GPU Render Speed (RX 470 & QSV)

Shinra Bansho wrote on 1/23/2017, 8:42 PM

Dear Forum members,

I recently bought an RX 470 and, using the Sony Red Car Benchmark material, measured the rendering speed on Sony AVC and MainConcept encoders, of which results in the attachment. My CPU (Intel i7 6700K) has an integrated GPU (HD530) and its QSV was also tested this time: Case 1 is where RX 470 is designated for Timeline and then rendered on the Sony AVC as well as MainConcept AVC encoders with RX470 or QSV. Case 2 is where QSV is designated for Timeline and rendered on the Sony AVC and Mainconcept AVC encoders with RX470 or QSV in the same manner. For comparison, as Case 3, I did the same test with CPU only. Each test is done twice and the average of the two is shown.

In my environment (my setup below), a few observations:

1. As has been reiterated a number of times here in this forum, there was little gain in rendering speed on MainConcept AVC with a modern GPU like RX 470. (My HD6870, in fact, did much better than this even on a lower- profile PC.)
2. Sony AVC render gained only a little bit if QSV is set for Timeline and RX 470 in encode mode ("Render using GPU if available"), but not the other way around.
3. QSV does NOT work for render on 1080 resolution spitting an error ("Intel QuickSync video is not available"). Only 720 resolution works but it's blazingly fast (>100fps).

To my surprise, the use of RX 470 yielded the best with MainConcept MPEG2 for Blu-ray (not GPU acceleratable format, tho'). I use Sony AVC and MC MPEG2 formats most of the time hence am deciding to continue using RX470. I do hope that MAGIX would speed up the development to deal with modern GPUs, including the QSV, soon. Oh, before I forget, in terms of timeline preview performance on the Sony Red Car benchmark, 29.97fps was kept in all the Cases 1-3.

Mobo: Gigabyte GA-Z170N-Wifi
CPU: Intel i7 6700K 4GHz with a Corsair H60 Liquid Cooler
GPU: Sapphire Nitro RX 470 4GB
SSD: Sun Disk 480GB
HDD: Seagate 4TB 7200 rps
OS: Windows 10 Pro
GPU Driver: AMD Radeon 21.19.407.0 (12/23/2016)
VEGAS Pro 14 ("VP14): Build 211

Lastly, if RX 470 is used for timeline, VP14 fails to load some projects files with a set of 4K footage from Sony AX100, Panasonic GH4 as well as WX970M always ending up with an unmanaged exception error. Although I am not entirely certain at this stage whether or not the complexity of projects is the culprit, as is presented above, small project files such as the Sony benchmark seem to be loaded without an issue. To make the matter clearer, if QSV is used (or disable GPU completely) for timeline, VP14 loads the problematic 4K project files without any issue. I did not have this problem even on my prior lower-profile PC (i7 3770, 32GB, HD6870) and am assuming that the problem is related to the GPU driver. If anyone can give me advice on the fix, I would highly appreciate as I intend to use RX 470 for timeline. Thank you for you help in advance.

Comments

JN_ wrote on 1/24/2017, 6:51 PM

 

RE: AMD vs NVIDIA modern/recent gpus.

I posted these values previously, anyway, by way of comparison, it shows only a 5 second difference between Shinra's AMD and my NVIDIA based system for render times. Of course the cpu's etc are not the same. But Nick, as I previously pointed out, your jumping to the conclusion of a 2x difference [Quote ... "Nearly 2x improvement with RX480 over GTX1080 is more than I expected! It more or less settles the discussion, unless alternative or future Nvidia drivers improve OpenCL. " ] for the Amd system based on a single test by "Mike" in a previous post just doesn't hold up. Also, even this (5s) comparison is far from perfect because the playback fps is topped out by both systems. Until someone makes up a better red car test, or modifies the existing one to make it harder, it's a little indeterminate, but at least the red car test is readily available for testing, unlike Mikes test. I suggest you modify your 2x conclusion as it's currently misleading, until better and more tests are available based on modern/recent GPUS.

 

GPU on (off in brackets)

i7-4790K + GTX 1080 + 16gb ram.

The fps was at max 29.97, preview setting was at best-full.

XDCAM .. 0:31s (2:58s)

MC .. 2:15s (3:11s)

Last changed by JN_ on 1/24/2017, 6:58 PM, changed a total of 1 times.

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.

Nick Hope wrote on 1/25/2017, 12:11 AM
But Nick, as I previously pointed out, your jumping to the conclusion of a 2x difference for the Amd system based on a single test by "Mike" in a previous post just doesn't hold up.

Obviously a sample size of just one test is not ideal, and it wasn't very controlled, but that test is still the only one I've seen that directly compares GPU acceleration of video processing between a GTX1080 and RX480 in the same machine. Anyway, which 2 figures exactly (with a 5s difference) are you referring to?

JN_ wrote on 1/25/2017, 6:13 AM

The 2 figures I am referring to are ..

Shinra's ... 1st. column in the middle "case 2" table, 5th row down ...

time ... 2:10s ... RX 470 (Mainconcept avc 30fps)

Jn_'s MC ...

time 2:15s (3:11s) gtx 1080

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.

Nick Hope wrote on 1/25/2017, 8:51 AM

There are loads of variables here as your figures and Shinra's figures are from 2 completely different machines, and Shinra is testing an RX470, not an RX480. If I was making a purchase decision between the GTX1080 and the RX480 I would give more weight to the test that compared those actual 2 cards in the same machine by the same people. I still stand by what I wrote in that comment as I was simply doing the maths. Anyway I've added a link to your comment in this thread.

JN_ wrote on 1/25/2017, 9:31 AM

Well, not sure it's wise to stand by an assumption that there's a 2x difference between the Nvidia and Amd cards, gtx 1080 and rx480, based on 1 test.

I do agree with you that making a purchasing decision for a PC, and given that the main software of interest was Vegas, and given that the Amd is far cheaper, and Amd's reputation with openCL, it makes sense to go with Amd for now.

I posted here though not based on what to purchase, rather to try and nail down what the state of play currently is between the 2 manufacturers with modern cards, within Vegas.

I just don't buy the 2x part that you are standing over.

The two different tests, Mikes and Red Car are different, another variable, also introducing a question mark over whether perhaps some openCL tests may favour one manufacturer over the other?

You know, the test by Mike that you are putting so much weight on didn't give a lot of hard data, a bit of talk about 2x, etc. At least these 2 examples, despite some differences in machines, in this post gives the normal data you would expect to find in any comparison, i.e. machine basic specs., times for render with/without gpu, timeline playback fps, the vegas codec used for render etc.

The 2 PC'S, Shinra's and mine, are not a million miles apart, spec wise, yet the render times, with gpu ON, are very close, frankly, I'd put more weight on this comparison.

 

 

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.

Nick Hope wrote on 1/25/2017, 10:58 AM

You're putting words into my mouth here. Nowhere did I make a definitive statement that an RX480 is 2x faster than a GTX1080. I merely took Mike's results and calculated that ("almost 4")/("slightly over 2")=(nearly 2), and therefore observed that in his test there was nearly 2x improvement. That's all I intended, and actually all I'm standing by is my reading and maths ability. I accept it may have been a little rash for me to say that "it more or less settles the discussion" but at the time I felt that it had, given a lack of any other data to compare.

JN_ wrote on 1/25/2017, 11:39 AM

Well, I didn't feel I was putting words in your mouth, or had any intention to, the quote from you ..."Nearly 2x improvement ... etc" implied to me that that's what you believe. I think you may be splitting hairs here.

Anyway, for sure I don't want to get into an argument with you here Nick about this, and like i'm sure most here, I really appreciate the tremendous contributions you make, especially the FAQ.

I just want to get a little closer to finding out the real difference between the modern Amd and Nvidia capabilities within Vegas.

Based on the previous "Mike" test and these 2 tests, they are just not looking close to each other at all, to draw any definitive conclusion IMHO.

If more data/tests with modern cards become available that may change. John-dennis had a clip of his red car and it showed also better results for the Amd cards, but not as dramatic a difference as Mikes. Of course as always, all comparisons were different machines. Unfortunately Johns test video clip wouldn't tax the cards as well and all round as the Sony red car test. I think it would be fair to say that the tests in this current post and the john-dennis post were closer together than the "Mike" test. I strongly suspect that the 2x is just not reality, but further info. may clarify.

Given that the Sony test is readily available, maybe some more contributors might chip in (pun intended) with some Red Car tests, with modern cards.

 

Last changed by JN_ on 1/25/2017, 11:40 AM, changed a total of 1 times.

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.

OldSmoke wrote on 1/26/2017, 2:00 PM

On my system, the Preview Ram setting greatly influences render speed. Anything higher then 200MB will slow it down. In order to have accurate data, one must also take into account the preview ram setting.

Plenty of testing was done a long time ago, maybe that needs to be updated for newer GPUs but I would wait until the underlying code has been changed.

By the way. After installing the GTX580 alongside the Fury-X, I can now render MC AVC with CUDA again. SCS Benchmark project takes 25sec (1:25 CPU). XDCAM EX 1080 30p takes 18sec (1:29 CPU) with the Fury-X, Mpeg 2 is highly optimized in Vegas.

Edit: With the GTX580 set for timeline and CUDA, MC AVC renders in 33sec. Surprisingly, XDCAM EX 1080 30p renders in 23sec. It seems Nvidia has improved their driver in respect to OpenCL. I would have to physically disable the Fury-X to confirm that.

Last changed by OldSmoke on 1/27/2017, 5:33 PM, changed a total of 4 times.

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

System Spec.:
Motherboard: Intel DX79SR
Ram: G.Skill 8x4GB DDR3 2133 (running at 1600 and lower latency)
CPU: 3930K @ 4.3GHz (custom water cooling system)
GPU: 1x ASUS Fury-X
Hard drives: 4x 2GB WD Red in RAID 5 (with Hot Spare), 2x Crucial 256GB SSD in RAID 0 (mulitcam project drive), 1x Samsung 850 Pro 256GB SSD (System), 1x Crucial 64GB SSD (temp files and swap file), 1x 3.5" Hotswap Bay, 1x LG BluRay Burner
PSU: Corsair 1200W
Monitor: 2x Dell Ultrasharp U2713HM, 1x Sony HDTV 32" preview monitor

JN_ wrote on 1/26/2017, 2:49 PM

These were the results I got, more than 6 months ago, older drivers, with cuda and gtx 580.

As you say in your edit the newer drivers may give better results, I since replaced gtx 580 card with gtx 1080. I've been thinking of putting the gtx 580 back in with the 1080 but I heard that it (gtx 580) would only work with a slightly older driver, not the very latest, and i'd like to just keep the latest driver with the gtx 1080. Power supply is 750w so should be ok.

i7-4790K + GTX 580 + 16gb ram.

XDCam 0:33

MC 1:40

Waiting until underlying code is updated, i.e. a newer Vegas update I assume you mean?

If I understand you correctly then that is probably the best way to go. Doing more tests now with only partial support for gpus is just frustrating.

Last changed by JN_ on 1/26/2017, 2:53 PM, changed a total of 2 times.

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.

OldSmoke wrote on 1/26/2017, 3:32 PM

Keeping drivers updated only makes sense if you play games.

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

System Spec.:
Motherboard: Intel DX79SR
Ram: G.Skill 8x4GB DDR3 2133 (running at 1600 and lower latency)
CPU: 3930K @ 4.3GHz (custom water cooling system)
GPU: 1x ASUS Fury-X
Hard drives: 4x 2GB WD Red in RAID 5 (with Hot Spare), 2x Crucial 256GB SSD in RAID 0 (mulitcam project drive), 1x Samsung 850 Pro 256GB SSD (System), 1x Crucial 64GB SSD (temp files and swap file), 1x 3.5" Hotswap Bay, 1x LG BluRay Burner
PSU: Corsair 1200W
Monitor: 2x Dell Ultrasharp U2713HM, 1x Sony HDTV 32" preview monitor

JN_ wrote on 1/27/2017, 7:27 AM

Good point, thanks.

 

Desktop and Laptop basic specs ...

Both run Win 10 ...

Running latest ver. of Vegas Pro with latest updates.

VP13 B453 also.

Vegaseur and Pluraleyes installed on both ...

Quicktime 7.79.80.95

PC ...

i9 9900K, Intel Graphics 630. Nov 2018.

Mem. 32gb DDR4 Nov. '18.

Graphics card .. Nvidia Rtx 2080 Ti

Nvidia Graphics driver .. latest Studio driver.

Intel Graphics driver

 

Laptop ... (Acer Predator G9-793-77AC)

CPU .. i7-6700HQ Skylake-H

Memory ..16GB DDR4 

Graphics card .. Nvidia GTX 1070, latest Studio driver.