Prepasses only using half my cores at most

[cgbeige]

I have been unable to get prepasses to use all 16-threads of my 8-core hyperthreaded Nehalem Mac Pro. Setting the thread count manually to sixteen doesn't help. The final render pass is always using all 16 threads but none of the prepasses. Has anyone got a fix for this? The screenshot below shows it using 440% of a potential 1600% (Photoshop CS5 beta is eating 100% though.)

[cgbeige]

ok - now I'm getting full usage on prepasses suddenly. Does it only kick in when images are above a certain size?

[Morbid Angel]

ok - now I'm getting full usage on prepasses suddenly. Does it only kick in when images are above a certain size?

did you put 16 in light cache pases instead of 8?

[cgbeige]

sorry - I'm a VRay newb. Does that affect how the job is divided among the CPU cores? Anyway, I'm not using light cache for primary or secondary bounces.

[Morbid Angel]

well typically lightcache passes determine the number of threads for lc processing. But if you are not using any...could be something else...

[cgbeige]

well now that I'm getting closer to final renders and higher res, it seems that the prepasses are using all the cores. Wish I knew how to avoid this though - is it related to bucket size and images?

[cgbeige]

spoke too soon - the final render pass is not fully using 16 threads and then all of a sudden it is. I'm new to VRay - is this just how it works? I'm used to renderers constantly saturating all cores unless it's doing something like a post glow pass that's only single threaded or something.

[vlado]

If you are using things like render time subdivision surfaces, or displacement, or VRayProxy objects, then this is normal. While the raytracing of such geometry is fully multithreaded, the generation of this geometry is not (for example, VRayProxy objects need to be loaded from disk; this can happen anywhere during the rendering and obviously cannot be multithreaded; once the geometry is loaded though, rendering continues at full speed).

There is also a limited amount of RAM that V-Ray uses for such render-time generated geometry; it is controlled through the "Dynamic memory limit" option in the Settings tab. If this value is too low, it will cause V-Ray to constantly generate and destroy geometry, which will slow down the rendering.

Best regards,
Vlado

[cgbeige]

ok - thanks for the elaborate reply. I have a few subdivision surfaces

[cgbeige]

Ok - this is driving me crazy again and it's wasting a ton of time. Can someone help me with the cpu usage in VRay - it's brutal for this scene. I don't know whether it's because it's using instances but it uses 1/4 of my total cores for 9/10 of the render and then spikes to half at the end (see linked movie below). It's pretty lame so I'm wondering if someone can look at my settings at the beginning and see if it's something there. It's basically the default setting with EXR 32-bit, some render passes and DMC sampling.

http://www.can-con.ca/cgtalk/vrayrender.mov

That number should always be at 1600% ideally (OS X measures it as 100% per core).

The same scene rendering in Maxwell saturates all cores:



I don't think instances should be completely killing the core usage.

[vladimir.nedev]

Try what Vlado said, go to Render Settings -> Settings -> System -> Raycaster params and increase the "Dynamic memory limit" from 500 MBs to around two thirds of your installed memory.

Greetings,
Vladimir Nedev

[cgbeige]

I did that - nothing changes.

[cgbeige]

ok - this is a serious problem. I have been watching this last part of the render for 15 minutes and it's been using ONE of SIXTEEN threads for the whole time. It's brutally inefficient and I think it should be considered a bug since it slows the renders down so badly.

[paulDryzen]

Hi CGbeige,

this is normal because your bucket size is pretty big. Try settings like 16*16 for this image.
One bucket is allways rendered bei only one core.

Greetings Paul

[snivlem]

Perhaps you are you oversampling too? That looks like it should render pretty fast.

Even on my old AMD X2 3800 I was able to render this scene out in under 3 minutes:

Refractions/HDR Reflections/Glossy maps/layered textures/bump maps you name it.