Welcome to our community

Be a part of something great, join today!

Register 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 :)

IBC - GPU Build REDCINE-X Beta 21

Les Dittert said:
Grant,
I was just able to test this theory out and it is not the case. The difference is magnified especially if you turn up the sharpness settings, there is a fixed pattern grid type of pattern that results from the GPU accelerated version.
this artifact is noticeable also in slightly out of focus regions of the image, not just the super sharp stuff that I like to shoot.
Click to expand...

Thanks for the report, Les. We'll check it out!
 
Hi Les,

Please email your *exact* debayer settings to redcinex at red.

Thanks,
David

Les Dittert said:
Grant,
I was just able to test this theory out and it is not the case. The difference is magnified especially if you turn up the sharpness settings, there is a fixed pattern grid type of pattern that results from the GPU accelerated version.
this artifact is noticeable also in slightly out of focus regions of the image, not just the super sharp stuff that I like to shoot.
Click to expand...
 
Just did some testing with 4KHD footage on my i7 3930 6 core CPU (slightly overclocked) with one GTX690 card.

With previous non-gpu accelarated builds I had to run at 1/4 debayer to get smooth playback with 1/2 debayer possible if I waited for the clip to buffer a little. Now I can play 1/2 debayer very smoothly. Full debayer is also possible if I wait for the clip to start buffering for a second or two, but occasional stuttering and image "tearing" can be seen. As soon as the clip loads, the CPU is at 100%.

When I monitor my GPU usage with GPU Shark, my single GTX 690 shows up as 2 cards. Its showing that only 1 of the two cards is being utilized by REDCineX and that its only using about 50% running full debayer for 4kHD footage. At 1/2 debayer, the GPU is only running at about 25%. This leads me to believe that currently my system's bottleneck is the CPU.

I don't have any 5k footage handy, but I will have to try that next. I'm sure I can run 1/2 debayer on my system but I don't think it will be good enough for full debayer.

This performance is simply amazing!!! I had been thinking of upgrading my system to a dual xeon but now with GPU accelerated debayer I don't need to.

EDIT: I've noticed that on fast pans there is image tearing for 1/4, 1/2 and full debayer.
 
Last edited:
Danal,

The 690 is showing up as 2 independent GPUs because it is actually 2x GTX 680 on one board. Currently, we only support using one GPU at a time in REDCINE-X, but that will change once we get a little further down the road. But that's why you're only seeing one of the GPUs being utilized when monitoring.
 
Just want to mention that a faster GPU or a multi GPU setup (once RCX supports that) is not gonna help if your CPU's are pegged at 100%. In that case you'll probably wanna upgrade to more and/or faster CPU's. Of course if you use some of the effects GPU usage will go up.
 
I also have a Quadro 600 in my system. With the Quadro 600, my system can do 1/2 debayer for 4KHD footage in real time. My CPU's are at 100% and my Quadro 600 is at about 50%. When I switch to full debayer I no longer get realtime playback. My CPU's don't even reach 50% and my Quadro 600 is pegged at 100%.

I also see image tearing with the Quadro 600 selected for GPU processing.
 
Danai: can you post a screenshot of the tearing?
 
Les Dittert said:
Grant,
I was just able to test this theory out and it is not the case. The difference is magnified especially if you turn up the sharpness settings, there is a fixed pattern grid type of pattern that results from the GPU accelerated version.
this artifact is noticeable also in slightly out of focus regions of the image, not just the super sharp stuff that I like to shoot.
Click to expand...

Same issue here with the "fixed pattern grid type of pattern" ​on rendered output.
 
Rob,

The image tearing is only showing when playing back the R3D clips in RedCineX. When exporting the clip to quicktime, the image is perfectly normal.

I tried to do a screen capture but the image tearing didn't show up in the capture. Or maybe it happens randomly and quickly enough that I just couldn't do a screen capture at the right moment. I also noticed that I could induce a similar image tearing artifact by moving a window around quickly inside Photoshop. I'm running fairly old nVidia drivers (version 310.9) since I'm using the recommended one in the Davinci Resolve config guide.
 
Les Dittert said:
there is a fixed pattern grid type of pattern that results from the GPU accelerated version.
Click to expand...

Red Rocket has a fixed pattern grid as well that is visible when you use nearest neighbor scaling to downscale the image. It was often described as a waffle pattern on the SCRATCH forum. It would not be visible when using bicubic. I don't recall if it was there with bilinear or not.
 
Red Rocket has a fixed pattern grid as well that is visible when you use nearest neighbor scaling to downscale the image. It was often described as a waffle pattern on the SCRATCH forum. It would not be visible when using bicubic. I don't recall if it was there with bilinear or not.
Click to expand...

Stacey: that should not happen with the GPU path
 
Guy Staley said:
Same issue here with the "fixed pattern grid type of pattern" ​on rendered output.
Click to expand...

Hi Guy,

Please email your *exact* debayer settings to redcinex at red.

Thanks,
 
I dont really speak computer.

WIll this improve native performance on a standard MacBookPro or only with the specified graphics cards. If only with the graphics cards, will a Sonnet Thunderbolt expansion chassis with a Titan be useable or is it "tower only" territory?
 
GPU differences

GPU differences

Here is a 400% blowup in a jpeg.

GPU_difference.jpg
 
Jett He said:
Anyone with a GTX 580 and/or GTX 680 that want to share their experience with these two cards? I'm certain the 580 has more compute power, but RCX might be utilizing 680 differently to achieve better results.
Click to expand...

Dual Xeon 5650 (12 Cores, 24 Threads @ 2,66 GHz) here with 24 GB of RAM and a single GTX 480 on Windows 7 Professional. I only own a Scarlet so I tested 4K footage @ 24fps and I got real time playback off an internal RAID 0 reserving 20 GB of RAM to Redcine-X. Impressive!

I believe that the 580 or the 680 will fare even better at 5K.
 
Nicolò Rebughini said:
Dual Xeon 5650 (12 Cores, 24 Threads @ 2,66 GHz) here with 24 GB of RAM and a single GTX 480 on Windows 7 Professional. I only own a Scarlet so I tested 4K footage @ 24fps and I got real time playback off an internal RAID 0 reserving 20 GB of RAM to Redcine-X. Impressive!

I believe that the 580 or the 680 will fare even better at 5K.
Click to expand...

What is your CPU usage like with 12 cores running 4k full debayer?
 
When we say that 'how fast the CPUs can throw data at the GPU' is the limitation - I wonder where the CPU bottleneck is?

If Danai's 2 x (6 core) Xeon 5650 @ 2.66 GHz ... with (I guess) 3 channel DDR3-1333 - are running at 90% CPU utilisation because all threads are blocked waiting to fetch memory to pass to the GPU... maybe the issue isn't number-of-Cores ... but memory bandwidth?

If the New mac pro also has 12 cores @ 2.7 Ghz (not including turbo) with 4 channels of DDR3-1866.
Is should process (roughly) 4 x 1866 / 3 x 1333 times more memory = 1.87 Times faster.

This bodes well for editing dragon footage.

AJ
 
You must log in or register to reply here.
Back
Top