- Posts: 1
- Joined: Sat May 15, 2021 10:16 am
- Location: Germany
- Real Name: Ringo Hoffmann
Sat May 15, 2021 10:50 am
Hey.
A friend of mine upgraded his PC yesterday with a new CPU and more RAM but with the same GPU. These are the specs of his system:
Ryzen 5 3600X, 16GB 3200MHz RAM, GeForce GTX 1060 (3GB VRAM). The footage was on a Samsung EVO 860 SATA SSD.
On rendering, the CPU load was only on average around 30-40% and the GPU was pin-pointed at 100% load with ~50-60 FPS during render (1080p60, QuickTime H.264 w/ CUDA). No effects have been applied to the raw 1080p60 H.264 MP4 footage recorded with OBS Studio.
That was way under my expectation for these specs when compared to my system with the following specs:
Ryzen 7 5800X, 32GB 3200MHz RAM, Radeon R9 380 OC (4GB VRAM). My footage was located on a Samsung EVO Plus 970 NVME PCI-e SSD.
When rendering there, my CPU is pin-pointed at 100% on all cores and my GPU is at around 10-12% load with the same output format & codec and the same footage format & codec, even when exporting onto a SATA HDD I have an output framerate of between 200 and 300 FPS.
Of course, because Radeon does not support CUDA, my GPU configuration is on OpenCL. When using OpenCL on my friends system, the GPU load drops down to around 20% and the CPU is barely even utilized with an output framerate of round 10 FPS.
I guess, when Resolve would not utilize on the GPU that hard on the system of my friend, he could also achieve up to 100 FPS during rendering when the CPU would be utilized 100% on all 6C/12T.
For clarification, neither the SSD with the footage on nor the SSD/HDD where the export location was set to was bottlenecking according to the usage statistics of the Task Manager.
So, is it possible to somehow bypass CUDA on low spec NVIDIA GPUs so that we can use 100% of our CPU power for rendering?
A friend of mine upgraded his PC yesterday with a new CPU and more RAM but with the same GPU. These are the specs of his system:
Ryzen 5 3600X, 16GB 3200MHz RAM, GeForce GTX 1060 (3GB VRAM). The footage was on a Samsung EVO 860 SATA SSD.
On rendering, the CPU load was only on average around 30-40% and the GPU was pin-pointed at 100% load with ~50-60 FPS during render (1080p60, QuickTime H.264 w/ CUDA). No effects have been applied to the raw 1080p60 H.264 MP4 footage recorded with OBS Studio.
That was way under my expectation for these specs when compared to my system with the following specs:
Ryzen 7 5800X, 32GB 3200MHz RAM, Radeon R9 380 OC (4GB VRAM). My footage was located on a Samsung EVO Plus 970 NVME PCI-e SSD.
When rendering there, my CPU is pin-pointed at 100% on all cores and my GPU is at around 10-12% load with the same output format & codec and the same footage format & codec, even when exporting onto a SATA HDD I have an output framerate of between 200 and 300 FPS.
Of course, because Radeon does not support CUDA, my GPU configuration is on OpenCL. When using OpenCL on my friends system, the GPU load drops down to around 20% and the CPU is barely even utilized with an output framerate of round 10 FPS.
I guess, when Resolve would not utilize on the GPU that hard on the system of my friend, he could also achieve up to 100 FPS during rendering when the CPU would be utilized 100% on all 6C/12T.
For clarification, neither the SSD with the footage on nor the SSD/HDD where the export location was set to was bottlenecking according to the usage statistics of the Task Manager.
So, is it possible to somehow bypass CUDA on low spec NVIDIA GPUs so that we can use 100% of our CPU power for rendering?