AMD VCE and NVidia NVENC Time Comparison (Render time comparison)

Reyfox wrote on 7/11/2018, 5:03 AM

The article is about comparing workstation GPU performance with Redshift, Blender & MAGIX Vegas 15.

The article compares performance with both workstation cards from both companies along with tossing in 1080ti, Titan Xp, Vega 64 and RX580. The test bed computer has Intel Core i9-7980XE (18-core; 2.6GHz) and 64GB RAM.

The Vegas Pro test is at the end of the article. Tested is the HEVC encode speed using Magix encoders for the graphics cards. No, it's not an end all be all test, and it would be nice if the clips had more FX added, but, it's a starting point.

Read about it here: https://techgage.com/article/performance-testing-redshift-blender-magix-vegas/

Comments

OldSmoke wrote on 7/11/2018, 6:34 AM

It seems that what has been established a long time ago is still true. Vegas is still using OpenCL for FX processing and AMD has implemented it much better then Nvidia.

Kinvermark wrote on 7/11/2018, 12:24 PM

Interesting. Thanks for posting. We need more of this kind of info to better understand hardware setups for the new Vegas engine. I would like a GPU upgrade, but with crypto mining market interference and not knowing what Vegas prefers, I have been waiting. Maybe soon now.

Reyfox wrote on 7/11/2018, 3:08 PM

I too would like to buy another graphics card (AMD), but the prices are way out of line. I can wait. Vegas with AMD VCE encoding is really fast. I am ok with it right now.

Reyfox wrote on 7/11/2018, 3:47 PM

The author of the article is looking for suggestions on what fx combinations should be tested. I am a novice with Vegas and defer to you "senior" Vegas editors. So please reply to the above link with your suggestions!

I have nothing to do with the author of the article for further clarification. Just passing on info.

fr0sty wrote on 7/11/2018, 6:28 PM

AMD's ProRender is also natively supported in Cinema 4D now, and while it works on any GPU, it is OpenCL based, so...

Their cards are beginning to look a bit more attractive for workstation use... if I can just see some reliability and solid drivers, I might give them another shot one day.

Kinvermark wrote on 7/11/2018, 6:40 PM

Looks like the author has found this forum...(maybe he can confirm?)

 

That being the case, I would suggest the following fx for testing:

Filmconvert (third party vegas plug in)

Neat Video ( ")

Prodad Mercalli (")

Vegas LUT (one or multiple instances)

The old Film Effects fx (I think this one may be single threaded / non gpu)

Also relevant would be tests using 32 bit mode and ACES color managed workflow.

Reyfox wrote on 7/12/2018, 7:22 AM

@fr0sty, I have been using AMD graphics cards since it's original ATi Radeon introduction. Currently, still using the RX480 8GB MSI card. The only time I had a driver issue was just recently. I no longer was able to see AMD VCE in the Render As listings with the newest driver. I went back to the previous one (I keep them until I am sure the latest one works fully). I can not speak on the "pro" workstation cards, but looking at the performance of the RX580, impressive in Vegas.

BTW... here is the video to accompany the weblink.

 

Deathspawner wrote on 7/12/2018, 6:38 PM

@Reyfox Thanks a lot for sharing the content; I'm glad it's proven useful :D

@Kinvermark Those are some great suggestions, so thanks a ton! I'm starting to make note of all this for when I can dive into testing again. If I gather enough useful benchmark information, I may revolve an entire content piece around Vegas, and if so, I'll be sure to take a look at Filmconvert since there's a trial available. 

As an aside, I also plan to test at some point the same projects on an Intel Coffee Lake system, since MAGIX rolled out updates in ver 15 for Intel Quick Sync Video. If the performance boosts are anything like they are with Premiere Pro, there could be some surprises. 

GJeffrey wrote on 7/12/2018, 11:29 PM

To compare apple to apple, I would suggest not to use any fx.

VCE and NVenc is only used during encoding with Magix avc and hevc codec.

Opencl is used for fx processing. AMD is obviously better at opencl than nVidia.

Either you compare VCE and Nvenc or opencl performance but not both.

fifonik wrote on 7/12/2018, 11:52 PM

To compare apple to apple, I would suggest not to use any fx.

I cannot agree more.

It would be nice to see:

- To check how good source filter on the system. Preview frame rates for couple of typical sources (what sources? this alone is already complicated). Plus generated media. Probably this would be almost the same for Nvidia/AMD, however this information is important for understanding the hardware's I/O and analyzing further results.

- VCE/NVEnc. Encoding without any filters.

- In-build filters support by GPUs. Preview frame rates with typical sources and a few in-build filters applied. Something simple and popular, like levels/color correction/sharpen/crop. NeatVideo should not be included as it has it own integrated benchmark.

Or simply use VP11 Benchmark :)

Kinvermark wrote on 7/13/2018, 10:25 AM

Yes. My bias is towards playback performance for editing rather than rendering (I can wait; within reason).

Most "tests" of this sort focus on render speed because it is easier to quantify than "is-playback-smooth" under varying conditions. There was a Davinci Resolve test I saw recently where the editor quantified by adding cc nodes until the playback frame rate started to drop. IMO, that method is OK, but a limited simulation of video editing.

I will take whatever information comes... and add a "grain of salt."

Reyfox wrote on 7/13/2018, 10:39 AM

Playback performance is a big one for me also. Still, AMD VCE exporting is a huge improvement in speed over the CPU rendering. It's a start, now on to timeline playback performance.

AVsupport wrote on 7/13/2018, 5:33 PM

+1 for playback performance. That's what makes most impact for me. Final renders can vary, and for me a little wait is expected / acceptable

BTW, I think it's VP's old OpenCL standard compatibility code base that's the biggest holdup. Technology has moved on, VP doesn't use the better accelerations available.

Also, have found [correction: Ignite Vibrance] to use GPU quite well, surprise! (tested with 1060/6 at CPU55%/GPU~30% resulting in usable playback..can anyone confirm? souces 2K mp4 from Mavic on 1080/25 timeline)

bob-h wrote on 7/13/2018, 6:49 PM

 

Also, have found [correction: Ignite Vibrance] to use GPU quite well, surprise! (tested with 1060/6 at CPU55%/GPU~30% resulting in usable playback..can anyone confirm? souces 2K mp4 from Mavic on 1080/25 timeline)

I just tested it with 4k video. It doesn't drop any frames when played on best(full). As you say that filtert makes great use of GPU.

walter-i. wrote on 7/14/2018, 4:00 PM

+1 for playback performance.

Thanks in advance
Walter

GJeffrey wrote on 7/14/2018, 4:48 PM

Looks like most of the people are interested in timeline preview performance.

I suggest to the OP  @Reyfox to modify the title of this post. Its misleading.

Reyfox wrote on 7/14/2018, 6:15 PM

@GJeffrey

How is the post misleading? The topic might have gotten derailed a bit, but the post certainly is not misleading.

Wolfgang S. wrote on 7/16/2018, 7:23 AM

The article compares render time only. But there is no appropriate measure for the playback performance of the footage in the timeline in that article. Both aspects are important, but a lot of users may tend to give the playback performance a higher priority, simply since that is a requirement to edit the footage. I have clearified the title a little bit.

Reyfox wrote on 7/16/2018, 7:26 AM

@Wolfgang S., thanks for the title clarification.