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 :)

Anyone else seeing poor GPU debayer performance in 11?

Ben Cannon

Ben Cannon

Member
Joined
Aug 23, 2014
Messages
23
Reaction score
0
Points
0
Several clients and my own testing indicates worse performance in 11 than 10 debayering 5k clips in multi-GPU configs.

11 is almost behaving as though 1 GPU is installed, even though 2 show up and display GPU for compute is selected.

Redcine-X debayers with nearly 2x the performance of Resolve.

Anyone else seeing this?
 
What platform are you on? Can you list some HW details?
 
Dual GPU Mac Pros mostly. New and old. Multiple systems, multiple reports and my own lab equipment. DRAGON footage that debayers realtime (24fps) in Redcine is at about 12 fps in Resolve 11. 10 was quicker.
 
Is the GPU debayer in Resolve (now) single-GPU only?
 
I have also seen this on my system. 5K half res plays back at the same fps as full res in Resolve (11-13 for MX, 16-19 for Dragon, which is another interesting thing). In RC-X I get realtime or faster at half res premium for playback. Exports from RC-X seem to be more in line with Resolve's playback speeds, though.

Asus P9X79 Pro
i7 4390k @4.2GHz
32GB RAM @1600MHz
GTX 770 4GB
 
We've got several reports of multi-GPU systems that will play back full res Dragon footage in RC-X, but 12fps in Resolve.
 
Same problem here. 5K Playback is smooth at 1/2 res in RCX but 12fps at best in resolve 11.
 
I've been having this issue since the 11.0 betas.

I have 2x 6GB GTX780's in my 2010 8-Core Mac Pro (1 in Mac Pro, the other in a Cyclone 2707 external PCIe chassis) and Resolve is simply unable to play back 4kHD R3D files in realtime at Half-Res Good and sometimes even Quarter-Res Good (in 1080p working timelines), even without any grades applied to them. I've had to rely on Smart Caching for every RED based project I do because of this, which is a major drag.

On the other hand, Adobe Premiere Pro CC on the same system is able to playback the same files at Full Res without dropping frames, which is extraordinary.

I've had these issues on both Mac OS X 10.8.5 and 10.9.4. I thought upgrading to Mavericks, and using the latest NVIDIA "web drivers" for the GTX780s would help performance, but there was no change in performance after the OS upgrade.
 
Hi, comparison in playback speeds between apps tells only part of the story as the apps are very different. So in practice the question is, does v11 play slower than v10 and in our tests it's the same.

If you don't have realtime playback for your r3d files try setting the camera raw decider rate for playback to a lower setting. Note it's CPU + GPU for realtime full res playback and so it's not just a GPU comparison.
 
Peter Chamberlain said:
Hi, comparison in playback speeds between apps tells only part of the story as the apps are very different. So in practice the question is, does v11 play slower than v10 and in our tests it's the same.

If you don't have realtime playback for your r3d files try setting the camera raw decider rate for playback to a lower setting. Note it's CPU + GPU for realtime full res playback and so it's not just a GPU comparison.
Click to expand...
Yes, the performance is the same (for me at least) between v10 and v11. However, performance at Full Res and Half Premium are identical to each other. Logically I would assume that Half Premium would be less demanding than Full, and REDCine validates that assumption - Half Premium performs at about double the speed of Full Res. This also bears out in CPU only decoding, where there is a performance difference between Half Premium and Full. I'm just wondering why Resolve would not gain better performance by dropping from Full Res to Half Premium, when REDCine doubles performance in that jump. Between the two applications, Full Premium, Quarter, Eighth, and Sixteenth all perform roughly the same. Half is the anomaly.

Again, this is only with GPU acceleration - CPU only performs as expected. This has been the case with multiple systems I've tested, each with substantially different hardware configurations.
 
At this point I'm betting my lunch money that Resolve's RED debayer is single-GPU only...
 
No change in pure RED debayer performance going from single GPU to quad GPUs. MacPro tower. Resolve 11.1 final.
 
As Peter mentioned, both CPU and GPU are involved with Resolve processing of the r3d files. If you don't have enough CPU power to decode the file in real time, adding more GPU power won't do anything. So likely you are CPU bound.

Resolve certainly uses all available GPU's for the debayering operation.
 
in addition to cpu id check you storage speed and ram... gotta try and find the weak link...

ps my rocket x works great with resolve...at a certain point its a valuable asset in your kit...i was getting faster than real time dnx encoding, four times real speed on a laptop to non raid single drive solutions off mx 4khd. its an amazing piece of any kit no matter your apps of preference
 
Dwaine Maggart said:
As Peter mentioned, both CPU and GPU are involved with Resolve processing of the r3d files. If you don't have enough CPU power to decode the file in real time, adding more GPU power won't do anything. So likely you are CPU bound. .
Click to expand...


Would there be any significant performance improvement, in regards to realtime R3D decompression & debayering, by upgrading my 8-Core 2x2.4 GHz MacPro5,1 (2010) system to a 12-core setup, using one of these CPU upgrade options from OtherWorld Computing?

I'm running Dual 6GB GTX780s, and it really sucks that I can't sustain reliable realtime R3D debayering in Resolve if the pre-debayer decompression is bottlenecked by the CPU (although it hurts even more to know that Premiere Pro CC on the very same setup is able to reliably sustain GPU debayered footage at Half/Full quality).
 
You can upgrade the cpu in your Mac pro yourself and get huge r3d speed increase when combining with multiple gpu. And prices for the xeons that go in Mac pros have recently dropped significantly.
 
Paul Provost said:
You can upgrade the cpu in your Mac pro yourself and get huge r3d speed increase when combining with multiple gpu. And prices for the xeons that go in Mac pros have recently dropped significantly.
Click to expand...

I haven't done an in-place CPU upgrade on a system since the early 90's, when I thought I was totally hot-shit with my 486DX2, so I'm not really sure what goes into Mac CPU upgrades in the modern world...are you saying that I can just buy retail Xeons and drop 'em in myself, via what I'm assuming is some sort of ZIF socket mechanism? I did some cursory research, and a lot of people strongly recommend against DIY CPU upgrades in Mac Pros, because it appears that it's not as simple as just flipping a ZIF socket lever and dropping a new CPU in. I assume this is why OWC wants you to send them your entire CPU sled to them, in order to do the upgrade?

Have you done this upgrade yourself in a 2010 Mac Pro?
 
Never since the day I bought it was I all that happy with the performance I had been getting from the Quadro 4000. Can anyone recommend a software tool to evaluate GPU performance?
 
You must log in or register to reply here.
Back
Top