Slow export - CPU usage at 10%

Hallo,
during export the CPU usage is approx. 10% only. I use a Dual Xeon with 32 cores. Could it be possible to get the whole CPU power? Ten times more speed would be great.  ;) 
Ciao,
Micha

Comments

  • From what FXhome staff have said, Hitfilm is designed to use your GPU more than your CPU. If you upgrade your graphics card you can up your performance. I also believe that Hitfilm uses a 32-bit memory address, so it's only using 4GB of RAM.
  • As Triem23 mentioned, Hitfilm uses your GPU heavily.  When I first downloaded the demo, I saw the same thing - running slow and little to no CPU usage.  What GPU do you have?
  • I use a GTX285 2GB and the usages are:
    GPU 20..40%
    GPU memory 400MB
    CPU usage10%
    RAM usage of Hitfilm 400MB
    I'm using Win7 64 with 32GB RAM.
    For me it looks like the hardware is ok.
  • Triem23Triem23 Moderator
    edited November 2013
    Your GPU is four or five years old. In fact, your card isn't on FxHome's list of compatible/recommended graphics carda. You would certainly get a performance upgrade with a newer card. Look at the NVidia 500 or 700 series. The 600's weren't NVidia's best.
  • The problem is that my CAD software (Rhino3D) is using OpenGL and Nvidia limited all newer GTX cards, so that user buy the much more expensive Quadro cards. And the CAD software developer doesn't spend time to get the newer GTX cards running at full speed.
    Is there a chance to get the full power from the GTX285? This card could be 2..4 times faster if the Hitfilm would use the power.
  • Well, you're sort of confusing memory and power. If you're doing a relatively simple animation--and it looks like in what you're working on, you only have one 3D model at the moment--then you're not generating 2GB of data. However, whether it's a simple animation or a complex animation it's the speed of the card that determines render speed, not the card RAM.
  • Most I dosn't work with 3D models, only frame based work, so maybe the GPU isn't needed so much and there is an other problem of the slow speed.
  • According to FxHome staff, most of the render work in Hitfilm is GPU-based.
  • All rendering in HitFilm happens on the GPU, not only 3D models - the CPU is used for particle physics simulation for example. The reasons your GPU's power isn't fully used can have many reasons depending on your project. I could imagine (only guessing here!) for example that there are some tasks in your project that require to be computed one after another, but GPUs are generally better for simultaneos tasks, so that might be one reason it simply can't use all of it's power. Or it has to wait on other slower components of your computer, it might have to wait for data to be read from the hardrive or something like that.
    You see I have no clue how HitFilm works internally, but there are a bunch of reasons why this could be happening. Let's see what some of the devs say about this.
  • So far my conclusion from my limited users view: CPU and GPU are at low usage only, memory isn't used, no 3D model needs to be rendered, particles are not used too ...  :( 
  • They started gimping the lines at the 600 series... so go get a nice GTX 580 classified and enjoy.
  • Would a GTX580 help me to get one of them, CPU or GPU, at full power? At the moment my impression is that not the hardware is the limit, since non of them is at 100%.
  • Well, you're sort of confusing memory and power. If you're doing a relatively simple animation--and it looks like in what you're working on, you only have one 3D model at the moment--then you're not generating 2GB of data. However, whether it's a simple animation or a complex animation it's the speed of the card that determines render speed, not the card RAM.


    Yes, a GTX 580 will speed up your renders. I've quoted myself here, because, when you're saying your card isn't using "100 % of it's power," again, you're confusing memory use with how fast the card is running. Again, the demo animation you're giving screenshots of is only a single object--you don't have enough going on in your animation to require 2 GB of memory at the moment, but, again-again-again, your memory use has nothing at all to do with how fast your GPU can process data--it's the SPEED of your GPU that's slowing your render down.


  • ....-it's the SPEED of your GPU that's slowing your render down.


    Sorry, I don't understand. That is what I say too, the memory isn't a problem, because the project is small. And the GPU is using 20..40% speed load only. I had a better feeling to buy a new graphic card if my old GPU would be at 100% speed load. Could the 32 cores of the Dual Xeon the problem?
     

  • Ah--My apologies--above, I thought you'd said your GPU was using 20-40% of it's memory. I misread something you wrote and that got me on the wrong tangent.
  • Hello,

    I have the same problem. My cpu only run on 20 % during the export.

    But I have a gtx 1060 with 3 gb vram and an intel core i7-8700k. 

    Why is that so?

     

    Lukas

  • Triem23Triem23 Moderator

    @Lukas200213

    Hitfilm's render/export process order. 

    GPU is idle.

    CPU reads source files off drive. 

    CPU decodes video and stills media. 

    CPU calculates particle physics (Pro Only) 

    CPU passes data to GPU. CPU is now idle. 

    GPU generates all effects filters, procedural media, lighting, compositing layers, etc...

    GPU passes final image data back to CPU. GPU is now idle. 

    CPU encodes video frame. 

    CPU writes frame to output drive. 

    Repeat for each frame. 

    This handoff between processors means neither a CPU or GPU will ever peg at 100%--and such is never desirable for a PC anyway. 

    20% CPU usage means you have two cores nearly pegged. Media decode in Hitfilm is single-thread, I think--one core per stream (I could be wrong here. @NormanPCN is our resident user export, @CedricBonnier is Staff who may correct me, here.). Without knowing your project I'd guess you have one or two video streams in your project and Hitfilm is working perfectly efficiently. 

  • Media decode is multi-threaded per stream.

    It is never a problem to have a CPU or GPU peg to 100%. It's a movie thing where computers flip their switches and overload when worked. If you have an application that fully loads things then you want to have top notch cooling for CPU and/or GPU. Upgraded cooling. Very high heat can reduce the servicable life of the semiconductor junctions.

  • There are countless threads regarding this. The fact that there is this back and forth between the CPU and the GPU means that when one is waiting for the other, it is not busy. This is highly dependent on the project that is being exported. A really effect heavy project will spend most of its time in GPU whereas a slideshow for example will spend most of its time waiting for the frame to get to the GPU then come back.

    And yes, as Norman pointed out, media decoding is multi threaded.

  • Hi, friends.
    I have the same on here.... CPU at 80% and GPU at 20% (never more than this, tho)

    So, I tried to export a  1'30''video using Boris Continuum Chroma Key Studio.
    It tooks 20 min in a 4-core Phenom II 3.2 GHz /12 GB / GTX 650 Ti with.

    BTW, when rendered with DaVinci Resolve, it tooks 3 min, with same configurations (but the GPU was at 70-100%)

    OK, I know that Hitfilm doesn't use CUDA (just OpenCL)
    This is why I ask: is there a plan to FXHome implement support to CUDA, or some copyright issues are stiil "alive"?

    TIA

     

Sign in to comment