Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

REDCINE-X Beta CUDA Build 23.0.0 and Release Build 22.2.0

Jarred Land

Jarred Land

Fire Chief
Staff member
Joined
Dec 27, 2006
Messages
10,644
Reaction score
33
Points
10
Beta Build 23.0.0 and Release Build 22.2.0

Enjoy :)

Release Build 22.2.0:

Added support for new Dragon OLPF.
Added option to enable PSF for video out.
Improved logging and error checking with OpenCL.
Improved OpenCL stability.
Sped up thumbnail generation in certain cases.
Added CAM and EXT indicators to audio track labels.
Fixed the delete key not working in the browser.
Fixed the timeline ruler disapearing when the last sequence was removed.
Fixed issue where setting the video slate plate could cause a crash.
Fixed issue on Windows where filename with illegal characters could not be saved.
Fixed black frames when exporting to OpenEXR with certain AMD graphic cards.
Fixed OpenEXR MultiView exports.
Fixed OpenCL stereo 3D exports.
Fixed OpenCL HDR simple blend operation.
Fixed random crash when exporting still clips.
Fixed frame rendering issue with simple render mode for older video cards.
Fixed off center crop framing issue when Rocket was enabled.
Fixed FCPX XML import video tracks being out of order Fixed issue where HDR SimpleBlend failed with OpenCL enabled.
Fixed issue where QT wrapper generation would fail with certain export presets.


New release version:

https://www.red.com/downloads/4fad7fd417ef027cd8000eb1

Beta Build 23.0.0 ads CUDA Support: ( OPEN GL disabled in this build )

https://www.red.com/downloads/4f56854e6261f95a570006af
 
Hmm, all other things equal (same clip, same preferences, same exact export preset, same exact hardware) CUDA (Build 23 Beta) is takes almost 5 times as long to transcode a clip compared to OpenCL (Build 22). CUDA drives are completely up to date, as well. Transcoding a minute long clip on my laptop to 1080p ProRes 422 HQ takes 1:11 in build 22, and takes 5:11 in build 23.

The RCX preferences don't seem to allow me to select OpenCL in Build 23. Now that it sees that I have CUDA cores, it offers "CUDA" or "Disable GPU". No option for OpenCL.

Playback, also, seems slower than it was in the OpenCL version. Though that's harder to quantify for me, since on even OpenCL, I would do 1/2 Debayer and only be able to play, say, 10-20 seconds. But the CUDA version seems to give me 2-3 before it hiccups.

Hmm. I can't imagine this is the kind of performance degredation that would be expected in CUDA versus OpenCL. The playback and transcode times are close to (if not the same or worse) to what I got before any GPU acceleration was added to RCX.


UPDATE: Following a reboot (and only a reboot) transcode time in Build 23 a much more comparable 1:20. Still not as good as OpenCL, but the first OpenCL build wasn't nearly as good as Build 22, so if we can expect any improvement before the CUDA version leaves Beta . . .
 
Opencl is disabled in the beta cuda build so we can focus on cuda testing :)

Don't worry, once things are merged together you'll have both!
 
Rob Lohman said:
Opencl is disabled in the beta cuda build so we can focus on cuda testing :)

Don't worry, once things are merged together you'll have both!
Click to expand...

Awesome. That's what I kinda figured as I was typing that up.

Does my performance seem to be, about, expected? Just trying to glean any insight as to whether OpenCL or CUDA should eventually be the kind of rule-of-thumb solution. Although I'm guessing I'll just have to be patient until it hits release version to test for myself.
 
Jarred Land said:
Beta Build 23.0.0 and Release Build 22.2.0

Enjoy :)

Release Build 22.2.0:

Added support for new Dragon OLPF.
Added option to enable PSF for video out.
Improved logging and error checking with OpenCL.
Improved OpenCL stability.
Sped up thumbnail generation in certain cases.
Added CAM and EXT indicators to audio track labels.
Fixed the delete key not working in the browser.
Fixed the timeline ruler disapearing when the last sequence was removed.
Fixed issue where setting the video slate plate could cause a crash.
Fixed issue on Windows where filename with illegal characters could not be saved.
Fixed black frames when exporting to OpenEXR with certain AMD graphic cards.
Fixed OpenEXR MultiView exports.
Fixed OpenCL stereo 3D exports.
Fixed OpenCL HDR simple blend operation.
Fixed random crash when exporting still clips.
Fixed frame rendering issue with simple render mode for older video cards.
Fixed off center crop framing issue when Rocket was enabled.
Fixed FCPX XML import video tracks being out of order Fixed issue where HDR SimpleBlend failed with OpenCL enabled.
Fixed issue where QT wrapper generation would fail with certain export presets.


New release version:

https://www.red.com/downloads/4fad7fd417ef027cd8000eb1

Beta Build 23.0.0 ads CUDA Support: ( OPEN GL disabled in this build )

https://www.red.com/downloads/4f56854e6261f95a570006af
Click to expand...

Hi Jarred,

Is the new OLPF shipping/colour science as well?

Thank you very much

Fury
 
Hm, running into the same problem again where on my 2012 MBPr w/ Mountain Lion, I have the OpenCL enabled for my GeForce GT 650M. Double-clicking a video file to preview closes the program and gives me this error a split-second before crashing:

Failure converting parameter host image to CL image: out of resources.

I tried switching off the "Energy Saver" preferences, but no luck.
 
This is so bad ass.

I'm bloody excited about the CUDA support for my Titan to do some .R3D chewing. Looks like it's working great after installation!

HOWEVER.

What I am most excited for in both of these release builds......

"Added support for new Dragon OLPF."

Hello beautiful. Jump in my Dragon Brain so I can shoot with you.
 
Nice
 
Started it up (the Cuda-only/OpenCL disabled version) on my 5,1 MP 64 GB's RAM, 500 MB/s internal Raid, GTX 680, 10.8.3 OSX

Installed the latest Cuda driver as RC-X wasn't seeing the card, after which it did

RedRaw 4k 5:1 compression, 2:1 aspect to 1080 ProRes 1080 LT, half-res debayer


Transcoded a 2:30 clip in 10 minutes (4:1 ratio), or about 6 fps

Meanwhile Resolve now goes faster than it did thanks to the Cuda driver update (apparently), getting 16 fps transcode
 
I'll do some side by sides with both installed versions installed, but I'm feeling like DPX export is pretty darn slow at the moment with CUDA compared to OpenCL. Like night and day. Quicktime export was speedy though.

The buffering is really neat with CUDA. A bit more snappy in how it loads, but OpenCL feels more "constant". Maybe I'm crazy.

What I do like is that I buffered some 6K FF at full debayer and it played back like butter. I'd be curious to see how this would perform on a bigger system with much faster disks actually.
 
You must log in or register to reply here.
Back
Top