DS 4.12 rendering slow sometimes. The official position?

IsaacNewton

There have been a number of threads on the subject of DS 4.12 rendering very slowly (by a factor of 3 to 5 times slower). I have noticed that this happens after I have done two or three renders. I found that closing and restarting DS usually fixes the problem temporarily. To me this suggests a memory management issue. Does anyone know what the DAZ official position on this is? Is the bug acknowledged or not? If yes, then are they working on it?

Has anyone found a solution or workaround?

My system should be fast enough (new in 2019):

Win 10, i7 8700K, 3.7GHz, 32GM DDR4 RAM, Dual GeForce 1080 ti 11 GB (driver 441.66). DS has access to the CPU and both GPUs for rendering.

fastbike1

Don't check CPU for Iray rendering. Don't leave multiple render windows open. The GPU will not release the associated memory. DS 4.12 is not inherently slower than 4.11, at least by my tests. It is actually about 20% faster. I have also not found that I needed to close and restart Studio to clear VRAM. Completely clearing VRAM does take several minutes, but I recover a significant percentage of a given render when I close a render window. I will typically a smallish (750x1100 or 500x750) to check lighting and pose etc. If I do several of these, I will close all but one for the final render. Leaving one render open eliminates most of the time needed to load the scene.

IsaacNewton

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

kenshaw011267

IsaacNewton said:

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

Test it and see for yourself. I did and renders on my 1080ti were faster than renders on my 1080ti plus 2700.

IsaacNewton

kenshaw011267 said:

IsaacNewton said:

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

Test it and see for yourself. I did and renders on my 1080ti were faster than renders on my 1080ti plus 2700.

OK, will do, but that doesn't explain why! It seems illogical to me.

Richard Haseltine

IsaacNewton said:

kenshaw011267 said:

IsaacNewton said:

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

Test it and see for yourself. I did and renders on my 1080ti were faster than renders on my 1080ti plus 2700.

OK, will do, but that doesn't explain why! It seems illogical to me.

Perhaps because it's trying to juggle ebtween using the CPU to manage the process and using it to actually render, the constant switching might mean that the GPU gets left idle some of the time.

fastbike1

@IsaacNewton "I don't see why NOT using the CPU would help. Could you explain"

When CPU and GPU are checked, some of the processing is offloaded to the CPU, which is much slower than the CUDA cores. The CUDA Cores are parallel processors which is why multiple Nvidia GPUs or GPUs with more CUDA cores are faster than GPUs with fewer CUDA cores.

kenshaw011267

IsaacNewton said:

kenshaw011267 said:

IsaacNewton said:

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

Test it and see for yourself. I did and renders on my 1080ti were faster than renders on my 1080ti plus 2700.

OK, will do, but that doesn't explain why! It seems illogical to me.

The CPU has to do some work to keep the CUDA process active. You can see that for yourself by watching CPU usage on a GPU only render. You'll see a pretty constant low level usage.

So when the CPU is rendering it uses every thread at 100%, or as close as the system can manage. Therefore what ever baton waving process the GPU needs is waiting for a timeslice a lot of the time.This introduces latency which slows down the GPU render.

IsaacNewton

kenshaw011267 said:

IsaacNewton said:

kenshaw011267 said:

IsaacNewton said:

Thanks for the reply. I generally close renders that are finished or were test renders, but that does not solve the problem for me. I should make clear that this issue does not always occur, but often when starting a new session of renders. If it does not happen then, then I can usually render normally.

I don't see why NOT using the CPU would help. Could you explain?

Test it and see for yourself. I did and renders on my 1080ti were faster than renders on my 1080ti plus 2700.

OK, will do, but that doesn't explain why! It seems illogical to me.

The CPU has to do some work to keep the CUDA process active. You can see that for yourself by watching CPU usage on a GPU only render. You'll see a pretty constant low level usage.

So when the CPU is rendering it uses every thread at 100%, or as close as the system can manage. Therefore what ever baton waving process the GPU needs is waiting for a timeslice a lot of the time.This introduces latency which slows down the GPU render.

OK that makes sense.

I wonder, do DAZ3d or Nvidia give advice NOT to use CPU when using GPU(s) to render? If they know that GPU rendering  works faster when the CPU is not used for rendering, why is that even an option?