Advantages of Dual GPU for ARRIRAW transcode

Hi,

A simple question I hope.

I am just setting up a rig with top end consumer parts, so will be a 6950x and 1080Ti (cant really see the value add of the Titan).

I have seen some benchmarks on the candle test with Dual GPU vs single and can see advantages under demanding workloads. However I am never going to be needing anything like this for real time workload.

My only real concern is if I will see any advantage during transcoding - will a dual GPU set up benefit me in terms of FPS transcode speeds in any significant way.

If anyone has this setup I would love to know some actual numbers for ARRIRAW to DNX36 with and without second GPU online.

Also - is there any other software commonly used for dailies which utilises a second GPU?

For such a task you may be better using ARRI own free ArriRAW Converter software, which can use 2 GPUs quite well as far as I remember. It doesn't do DNxHD, but it does all ProRes modes (also on PC as far as I understand).
Exported MOVs will have all metadata passed from RAW files.
There is GUI and cmd line version for all OSess, it's very flexible. For proxies you can use lower debayer setting which should give you very good speeds.

On a 1080p timeline, I rendered an Arri raw sequence 3424x2202(33 seconds long, 24fps) at DNX 36 8 bit in 18 seconds. DeBayer set to full res Resolve(not Arri). Done with a gtx 1080 ti on a lowly i7 4770k. Don't have a second Ti card.

Sounds okish. Your CPU is probably fast enough. Does it go above 50% usage? Encoding DNxHD 36 is not that demanding at all.

Hmm... not sure as I used ArriRaw Converter long time ago on Mac. I think it may be faster and has cmd line, so you can script things if you deal with a lot of footage.
These are proxies, so debayer quality is not that important at all. You rather want things done quickly.

Arri debayer is much slower than Resolve debayer. Inside Resolve you get real time playback on full Resolve debayer, while using Arri debayer gets 5 fps. I tried the Arri convertor and it was getting over 2 minutes(The 33 second clip) on all the quicktime prores codecs.

Dan - thanks for those figures, they are not bad at all!

I actually pulled the trigger on a 1080Ti today so will get to check out what kinda speeds it runs at with my current CPU which is a 5820k... still hovering over the purchase button of the 6950x.

Andrew - Cheers, I'll have to check out ArriRAW converter if it is indeed available for PC and see how it compares to Resolve, although it is more than likely DNX I will be delivering so Resolve will skip that extra step!

As you said, I dont need full debayer quality, purely focussed on media tunaround on set, so transcode speed is essential.

Well, I just tried and I get (2.8K RAW sample):

- 6.5fps for full quality (SW5 debayer) 2.8K ProRes Proxy export
- 5fps in Resolve full quality (Resolve debayer) ProRes Proxy export

- 14.3fps Arri in Proxy mode full resolution, export to HD
- 11fps in Resolve half resolution debayer, export to HD

When it comes to preview (playback):
- 9fps full quality/resolution in Arri converter
- 4fps full quality/resolution in Resolve
- 10fps with SW5 debayer at full resolution in Scratch

- 25fps for half size proxy setting preview in Arri
- 13fps half resolution debayer in Resolve
- 87fps with Fast debayer in Scratch at full resolution (looks absolutely enough as for preview or proxy)
Maybe data was cached after 1st playback. Need longer sample to get real max speed.

Arri Converter is faster in my case and my machine is just Macbook Pro dual i5 with Intel 6100.
This maybe not representative, but I think in case of proper GPU difference maybe even higher (remember people being so happy with speed on dual D700)

If you have 5fps in Arri Converter then something is not right with your settings/setup.
For fastest turn around Scratch will probably blow anything else by miles- don't have license to check export speed.

Blimey, I was going to test out scratch but thats astonishing, will definitely be testing it out now Iv'e seen that, cheers Andrew.

87fps could be due to some caching- small sample. Need to verify, but I had no issue to play at 24fps first time.
Scratch own Fast debayer is really fast.

Ok, I restarted machine, loaded sequence again, set fps to free run and 1st time hit play. Got 70fps.
At 2nd play got 80fps. Looks like after 1st play data is in RAM, so plays faster, but 70fps should be real value.

Other than this- in best modes both debayers are so similar. Set Resolve to sharp, Arri sharpness to eg. 90 and they look almost identical. By the default Arri is sharper, but this doesn't really mean better.

Thanks for the tests Andrew, really appreciated!