Rocket X transcode speeds

[Eric Haase]

Now that there are some rocket X cards out there can those who have them please post transcode speeds for 6k and 5.5k dragon footage (full debayer)?

in another thread it was stated that it took 1:52 to transcode 15s worth of material, which is 7x realtime. I find this hard to believe considering the card has been advertised as 3x-5x faster than the old Rocket card.

It it would be great to get a sampling from various users of transcode times. If you can, please include media and connection used (Ssd, external drive, esata, USB 3, etc).

thanks!

 

[Bob Gundu]

I'm getting the feeling there is still optimization work to be done.

 

[Neil W. Smith]

Now that there are some rocket X cards out there can those who have them please post transcode speeds for 6k and 5.5k dragon footage (full debayer)?

in another thread it was stated that it took 1:52 to transcode 15s worth of material, which is 7x realtime. I find this hard to believe considering the card has been advertised as 3x-5x faster than the old Rocket card.

It it would be great to get a sampling from various users of transcode times. If you can, please include media and connection used (Ssd, external drive, esata, USB 3, etc).

thanks!

Eric, these are early days ... RedCine-X Pro is still in beta ... new firmware and drivers for the RR-X are coming out daily .. the testing I did on Phil's DRAGON PAINT footage was more about the render quality between RR-X and software only on FULL debayer ... we're nowhere near the final optimization of RR-X ... the most important take-away from the rendering I did is that without a RR-X card it took over 7 minutes ... with the RR-X it took under 2 minutes ... that differential is only going to get wider.

Just relax and enjoy the wonderful images coming out of the DRAGON and know that when it comes to rendering and finishing DRAGON projects there's a means to do it fast and efficiently.

Cheers,
Neil

 

[Johnny Friday]

Why was your thread closed Neil? i thought there was some good info. coming from there. After thinking more on the rending quality tests you did....and thanks for those, but who really does a final render/output from Redcine-xPro? At least for a final outputted movie.....I think it's good to know the quality is better when outputting from RCX with the new RR-x card, but in a real world situation, i think some further testing of outputting via NLE's with and w/out RR-x card would be far more valuable.

....not that i don't appreciate all of your hard work and effort...far from it. It merely got me thinking of the alternatives--as stated above. So if debayering and transcoding within RCX-pro with the RR-x card produces a superior image vs. the same test w/out a RR-x card, then what does that then mean when transcoding or outputting a final movie file with and w/out a RR-x card.

I hope there is some time to get to that soon and see some even more real world examples....
thanks again Neil.

 

[Gabriele Turchi]

i guess that the Software vx R-X quality difference in RCX would be the same on any system that use RED SDK ?

am i wrong ?

thanks

 

[Blair S. Paulsen]

IMO its way too early to benchmark RR-X speeds. Moreover, the alternative option - multi-GPU acceleration - is still incubating.

I'm as curious as the next person, but IMO extrapolating from early implementations is a waste of time.

Cheers - #19

 

[joe hedge]

So I ran essentially the same test that Neil did, using a 5,1 8-core MacPro/GTX 680 accelerated, no RR card, to/from 500mb/s internal raid

The 15 seconds worth of clips took 56 seconds to transcode to ProRes 4444 1080, or slightly less than 4:1 transcode ratio, full res debayer

I'm mystified why it took Neil 7 minutes to do the same thing (I assume) on a Titan-equipped system. That's 28:1!


Also rendered out 6144 x 2592 DPX files (same res as the .R3D's), one of them took 50 seconds, then RC-X crashed - so all 5 would have taken 4:10

 

[Shiblon Wixom]

I believe they did CPU only, no GPU support.

 

[Bob Gundu]

Joe, Neil transcoded to the original size of the clips and not to 1080 as you did. I'm curious about your speeds with the same setup. BTW, on my macpro retina, the test took 8min 22seconds.

 

[joe hedge]

Joe, Neil transcoded to the original size of the clips and not to 1080 as you did. I'm curious about your speeds with the same setup. BTW, on my macpro retina, the test took 8min 22seconds.

Out to 6144 x 2592 took 1:33 for all 5 clips with the same machine and transcode parameters as above (ProRes 4444)


So 6:1 with no RR - but again, why 28:1 with a Titan system?

 

[Robert Ruffo New]

Why was your thread closed Neil? i thought there was some good info. coming from there. After thinking more on the rending quality tests you did....and thanks for those, but who really does a final render/output from Redcine-xPro? At least for a final outputted movie.....I think it's good to know the quality is better when outputting from RCX with the new RR-x card, but in a real world situation, i think some further testing of outputting via NLE's with and w/out RR-x card would be far more valuable.

....not that i don't appreciate all of your hard work and effort...far from it. It merely got me thinking of the alternatives--as stated above. So if debayering and transcoding within RCX-pro with the RR-x card produces a superior image vs. the same test w/out a RR-x card, then what does that then mean when transcoding or outputting a final movie file with and w/out a RR-x card.

I hope there is some time to get to that soon and see some even more real world examples....
thanks again Neil.

Red Rocket does a line-skip - not sure of Rocket-X is any different. It gives output which look almost identical to half-rez debater. if you think that kind of cheap sharpening effect looks "better" than a real full debayer you and I have very different ideas of what good is. GPU debayer may also skip corners, and be even worse - that I do not know.

 

[Johnny Friday]

Red Rocket does a line-skip - not sure of Rocket-X is any different. It gives output which look almost identical to half-rez debater. if you think that kind of cheap sharpening effect looks "better" than a real full debayer you and I have very different ideas of what good is. GPU debayer may also skip corners, and be even worse - that I do not know.

Robert, what are you stating? what is a "real full debayer"? You are stating that outputting via redrocekt or GPU is not good, but what other method can then be used? Straight debayer from CPU? is that what you mean?

 

[M Most]

Red Rocket does a line-skip - not sure of Rocket-X is any different. It gives output which look almost identical to half-rez debater. if you think that kind of cheap sharpening effect looks "better" than a real full debayer you and I have very different ideas of what good is. GPU debayer may also skip corners, and be even worse - that I do not know.

And you got this information where?

 

[Johnny Friday]

Red Rocket does a line-skip - not sure of Rocket-X is any different. It gives output which look almost identical to half-rez debater. if you think that kind of cheap sharpening effect looks "better" than a real full debayer you and I have very different ideas of what good is. GPU debayer may also skip corners, and be even worse - that I do not know.

Robert,
Maybe posing a grab frame example of both that illustrates your point. I've not noticed any discernable difference in a debayer w/out my rocket vs. with my rocket card. Maybe i need to pay closer attention, but from your post it seems that you see a drastic difference in the debayering process w/out a red rocket card. I would be very interested in knowing how bad this is if in fact what you are stating rings true.

 

[Michael Lindsay]

Red Rocket does a line-skip - not sure of Rocket-X is any different. It gives output which look almost identical to half-rez debater. if you think that kind of cheap sharpening effect looks "better" than a real full debayer you and I have very different ideas of what good is. GPU debayer may also skip corners, and be even worse - that I do not know.

?? I cheked (when I 1st got a rocket) exactly what you are saying and did not find it to be the case... Maybe things have changed but it seems unlikely?