2
$\begingroup$

In system prefs the CUDA tab shows my card GeForce RTX 2080 SUPER. I've selected this option, and de-selected my CPU (3900x). I then select via the inspector tab the Scene options, then select :

Render Engine : Cycles Feature Set : Supported Device : GPU Compute

I then render the scene via

Render->Render Animation

The CPU is being used first (can see via task manager), and the GPU goes to about 8% per png image rendered. Switch to CPU not GPU and the the CPU goes up to 90% and GPU still uses 8%. It takes the same amount of time to generate an image ~30 seconds.

What is the expected speed up for a 7.5 rated CUDA card, with > 3000 cores on it ? Does anyone else have any stats when using CUDA with Super 2080 please ?

I'm using

Blender 2.83 LTS Windows 10 CUDA 11 Installed NVidia Driver : 451.48 for RTX Super 2080

I've noticed the fastest rendering settings are Cycles, Supported, GPU Compute - but setting the OptiX in the system settings - with this each PNG renders in around 20 seconds.

Improve this question
$\endgroup$
7
  • $\begingroup$ Blender 2.80.xx or Blender 2.83.xx ? You might need to clarify your version more specifically. And what scene? Rendering Performance Tab? If you don't need some feature, 2080s should use OptiX backend. $\endgroup$
    – HikariTW
    Commented Jul 7, 2020 at 7:36
  • $\begingroup$ Hello :). Regarding stats, you can use Blender's benchmark page $\endgroup$
    – jachym michal
    Commented Jul 7, 2020 at 10:59
  • $\begingroup$ @hikariTW - is there a difference between 2.80 and 2.83 wrt handling render processing then ? And how many ways are there to render a scene, I was only aware of 1 way - ie Render->Render Animation. $\endgroup$
    – Marcus O'Brien
    Commented Jul 8, 2020 at 3:58
  • $\begingroup$ @JachymMichal - many thanks for the info, I'll take a look. $\endgroup$
    – Marcus O'Brien
    Commented Jul 8, 2020 at 3:59
  • $\begingroup$ @MarcusO'Brien don't ask new questions as comments, open a new question. This is not a forum. Please read the tour to understand how to make the best use of this site. $\endgroup$
    – susu
    Commented Jul 8, 2020 at 4:03

1 Answer 1

Reset to default
2
$\begingroup$

Brainscattered answer:

Use RTX not CUDA if you want to see an improvement in render speed.

The CPU is never idle when rendering in GPU, there will always be activity.

If the scene uses only 8% of the GPU it means that the scene is not very complex, hardly a challenge for the graphics card. Blender will not use more than what is needed. With a more complex scene and more samples, etc you will see the GPU work harder (Can you make the cpu go to 100% when typing in word or when reading email? )

In GPU one tile per GPU is rendered, whereas the CPU renders many tiles simultaneously, one per thread, so it is perfectly possible that a multicore-multithread CPU can be faster than GPU.

And forget about measuring CPU and GPU usage, those numbers say very little when it comes to the overall performance of blender.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks for the answer, and the info. I knew posting on here would garner a bad rating (I seem to have attracted a -1 already about not being clear enough). I was hoping the GPU would be used in parallel just as the multi-threading on the CPU as you mention ie push multiple tile processing to the GPU at the same time, now I know the tool is not designed for that then I will just stick to using the OptiX backend - is that what you mean when you say use the RTX not CUDA ? $\endgroup$
    – Marcus O'Brien
    Commented Jul 8, 2020 at 3:54
  • $\begingroup$ Don't worry about the downvotes. All of the cores in the GPU work as a single unit, the CPU will use all of the threads as separate process. You can certainly render with GPU and CPU simultaneously. Yes Optix i meant. $\endgroup$
    – susu
    Commented Jul 8, 2020 at 4:01
  • $\begingroup$ A 20~30 sec rendering should consider as "complex" and one should notice system hardware usage. OP said that CPU is up to 90% when rendering his scene. Then the 8% usage for GPU is trust-able in this situation. CPU might struggle on calculating something and GPU didn't get enough jobs to done. Thus, CPU is then considered as a bottleneck here. Measuring CPU and GPU usage is a big thing when knowing how Blender use your hardware properly. $\endgroup$
    – HikariTW
    Commented Jul 8, 2020 at 4:27

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .