AMD VCE Rendering Questions

Comments

OldSmoke wrote on 1/21/2019, 9:56 AM

@gpasko You also have to realize that eventually the GPU will have to wait for the CPU to feed it the frame and that is why higher clocked CPUs are doing better in Vegas compared to lower clocked high core count CPUs. Also, the Fury X is VCE 3.0.

I understand that there's a balance between components. That's part of the fun w/ new technologies. What would your best guess be between running a VCE rendering comparison on the same PC between a R9-380 (4GB) vs a Fury X (4GB)? Would we get the same performance results? I suppose it might be slightly in favor to the Fury X because of its RAM speed. I'm just not sure how much a difference the GPU clock speed makes.

That would depend on the CPU, it would have to be fast enough to load up the higher spec card to see a difference, if it can’t, there will be as good as no difference. I guess on a 10core i9 at 5GHz you should see a difference between the two cards but you won’t on a 3GHz Dual core CPU. What CPU would just be sufficient is hard to say and may require a lot of testing.

Proud owner of Sony Vegas Pro 7, 8, 9, 10, 11, 12 & 13 and now Magix VP15&16.

System Spec.:
Motherboard: ASUS X299 Prime-A

Ram: G.Skill 4x8GB DDR4 2666 XMP

CPU: i7-9800x @ 4.6GHz (custom water cooling system)
GPU: 1x AMD Vega Pro Frontier Edition (water cooled)
Hard drives: System Samsung 970Pro NVME, AV-Projects 1TB (4x Intel P7600 512GB VROC), 4x 2.5" Hotswap bays, 1x 3.5" Hotswap Bay, 1x LG BluRay Burner

PSU: Corsair 1200W
Monitor: 2x Dell Ultrasharp U2713HM (2560x1440)

gpasko wrote on 1/23/2019, 8:47 PM

I've been following the forum for a while and I am generating some theories about the bottlenecks we ultimately will always have while using computers. l also have to say that I am really excited to be down to a single power-efficient GPU; down from an older GTX590 and an older AMD. I've recently tried two AMD GPU's to see if having one act only as the time-line OpenCL GPU (my older R9 270) and the better of the two cards RX480/580 as the VCE renderer -- I saw no performance difference.

Anyway, I have one crucial question for anyone that can shed some light on this for me. We all know there is a lot of "assembling" of the video & audio on the timeline, right. We all know there is a lot that OpenCL does for us using Vegas -- AMD cards being the OpenCL saving grace. I also realized that if I am watching the timeline playback during editing in real-time or better, including FX, that those videos should render, at least, realtime w/ VCE. My BIG question is this -- with the playback slider I can watch the timeline playback at 4.0x w/ sound and even 5.0x w/o sound and all the FX and at BEST FULL settings -- what is the bottleneck? If the timeline is truly running with FULL BEST at 4x 30fps we should be capable to render at a potential of 120fps. Is the bottleneck the part of our system that compresses the video rendering, currently looking at you VCE/GPU? :)

...I hope this made sense ;)

BTW - I have a new video card that I've ordered to test this theory. It's an R9 Nano for my small PC that I use for my racing gaming rig. I'm hoping it arrives before this weekend to test out my bottleneck theory on my Vegas Workstation. From what I've learned, as far as speed goes, it would not be my first choice for VCE rendering. It is a good choice and it should be equal to a Fury X (as far as VCE). It is a VCE 3.1 chipset and if my researched data is correct it should perform about 20% faster than my RX480/RX580.

Keeping my fingers crossed!

 

Kinvermark wrote on 1/23/2019, 8:56 PM

Yes, clearly rendering means having to playback the timeline and encode a new file, so it should take longer. there may also be "better quality = slower" algorithms involved during encode.

bob-h wrote on 1/24/2019, 12:59 AM

My BIG question is this -- with the playback slider I can watch the timeline playback at 4.0x w/ sound and even 5.0x w/o sound and all the FX and at BEST FULL settings -- what is the bottleneck? If the timeline is truly running with FULL BEST at 4x 30fps we should be capable to render at a potential of 120fps. Is the bottleneck the part of our system that compresses the video rendering, currently looking at you VCE/GPU? :)

I think I've read that 480/580 generateS around 60fps at 1080p via VCE (h.264) using Quality preset. You could try the nightly build of Handbrake & do a transcode using VCE & confirm vegas isn't any slower. That would probably indicate your render speed in limited by VCE.

With NVENC I get 57fps in Vegas & 110fps in Handbrake(constrained by 100% CPU), but with Vegas I notice it pausing approx once a second & I don't think that's supposed to be happening.

gpasko wrote on 1/24/2019, 10:00 AM

My BIG question is this -- with the playback slider I can watch the timeline playback at 4.0x w/ sound and even 5.0x w/o sound and all the FX and at BEST FULL settings -- what is the bottleneck? If the timeline is truly running with FULL BEST at 4x 30fps we should be capable to render at a potential of 120fps. Is the bottleneck the part of our system that compresses the video rendering, currently looking at you VCE/GPU? :)

I think I've read that 480/580 generateS around 60fps at 1080p via VCE (h.264) using Quality preset. You could try the nightly build of Handbrake & do a transcode using VCE & confirm vegas isn't any slower. That would probably indicate your render speed in limited by VCE.

With NVENC I get 57fps in Vegas & 110fps in Handbrake(constrained by 100% CPU), but with Vegas I notice it pausing approx once a second & I don't think that's supposed to be happening.

Testing using HB is a great suggestion! Yes, my RX480/580 does the RED CAR 1080p video render at an average 58 FPS; ranging for 40fps to 80fps directly in Vegas. I believe I get the same speed when I turn off all the FX, so that isn't the limiter. I'll try Vegas+HB and report back.👊

With my AMD cards, I don't see any pausing. It may be an Nvidia symptom with OpenCL. I seem to remember seeing some pausing when rendering when using an old GTX590 card as the timeline playback GPU & the legacy rendering GPU. The driver version might have helped that too.

IAM4UK wrote on 1/29/2019, 12:59 PM

Having just acquired a Radeon Vega Frontier Edition, I can say: "WOW! That really changes the preview experience and the rendering times, versus the previous card (R390)."

Nice to be able to use VCE for HEVC 4K renders, too.

Early experiments show that certain GPU-enabled plugins slow down VCE renders more than others (and curiously, "engage" a lower percentage of the GPU, per Task Manager performance monitoring). Ignite plug-ins are particularly slower than Vegas ones, for example.

gpasko wrote on 2/4/2019, 8:46 PM

Sorry, it has taken me this long to respond. I’ve been doing some testing on 5 different AMD GPU’s w/ Vegas Pro 15. This should show some comparisons between VCE 1.0, 2.0, 3.0, and 3.4 versions. I’m using my Dell T5600 workstation, dual E5-2690 CPU’s, 32 GB ECC RAM, Windows 7 Pro. All testing has Hyperthreading OFF (16 real cores), all memory banks filled 4G x 8 DDR3-1333Mhz. The memory in this configuration runs at quad speed. My research shows that other’s memory testing doesn’t show much if at all performance increase w/ faster memory or fewer banks filled. I can upgrade up to DDR3-1666Mhz, but I don’t think it will show any performance gains on MY system. Vegas Pro 15 (Build 416) w/ 200MB dynamic RAM used (default), and 32 rendering threads (default) set. AMD driver version 18.12.2.

 

The test results are for the Sony Red Car test in 1080P using the preset for Magix AVC 1080P Internet 29.97 fps (AMD VCE):

 

R9-270 (VCE 1.0) = 34 sec

R9-280X (VCE 1.0) = 32 sec

RX480/580 (VCE 3.4) = 31 sec

R9-380X (VCE 3.0) = 30 sec

R9 Nano (VCE 3.1) = 29 sec

 

My fastest rendering results had been using the “legacy GPU’s” on Vegas 11 Pro on the same PC system years ago:

 

R9-280X (OpenCL timeline) + GTX580 = 27 sec

 

This lead me to think of another way to test for bottlenecks. I started a new Vegas project and created one track and added a solid black color FX and stretched the event to 1 minute. I ended up rendering that on the 5 GPU’s and here are their rendering times:

 

R9-270 = 26 sec

R9-280X = 26 sec

RX480/580 = 19 sec

R9-380X = 13 sec

R9 Nano = 14 sec

 

Of course, this is really simplifying all that Vegas is doing, but this would be pushing the VCE GPU as fast as it can go and leaves out any video conversion, transitions, filters, color corrections, GPU FX, etc. I am starting to think we have a bottleneck most people don’t think of. We have a software bottleneck. When rendering the Red Car w/ the RX480/580 my CPU is teetering between 40-60%. The GPU usage is hard to tell because Windows 7 doesn’t show the encoding usage of the GPU as Windows 10 does.

 

If I’m overlooking something please let me know. Or if you want me to test something else let me know as well. I still have some of the GPU’s, but will be selling off the ones I am not using. I’ve settled on the RX480/580 (VCE & HEVC) for Vegas Pro and R9 Nano for my gaming PC. I may eventually get a Vega card for testing, but don’t think the performance gains will be much w/ MY system. I am still impressed by the rendering performance combination of Vegas w AMD’s VCE processing. I am still very happy with the performance of my PC. We spend a lot more time editing in Vegas compared to rendering. This rendering performance increase is always welcome and if more can be squeezed out of it then great. I think the Magix team developing the software is doing a great job! Keep it going as we all love the flexibility and speed we can edit! Rendering at a reasonable speed for most is a bonus now!

 

Btw - My Vegas2HandBrake performance was relatively slow. I was only getting around 40 fps using Red Car and HandBrake VCE. I get better performance using Vegas VCE rendering directly or CPU rendering. Might be some issue on MY system w/ Vegas -> Frame Server -> HandBrake setup.


 

james-ollick wrote on 2/6/2019, 9:25 AM

gpasko thank you for your tests. I purchased the RX580 and have been very happy with its performance.

 

Last changed by james-ollick on 2/6/2019, 9:26 AM, changed a total of 1 times.

Home built PC - Corsair case, ASUS ROG Maximus XI Code motherboard, i9 9900k, 32 gigs Corsair Vengeance RGB 32GB DDR4 DRAM 3200MHz,  Sapphire Nitro+ Radeon RX 580 graphics card, Corsair 1000 watt power supply.

DesertSweeper wrote on 5/24/2019, 1:57 AM

Inqould tryband leave hyperthreading ON and reduce the amount of threads inside Vegas under preferences.

Surprising results IMO w/ little variance between allowable rendering threads:

Magix VCE still using R9 270

Hyper Threading ON

39 sec (8) threads 200MB RAM Preview (on all)

 

Hyper Threading OFF

34 sec (8) threads 200MB RAM Preview (on all)

 

 

This is most likely as a result of recent software/firmware updates that mitigate Meltdown/Spectre/Branchscope/PortSmash