Benchmarking Hitfilm?

(I have re-uploaded this conversation to this category because it seemed to me that the "Hitfilm Pro support" category was not suitable for this thread. I apologize.)

Hi,

I decided to change my graphics card, but thinking about the performance I will gain, I came up with an idea:

Could I benchmark hitfilm?

It would be a perfect method to know if there has been a substantial change when changing the hardware.

I came up with certain ideas to do it:

  • -To measure the impact that RAM has on HF, it is enough to create an intensive project using different compositing techniques and make a RAM preview. That way, you can measure the time taken to make the ram preview.(I did the test and moving from RAM ddr3 1333Mhz to DDR4 3000Mhz provides a difference of 2 minutes less with ddr4 in an 8:30 min video.)

 

  • -You can also measure the impact of several CPUs in hitfilm by exporting different videos and measuring time.

For the test I made 3 projects:

Project1: basic 2k editing with color correction, light and flares e.t.c

Project 2: Particles and simulator (various emiters e.t.c)

Project  3: VFX 4K shot. Chroma key, light flares, color correction, masking e.t.c

Intel i9 9900K 8 cores EXPORT:

  1. Project 1: 12 minutes.
  2. Project 2: 1 minute.
  3. Project 3: 8 minutes

 

2X Intel x5670 12 cores EXPORT:

  1. Project 1: 30 minutes .
  2. Project 2: 2 minutes.
  3. Project 3: 9 minutes

These tests are quite reliable and accurate because performance is measured by measuring the time that hardware uses to complete a task: RAM preview, export e.t.c

But, could the performance of a GPU be benchmarked accurately?

The only way I've found to do it is by eye looking at the performance on the timeline. Is there any more accurate way to do it?

Thanks.

Comments

  • Triem23Triem23 Moderator

    I was about to merge your threads to delete your duplicate, but you beat me to it. Thanks for helping keep the forum focused. Tip of the hat to you! 

     

    I'll tag @NormanPCN-san* in to SAN check me, but, I believe it's very difficult to truly isolate the differences in GPUs because of system interdependence. In other words, slow CPU hampers GPU. This is why test/benchmark sites literally swap all the cards on a single system. To reduce the number of variables. This only is accurate for an identical system.

     

    FWIW I use userbenchmark.com for my comparisons. They crowdsource benchmarks, which means the final scores are calculated from a larger database. 

     

    Even more complex is comparing functions. If card X has blazing fast OpenGL but weak Cuda and card Y has weak OpenGL but amazing CUDA, card A is better for Hitfilm (OpenGL), but worse for things using CUDA (a lot).

    So, benchmarks are imperfect, specific benchmarks better - for related tasks. More than that is beyond my skills. 

     

  • @Triem23;

    Thanks!

    Please, feel free to correct me if I´m wrong but I understood this:

    Due to hardware bottlenecks e.t.c we are not going to be able to benchmark HF GPU. But hitfilm is based on OpenGL, so any OpenGL benchmark (preferably ones like userbenchmark that are based upon a large database) will work fine for determining HF speeds between cards.

    Am I correct?

     

    So, which GPU benchmark is the best for OpenGL?

    (I have thought that the Nvidia RTX 2070 super is the best option for hf rather than AMD due to hardware decod. e.t.c, but I am still in time to purchase the RX5700XT. Which card is better with OpenGL tasks?

     

    Thank you.

  • "To measure the impact that RAM has on HF, it is enough to create an intensive project using different compositing techniques and make a RAM preview. That way, you can measure the time taken to make the ram preview.(I did the test and moving from RAM ddr3 1333Mhz to DDR4 3000Mhz provides a difference of 2 minutes less with ddr4 in an 8:30 min video.) "

    Are you stating that you had the exact same CPU+motherboard using DDR3 and DDR4? I ask because this really cannot happen. There are select cases where a CPU+MB can be compatible with both DDR3 and DDR4. If it is not literally the same CPU+MB then the comparo is not valid.

    ---

    "Due to hardware bottlenecks e.t.c we are not going to be able to benchmark HF GPU. But hitfilm is based on OpenGL, so any OpenGL benchmark (preferably ones like userbenchmark that are based upon a large database) will work fine for determining HF speeds between cards.

    Am I correct?"

    No. Only a Hitfilm test is valid for measuring Hitfilm. Any other OpenGL test does not necessarily apply to Hitfilm. Largely because Hitfilm is doing things differently than virtually every other OpenGL app. OpenGL apps are typically 3D VR/game type apps.

    ---

    It is difficult, or impossible, to really separate out specifics like RAM speed, CPU speed and GPU speed in a test. The only thing one can do is create some representative projects and go with that. But really, unless you are a review site like Anandtech where you test all CPUs and GPU in combo against a suite of tests with what would your results really mean. Use benching reports can be iffy since virtually all users will not have proper standardized and rigorous test methodology. A lot of review sites don't.

     

     

Sign In or Register to comment.