Brian Boyer
Well-known member
- Joined
- Jan 4, 2011
- Messages
- 943
- Reaction score
- 6
- Points
- 0
It seems CUDA is not functioning at all in RCX or the RED Player on my machine. OpenCL breezes through caching clips like it always has. Switching to CUDA, the caching slows to a crawl.
No matter what I set the debayer resolution to it caches the frames at the same speed. At 1/16 it's supposed to fly by with the most modest of graphics cards. It's as if I've set the card to use no GPU at all and it's relying solely on the CPU.
I've tried the "Disable GPU" setting and it renders the same speed as when CUDA is selected, which is why I think CUDA isn't doing anything.
With the same 780 Ti card installed but running El Capitan (10.11) on another drive in the same machine and using Redcine-X Pro 33, CUDA behaves the way it's supposed to, even a little faster than OpenCL.
I'm using Redcine-X Pro 50.2.44748 in macOS High Sierra 10.13.4 running on a mid-2010 Mac Pro Tower with 24 GB RAM and an EVGA 780 Ti graphics card with 3 GB of RAM. The NVIDIA Web Graphics Driver version is 387.10.10.10.30.106 and the CUDA Driver is 387.178
I know the RAM in the card is less than the suggested 6GB minimum. I have a 1080 Ti with 11GB ready to be installed but I'm not sure it will make a difference since CUDA doesn't seem to be doing anything.
The 3GB of RAM doesn't affect the OpenCL caching. Perhaps it will make the difference but I didn't want to make a change until I knew for sure.
I tried reinstalling RCX and trashing the preferences file but neither worked.
Does anyone have any insights?
Thanks
UPDATE
Ok, Redcine-X version 50.2 is definitely the culprit.
As I mentioned before, RCX 33 running on El Capitan (now upgraded to 10.11.6) works as expected with the GTX 780 Ti, even with just 3GB of RAM. The debayer for 4KHD (3840 x 2160) clips at 1/2 res and 6:1 outpaced the playback head.
For 4KFF (4096 x 2160) clips of the same quality the debayer was just a hair slower than the playback such that, if you cached for maybe 5 seconds, the clip would play for a long time before stopping and having to re-cache.
I did end up installing the 1080 Ti but could only use it when booted on the High Sierra drive. I tried booting on the El Capitan drive as well but I couldn't get it to work, even with the latest NVIDIA drivers for 10.11.6.
I think I read it has to do with the Pascal architecture not being supported in El Capitan but I'm not sure.
Anyway, in RCX 50.2 on High Sierra, the 1080 Ti plays back better than the 780 Ti's excruciatingly slow performance. However, for a card that's 1.9x as fast at OpenGL renders according to Octane Bench, it plays back slower than the 780 Ti does in RCX 33 on El Capitan.
The sheer speed of the 1080 Ti is all that's allowing it to overcome 50.2's poor CUDA performance and keep a modest pace.
So, to make sure High Sierra wasn't the problem, I updated RCX on the El Capitan drive from 33 to 50.2. Sure enough, the 780 Ti CUDA performance slowed to a crawl where a mere 3 minutes before it was cruising along.
And, like on the High Sierra drive, the caching took the same amount of time no matter which resolution I chose. 1/16 was as slow as 1/2.
I thought 50.2 was slow AF (with the 780 Ti) before I upgraded to High Sierra but I thought it was a combination of my imagination and the fact that I've only been looking at 8K stills and clips lately.
As it turns out, I was right. I can barely play my old 4K Scarlet-MX clips.
No matter what I set the debayer resolution to it caches the frames at the same speed. At 1/16 it's supposed to fly by with the most modest of graphics cards. It's as if I've set the card to use no GPU at all and it's relying solely on the CPU.
I've tried the "Disable GPU" setting and it renders the same speed as when CUDA is selected, which is why I think CUDA isn't doing anything.
With the same 780 Ti card installed but running El Capitan (10.11) on another drive in the same machine and using Redcine-X Pro 33, CUDA behaves the way it's supposed to, even a little faster than OpenCL.
I'm using Redcine-X Pro 50.2.44748 in macOS High Sierra 10.13.4 running on a mid-2010 Mac Pro Tower with 24 GB RAM and an EVGA 780 Ti graphics card with 3 GB of RAM. The NVIDIA Web Graphics Driver version is 387.10.10.10.30.106 and the CUDA Driver is 387.178
I know the RAM in the card is less than the suggested 6GB minimum. I have a 1080 Ti with 11GB ready to be installed but I'm not sure it will make a difference since CUDA doesn't seem to be doing anything.
The 3GB of RAM doesn't affect the OpenCL caching. Perhaps it will make the difference but I didn't want to make a change until I knew for sure.
I tried reinstalling RCX and trashing the preferences file but neither worked.
Does anyone have any insights?
Thanks
UPDATE
Ok, Redcine-X version 50.2 is definitely the culprit.
As I mentioned before, RCX 33 running on El Capitan (now upgraded to 10.11.6) works as expected with the GTX 780 Ti, even with just 3GB of RAM. The debayer for 4KHD (3840 x 2160) clips at 1/2 res and 6:1 outpaced the playback head.
For 4KFF (4096 x 2160) clips of the same quality the debayer was just a hair slower than the playback such that, if you cached for maybe 5 seconds, the clip would play for a long time before stopping and having to re-cache.
I did end up installing the 1080 Ti but could only use it when booted on the High Sierra drive. I tried booting on the El Capitan drive as well but I couldn't get it to work, even with the latest NVIDIA drivers for 10.11.6.
I think I read it has to do with the Pascal architecture not being supported in El Capitan but I'm not sure.
Anyway, in RCX 50.2 on High Sierra, the 1080 Ti plays back better than the 780 Ti's excruciatingly slow performance. However, for a card that's 1.9x as fast at OpenGL renders according to Octane Bench, it plays back slower than the 780 Ti does in RCX 33 on El Capitan.
The sheer speed of the 1080 Ti is all that's allowing it to overcome 50.2's poor CUDA performance and keep a modest pace.
So, to make sure High Sierra wasn't the problem, I updated RCX on the El Capitan drive from 33 to 50.2. Sure enough, the 780 Ti CUDA performance slowed to a crawl where a mere 3 minutes before it was cruising along.
And, like on the High Sierra drive, the caching took the same amount of time no matter which resolution I chose. 1/16 was as slow as 1/2.
I thought 50.2 was slow AF (with the 780 Ti) before I upgraded to High Sierra but I thought it was a combination of my imagination and the fact that I've only been looking at 8K stills and clips lately.
As it turns out, I was right. I can barely play my old 4K Scarlet-MX clips.
Last edited: