for video decoding/encoding. Differant from 950 / 960 and the newer 1000 series.

I did some conversions in Magic. one 4k uhd and one HD. Each played flawless. I decoded another set of 4 camera's in hd and it dosen't work. Even had a seperate audio track for the HD from the concerts sound board.

I am about to load a new system drive to make sure there isn't any other problems I am missing. My system drive is a old ssd and part of why i am doing a fresh load on a new drive.

Last but not least I guess I should be doing lower res prints. Then edit, rename and print..... off to the races...

Thanks for the time and input. Your numbers are in line with a few others posted.

SoNic67

Senior Contributor Joined: Sep 27, 2014 14:14 Messages: 1308 Offline

Sep 03, 2016 04:15

I think that "audio lagging" is other issue, not related to the video card.

Ray Mc

Member Joined: Dec 23, 2014 19:01 Messages: 61 Offline

Sep 03, 2016 04:28

just put a new ssd sys drive and loaded windows and Pd 14. first pass in multicam was in syc with Hd Magic files. No shadow files. HD quality preview.

JL_JL

Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline

Sep 03, 2016 07:38

Quote: just put a new ssd sys drive and loaded windows and Pd 14. first pass in multicam was in syc with Hd Magic files. No shadow files. HD quality preview.

As I had mentioned above, an intermediate codec playback will typically provide a very fluid playback. Magic files are an uncompressed AVI file, not mp4 as you previously mentioned.

Again, any real details on "card has a bottleneck on the backside of its processing bus" statement. I'm well aware of the NVENC ASIC IP block encoding/decoding differences which have nothing to do with bottleneck or bus.

Jeff

SoNic67

Senior Contributor Joined: Sep 27, 2014 14:14 Messages: 1308 Offline

Sep 03, 2016 08:01

I always had WMP-HC installed on my PC and it comes with the same filters like the K-lite ones (lav splitter).
I wonder if that's why I never had issues with PD like other people say.
I am not editing uncompressed avi and seems that Ray Mc didn't either go trough that step.

BTW, I think that the MagicYUVloseless in trial mode is not decompressing anything, it just pases the video as-is.

JL_JL

Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline

Sep 03, 2016 08:24

Quote:
BTW, I think that the MagicYUVloseless in trial mode is not decompressing anything, it just pases the video as-is.

SoNic67, something really amiss then, as it is just a intra frame codec. Yes it would simply have each frame compressed separately as defined in the original clip, hence loseless. So every frame in the output video will be an I frame, no P or B predictive frames to decompress during playback, so in the end it's similar to just a basic image sequence. So a typical IBBPBBPBBPBBP GOP structure becomes all I frames, hence the ~10x increase in file size as all the smaller P and B frames are now full content I frames.

So your processed video into a intermediate codec produced the same file size as original? That would be very strange.

Jeff

SoNic67

Senior Contributor Joined: Sep 27, 2014 14:14 Messages: 1308 Offline

Sep 03, 2016 08:38

I confessed that I didn't go trough that step since my system can edit fluidly the IBBP encoded files. I said that by just looking at the settings of that codec.

Well... not using multicam here.

This message was edited 1 time. Last update was at Sep 03. 2016 08:39