Blackmagic Forum

John Paines wrote:Am not sure how you tested NR, since Lite doesn't offer it(?)....

DaVinci Resolve (previously called Lite) offers Studio only features with a watermark. As much as possible.

Studio only features with a watermark

silly question - would the added watermark alter the render speed?

Leslie Wand wrote:Studio only features with a watermark

silly question - would the added watermark alter the render speed?

It's a good question!

The difference is very minor.

Rohit Gupta wrote:DaVinci Resolve (previously called Lite) offers Studio only features with a watermark. As much as possible.

Thank you for this change! It helps to be able to test stuff for example when on the go without the dongle. Like when building DCTLs or so.

I'm still a little confused on how resolve uses multiple GPU's? If I had three in the system would it use all three including the GUI gpu if I had it checked in the options? If I did would it be worse for performance? It doesn't make sense to have one wasted on just running the UI?

If you have the Studio version then you can have as many GPU's in your system as the operating system will support, and as long as they all match you can tell it to use the GUI GPU for rendering as well. If the GUI GPU is less powerful and has less RAM then you wouldn't want to use it for rendering as it will slow down and limit the RAM used on the other cards

Adam Simmons wrote:If you have the Studio version then you can have as many GPU's in your system as the operating system will support, and as long as they all match you can tell it to use the GUI GPU for rendering as well. If the GUI GPU is less powerful and has less RAM then you wouldn't want to use it for rendering as it will slow down and limit the RAM used on the other cards

Excellent thanks! We have the studio version that we got with the Ursa 4.6k. We've been going through the ripple training videos and learning, but not really come across any that mention multiple GPU usage. I'm tempted to buy 3xRX480 and place them in my HPZ820 system. I know the AMD cards also have 10-bit output too, but I should be able to run a 10-bit output from one of the cards or would I have to buy another BMD Intensity Pro 4k card?

Definitely get a BM I/O solution, like the Infinity. Resolve is relying on that monitoring path.

John Paines wrote:Am not sure how you tested NR, since Lite doesn't offer it(?)

Anyway, at what node does the blur test lose real-time? Is it 9 nodes -- or more? People report this inconsistently, so it's hard to compare. If it drops frames on node 10, then I'd be concerned.... But of course there's also your CPU, which is on the slow side, by current standards.

It did run the NR. I do see the watermark on it. At first I though it was the part of the test till someone pointed out here its the watermark.

Its does NOT drop frame on the 10. Its good even on the 11th node
Its drops frame on the 12th node(it shows 22-24).

I noticed that 4K footage from Panasonic G7, during editing when I move back and forth on the timeline its slows down(). The very same issue stated in the video. But when I used optimized media it was good. So would a card like 480 solve it for me? Or do I continue to use optimized media. Should I stick with GTX 1060 or get the 480 RX? Sorry I did not get an answer earlier.

I didn't watch the video, but the GPU won't help with decoding highly compressed 4K video, such as the Panasonic G7 footage -- timeline responsiveness isn't likely to change. For whatever reason, Resolve doesn't handle long GOP formats as well as Premiere or FCPX. Transcoding or optimized media is the best approach.

As for the candle test, results of other users above suggests you'd do better with the 480, but there's no way to know for sure, without actually testing it. Also, you'd want to test other GPU-intensive effects, like 3D keys and raw decoding.

I've benched my overclocked Gigabyte Gaming G1 1060 card:

Blur:
09 Nodes: RX480:24fps TitanX:24fps 1060: 24fps
18 Nodes: RX480:20fps TitanX:16fps 1060: 14fps
30 Nodes: RX480:13fps TitanX:11fps 1060: 9fps
66 Nodes: RX480:6fps TitanX:5-6fps 1060: 4fps

TNR:
1: RX480:24fps TitanX:24fps 1060: 24fps
2: RX480:18fps TitanX:17-20fps 1060: 17fps
4: RX480:9fps TitanX:11fps 1060: 8fps
6: RX480:7fps TitanX:8fps 1060: 6fps

I also did some tests with decoding RED footage, the files that come with Patrick's Resolve 11 training, and the 1060 is about 15% slower on RED decode than the RX480 in the same system, the 1060 is also slower with Neat Video, however the heavier tests we did with Neat Video failed on the 1060 as it ran out of RAM, it has 6GB vs the RX480s 8GB and the Titan X with 12GB.

The 1060 I have is also huge, it was a real pain to fit it inside the case, I believe there is a short 1060 model available, it isn't overclocked, but would be easier to fit.

In short, the RX480 is faster with every test we throw at it in Resolve than last years Titan X, and much faster than the 1060. Looking at the specs for the new Titan X, it will be a bit quicker than the 480 in resolve, but a pair of 480 cards should be a lot faster than the new Titan X, at considerably less than half the price.

davidanderson wrote:I've benched my overclocked Gigabyte Gaming G1 1060 card:

Blur:
09 Nodes: RX480:24fps TitanX:24fps 1060: 24fps
18 Nodes: RX480:20fps TitanX:16fps 1060: 14fps
30 Nodes: RX480:13fps TitanX:11fps 1060: 9fps
66 Nodes: RX480:6fps TitanX:5-6fps 1060: 4fps

TNR:
1: RX480:24fps TitanX:24fps 1060: 24fps
2: RX480:18fps TitanX:17-20fps 1060: 17fps
4: RX480:9fps TitanX:11fps 1060: 8fps
6: RX480:7fps TitanX:8fps 1060: 6fps

I also did some tests with decoding RED footage, the files that come with Patrick's Resolve 11 training, and the 1060 is about 15% slower on RED decode than the RX480 in the same system, the 1060 is also slower with Neat Video, however the heavier tests we did with Neat Video failed on the 1060 as it ran out of RAM, it has 6GB vs the RX480s 8GB and the Titan X with 12GB.

The 1060 I have is also huge, it was a real pain to fit it inside the case, I believe there is a short 1060 model available, it isn't overclocked, but would be easier to fit.

In short, the RX480 is faster with every test we throw at it in Resolve than last years Titan X, and much faster than the 1060. Looking at the specs for the new Titan X, it will be a bit quicker than the 480 in resolve, but a pair of 480 cards should be a lot faster than the new Titan X, at considerably less than half the price.

Your test matches mine. I have the Zotac Mini(Got it for $229). Its not overclocked and the installation went smooth and its small. For 15% loss in performance, I guess I will stick with 1060 since if I sell it off and get the cheapest 480 I will be losing money. I wish I had got this review this Monday, I could have cancelled my 1060

The performance varies between 15% to 45% faster than the 1060, depending on the plugins, source files etc.

I agree, if you already have a 1060, the performance gains might not be enough for it to be worth your hassle, although I would still consider it as adding a second card later becomes an option and then the performance difference is more noticable between the pair of 1060s and the pair of 480s.

This is the last benchmarking we will do, we are a production company, not a PC test site and this took up a surprising amount of time, but we did want to share our experience as it is a great boon to the smaller shops to be able to get what just 12 months ago was the fastest performance available, for well under $300.

The 480 is a really amazing value proposition, and even when stuck in an 8x slot, the performance dropis only a few percent, so it is also working great in our old Dell Precision workstations that were about to be put out to pasture.
So there is little reason for even the most budget conscious colourist to go without cutting edge performance with Resolve now.

I'm surprised to hear you say it took up a long time for the testing. I found it only took around 15-20 minutes per card and that includes removing the old one and installing the new one

davidanderson wrote:The performance varies between 15% to 45% faster than the 1060, depending on the plugins, source files etc.

I agree, if you already have a 1060, the performance gains might not be enough for it to be worth your hassle, although I would still consider it as adding a second card later becomes an option and then the performance difference is more noticable between the pair of 1060s and the pair of 480s.

This is the last benchmarking we will do, we are a production company, not a PC test site and this took up a surprising amount of time, but we did want to share our experience as it is a great boon to the smaller shops to be able to get what just 12 months ago was the fastest performance available, for well under $300.

The 480 is a really amazing value proposition, and even when stuck in an 8x slot, the performance dropis only a few percent, so it is also working great in our old Dell Precision workstations that were about to be put out to pasture.
So there is little reason for even the most budget conscious colourist to go without cutting edge performance with Resolve now.

Thank you very much for your efforts. I think it will indeed help a lot of people to make a decision.

another thank you too. your time and efforts are greatly appreciated indeed.

i think with both your report and what else i've read that the 480 will suffice for now especially since i'm working hd mostly and, when circumstances dictate, i'll revisit the whole question again with 4k and the attendant cards available at that time in the future.

Tim Mark wrote:I test the GTX 1060

Resolve 12.5 Lite, Z87 Motherboard, 4770K.

Blur:
09 Nodes: 24fps
18 Nodes: 14fps
30 Nodes: 9fps
66 Nodes: 4fps

TNR:
1: 24fps
2: 17fps
4: 8fps
6: 5fps

Those scores are not very good considering the 1060. This was my result from last year using a GTX 970 3.5GB:

z87 'Haswell' 3.6Ghz 4 Core i7 PC, 16GB RAM, 2 x 256GB SSDs Striped RAID 0 under Win 7 64bit OS
Resolve 'lite' 11.2 with 1 x Maxwell GTX 970 3.5GB:

11 Blur @ 24fps
13 Blur @ 19fps
15 Blur @ 17fps
18 Blur @ 14fps
30 Blur @ 09fps

Compute wise the 1060 is about the same performance as the 970 in nearly every application, so that sounds about right.
The main advantage would be the 6GB vs the 970, but speed wise it would be about the same.

Adam Simmons wrote:I'm surprised to hear you say it took up a long time for the testing. I found it only took around 15-20 minutes per card and that includes removing the old one and installing the new one

Our machines are wired into racks and normally stay untouched. I had to pull the machine, and take it up to a workbench. These machines are not internet connected, so we had to pull the files down via laptop, check them and then transfer them to the machine.

The 1060 card was so long that it obstructed some HDDs that had to then be moved, requiring re-routing the cables, and remounting the drives.

The system previously had the Titan X, so drivers had to be completely removed, initially Resolve wouldn't switch to OpenCL from CUDA, so this took a bit of messing around when the RX480 went in. Then there were initially concerns re power, so we had to contact AMD and sort out that issue (they provided patched drivers before they were released to the public, which was great service), then we had to find the footage for the Standard Candle test, download it, run those tests and record the results. Also load some RED footage and do some plugin testing to see if the performance still held across different usage patterns. We also then locked it down to 8x instead of 16x to check the performance.

Then the RX480 had to come out and the 1060 go in, which again meant driver removal, downloading the latest drivers, the aformentioned trouble getting the 1060 to fit, installing the new drivers and then running all the tests again, then coming back here to post results, and trying to fit this all in between paying jobs.

Also, this machine has a RAID card which makes for a very long boot process with each restart, of which there were *many*.

Then the RX480 was put back in the machine, Nvidia drivers removed, AMD drivers installed and the machine placed back in the rack, scanned and put back to work.


So all up, yeah, it took longer than expected.

davidanderson wrote:So all up, yeah, it took longer than expected.

You're a dedicated man! (Isn't that a quote from Ridley Scott's 'Bladerunner'?)

I have Resolve 12.5 Studio and a Gigabyte R9 390X 8gb gpu card. Can I add a second amd rx480 8gb of any vendor brand and how will it be best set up? Thanks in advance.

I have Resolve 12.5 Studio and a Gigabyte R9 390X 8gb gpu card. Can I add a second amd rx480 8gb of any vendor brand and how will it be best set up? Thanks in advance.

It's better to use identical compute gpu cards. Non-matching compute cards are limited by the speed and ram size of your worst gpu.

At least your 390x has 8gb of ram but unfortunately an additional rx480 will run at 390x speeds.

A better setup would use TWO rx480 as compute cards and the 390x solely as GUI. Assuming your motherboard can handle 3 GPU's.

Anyone tried the new Titan X yet?

Neil Brassington wrote:Anyone tried the new Titan X yet?

See this:
viewtopic.php?f=21&t=50957

We haven't tested the RX480 with dual cards, but the triple RX480 card benchmarks are:

Blur:
09 Nodes: 24fps
18 Nodes: 24fps
30 Nodes: 24fps
66 Nodes: 17fps

TNR:
1: 24fps
2: 24fps
4: 24fps
6: 19fps

The cards we used were USD249 each, they come with a mild overclock as 'standard' so it was USD750 for the triple card setup.

Was this still on the X99 board or did you have to go up to a dual Xeon to get the speed?

This is on a Gigabyte X99 motherboard, with 64GB DDR4 RAM.

That's really good performance/price ratio. Anything over two TNR @24fps is excellent so achieving four on an X99 i7 mobo is probably all the performance most people working with 4K and HD source will ever need.

David,

Have you run into any issues with current AMD driver? I have 4 installed and trying troubleshoot some odd behavior. For starters if I select Use Display GPU For Compute only one is recognized. I tested this with Dual cards, 3 cards, & 4 cards. If I uncheck Use Display GPU For Compute all four cards show however one is GUI only.

I'm also seeing choppy performance in the GUI while mostly maintaining realtime playback on reference monitor.

Lastly I tried to render a 60sec commercial (R3d source material) at source resolution to DNxHR 444. I got Your GPU memory is full errors. About a dozen clips have temporal noise reduction.

I have Radeon Software version - 16.8.2

Here is my standard candle benchmark

24
24
24
16
24
24
15
10


Patrick Woodard

I assume the cards are in the Xeon system in your signature. With Xeon boards sometimes the slots that the cards in can make a difference as different slots go through different CPU's and therefore can react differently.

I also assume you are using the full Studio version and not the free version.

Did you buy the 8GB or 4GB RX 480's?

davidanderson wrote:We haven't tested the RX480 with dual cards, but the triple RX480 card benchmarks are:

Blur:
09 Nodes: 24fps
18 Nodes: 24fps
30 Nodes: 24fps
66 Nodes: 17fps

TNR:
1: 24fps
2: 24fps
4: 24fps
6: 19fps

The cards we used were USD249 each, they come with a mild overclock as 'standard' so it was USD750 for the triple card setup.

Hey did you use a GUI card? or was on dedicated to GUI and 2 for processing ?

I would suspect they used 1 of the card as GUI and Render and the other 2 just as render