|
|
Quote
Why does PowerDirector 17 consistently use less than 50% of the available CPU capacity on my PC when producing?
My CPU utilisation when producing: a maximum of 51% (48% average)!
Video producing could be twice as fast, if PD17 used the available processing capacity.
PD17 takes on averge approx. 30m to render a 60m H265 video on my workstation, but it could be twice as fast if it fully utilised the available CPU cores and speed. RAM usage was consistently around 5%. SSD read/write was also minimal.
Why does PD17 seriously under-utilise the available hardware resources? What is the limiting factor here? Is there anything that can be done to configure PD17 to use the hardware more fully?
Also, why is "Fast video rendering" (including both "SVRT" and "Intel Quick Sync" unavailable, with this option disabled on the Produce screen?
The spec of my machine is as follows.
- CPUs: 72 (3.0 GHz Intel Xeon Platinum 8124M with Intel Advanced Vector Extension 512 instruction set)
- CPU Cores: 36
- RAM: 144 GB
- SSD2 (location of project and videos): 160MB/sSSD1 (location of OS and PD17): 160MB/s
- OS: Windows Server 2019 (Datacentre Edition) v10.0.17763 [the same build as Windows 10 version 1809]
- No graphics card.
PowerDirector version: 17.3.2721.0 (PowerDIrector 365 type installation)
PDR.exe process priority setting in WIndows: Realtime
PDR.exe Graphics Performance setting in Windows: High Performance
Tested with and without CPU Turbo (3.5 GHz) on a range of projects producing 60 minute videos, both and without extensive effects (e.g. skin smoothing) and enhancements (e.g. colour), in H264 and H264 with various 1080p profiles.
I also ran my benchmarks on comparable rigs including one with an Intel Xeon E5-2686 v4 CPU:
- 64 CPUs, with 4 NVIDIA Tesla M60 GPUs, and and measured no significant performance difference. I used the latest Tesla drivers, and had hardware acceleration enabled and optimised in PD17. PD17 gave me the OpenCL and hardware decoding options. The GPUs were optimised and configured to run at their maximum speed. Each CPU has 2048 parallel processing cores and 8 GB of video memory, plus an on-board hardware video encoder/decoder supporting up to 10 H.265 (HEVC) 1080p30 streams and up to 18 H.264 1080p30 streams simultaneously.
- 96 CPUs, and 8 NVIDIA Tesla V100 GPUs, each pairing 5,120 CUDA Cores and 640 Tensor Cores! I used the latest CUDA drivers, and had hardware acceleration enabled and optimised in PD17. PD17 gave me the OpenCL and hardware decoding options. The GPUs were optimised and configured to run at their maximum speed.
CONCLUSION
As you upgrade your hardware, PD17 hits a performance wall where it cannot go any faster. When you hit that limit, which is surprsingly low, no hardware upgrades can make video production any faster. CyberLink customers need to be aware of what these limits are, so they don't waste money on hardware capacity that will not be used. You could invest in more CPUs, better CPUs, more CPU cores, and you could install 24TB of RAM, and you could add more GPUs, faster GPUs, with more on-board graphics memory, and after a point none of it will speed-up video production because PD will not use it.
In 2023 I have similar problem:
https://forum.cyberlink.com/forum/posts/list/0/118604.page#446365
Did you find any solution to use the full power of the CPU?
|
|
|
Quote
100% load as you show is very typical for H.265 CPU encoding, is that what the example is?
However, you don't get the 2X reduction in elapsed time claiming one has 8 procs at 100%, I believe that's the common thought on the forum. If one turns HT off, you will see you are again 100% on 4 cores in this case but elapsed time will be maybe be only 10-15% better with HT on, not 2X.
HT is nowhere near linear, it's typically better on low core count CPU's as your example, you may even get a 20% boost, and much worse at high core counts more typical in modern high end CPU's. At best, HT in a large core count CPU asymptotes at about a 25% improvement with very few real applications ever achieving that.
For me it's more like if you want to look at real CPU loads, turn HT off, if you only casually care about loads and want a slight boost in performance, turn HT on. I've never seen HT hurt PD performance, however, it's benefits can be very marginal to maybe 10-15% benefit for some tasks and benefit can be CPU core count sensitive.
Jeff
I use only H.264. Original recorded files are in H.264 and I produce H.264 too.
But as I wrote - If long working computer process doesn't consume 100%, thats means (as I think), that process waiting for something. But waiting for what?
|
|
|
Quote
I don't understand what you changed between your tm and tm2 pics so can't comment at all.
There is no setting you can adjust to change load for a given set of produce conditions. It is what it is. Likewise, H.264 encoding will have a totally different load than H.265 encoding.
In my view, it's not proper to expect high CPU load when SMT or Hyper-Threading is active as PD/CD doesn’t effectively use those processors. In my experience, if you CPU produce a project with SMT on and off you should see dramatic CPU % load difference, but the elapsed time to do the exact same produce task with the exact same settings will only be a few % different. Not 2X different as the logical processor increase.
Jeff
tmp2 is after a few minutes, when the load on the GPU has decreased and the CPU is still not fully utilized. I am based on my assumption that if the CPU load is not at 100%, then something must be waiting somewhere. I.e. is there some component somewhere that is the bottleneck. And tm2 shows that the GPU (integrated in the CPU) is not. If it is not waiting for the hardware, then there must be some problem in the software.
After turning off SMT (hyperthreading on AMD), the CPU load indicator in the task manager rose to 85%, however, the video production time did not change. What's more, CPU wattage hasn't gone up either. At the same time, the CPU stress test utility gets my CPU to 100% and the CPU consumption in watts increases by a few tens of watts.
The time to export my project on the new PC is 48 minutes. On an old PC it is 2 hours 8 minutes. The new PC is therefore 2.66 times faster.
New PC
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+7900X&id=5027
Benchmark value 52285
Old PC
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-8700+%40+3.20GHz&id=3099
Benchmark value 12964
I don't know exactly which operations I should take to compare performance, so I take the total value.
The ratio of benchmark values is therefore 4 times. Simply put, the new PC could be 4x faster.
From this and from the reported values of CPU consumption (using the ASUS utility), I believe that there is a problem somewhere in the cyberlink software.
But where?
|
|
|
Quote
I compared everything and I didn't see a difference. I also tried ColorDirector separately and PowerDirector separately, and during the production event I get to a maximum CPU load of 50-60%. I don't see a narrow place anywhere. I am adding a load in the task manager.
And now, in 10% ColorDirector encoding, there is no GPU usage. Bud CPU usage is still not full. Why?
|
|
|
Quote
You might just verify both your pref > Hardware Acceleration settings and you have the same "Produce" settings, in particular "Hardware video encoder" between the two computers for proper comparison of elapsed times to Produce.
It's not that CD doesn't use the GPU, it can use it for decoding and encoding, it's just that the color corrections are handled by the CPU after decoding and prior to encoding so that is typically the time consuming side of the operation. The GPU doing the decoding can significantly affect device load.
Jeff
I compared everything and I didn't see a difference. I also tried ColorDirector separately and PowerDirector separately, and during the production event I get to a maximum CPU load of 50-60%. I don't see a narrow place anywhere. I am adding a load in the task manager.
|
|
|
I have hyperthreading enabled on both computers. I can see 2x more cores in TaskManager than I have physical cores. That is, on the old PC I have 6 physical cores and therefore a total of 12. On the new PC I have 12 physical cores and therefore I see 24 cores in taskmanager.
On the new PC it is obvious that the PC is not running at full power. It shows 105W CPU power consumption. If I run the CPU Stress test program, it can actually load the CPU more and more consumption is seen.
Translated with www.DeepL.com/Translator (free version)
|
|
|
I bought a new PC with an AMD Ryzen 9 7900x for maximum performance. I didn't buy a graphics card because ColorDirector doesn't use it.
The original 6-year-old Intel i7 PC was loaded at 100% when rendering on all cores. On a new PC with SSD disks and DDR5, I only have about 50-60% rendering load. Temperature CPU is about 75 degrease C.
I would like to use all the power, after all, that's why I bought a new PC. Can you think of any tips on what to do with it?
|
|
|
|
And I'll add: I always use ColorDirector to fine tune the colors from different cameras. I'm not sure if ColorDirector uses a GPU and if not, the question is whether I even need a GPU. Wouldn't a GPU integrated in the CPU be enough?
|
|
|
|
I record concerts, I usually have 4-5 cameras in 4K and use powerdirector for editing. The existing Intel Core i7-8700 PC is at the limit of usability. I would like to buy a new PC for this work. What configuration of PC would you recommend for me at a price under $1500?
|
|
|
Hi,
is there disable switch for temporary disable all fix/enhance adjustments in preview windows.
When I check the final cut and already have the color adjustments on, the preview is too slow. To check the editing of the entire video and a possible final revision, it would be enough to temporarily turn off the adjustments.
Is it possible now? Or is it possible to add it to the next version?
Thanks
|
|
|
Hi,
is there disable switch for temporary disable all fix/enhance adjustments in preview windows.
When I check the final cut and already have the color adjustments on, the preview is too slow. To check the editing of the entire video and a possible final revision, it would be enough to temporarily turn off the adjustments.
Is it possible now? Or is it possible to add it to the next version?
Thanks
|
|
|
