Render time comparison - maya geo vs. alembic vs. vrayProxy

[MFlemming]

Hi,

I just did a render time comparison with a geo heavy asset. (VRAY 3.40.01)

1. Maya geo
2. Vray proxy
3. alembic (loaded through vray proxy)

I figured out, that in my case a single vray proxy takes 1.2 to 1.3 times longer than regular maya geometry. Alembic was even longer with about 1.3 to 1.4 times of the original render time. Now, while I know that the primary purpose of vray proxies is to reduce memory consumption, an increase in rendertime by a factor of 1.4 is really an argument against an alembic based pipeline with vray in maya. Switching the default geometry type to static didn't help.

Am I missing something or is this a known fact?

Best regards,
Manuel

[tashko.zashev]

Hi,

Usually the render of proxy geometry is slower and the results you are getting should be normal. Just in case, please provide some more information about your tests. What is the polygon count? Which V-Ray image sampler you are using ?

[MFlemming]

Hi and thanks for the reply,

the poly count in my test series was 9 mio faces and it has been rendered with the adaptive (DMC) sampler. One domelight as light source, brute force GI (as primary bounces, no secondary bounces). All other settings are default (VRAY 3.40.01), except for the dynamic memory limit which I've raised to 40 gigs. I can't provide the geometry for testing.

I was expecting the proxy to take longer, but as I saw the numbers with alembic I really had to think about the alembic workflow. Obviously there are some great benefits with baking everything down into clean alembic caches before handing things over to lighting.. and software like katana is based around that concept for a reason. However I wasn't aware that in vray for maya, simply switching from standard geo to alembic would increase rendertimes that much.

I didn't expand my tests yet to validate those results with multiple scenes.. I just wondered if this is a known thing and what is the best way to deal with it. Loading alembics through vray proxies seems to be the obvious way right now to use vray and maya in a "deferred loading /processing" type of lighting workflow. My test asset is only one part of the original scene.. having all geometry loaded in maya eats up about 24 gigs of my 64 gigs of ram. Starting vray RT will take quite some time and also fill up my ram almost entirely (as long as maya is running at the same time). It's so much better to have the geometry just be evaluated by the renderer, however it's quite a trade-off if this will cause additional rendertime just because of using alembic. Now please don't tell me, that it's only possible with katana or houdini..

Cheers,
Manuel

[tashko.zashev]

Thank you Manuel and excuse me for the delay on my reply.

I don't know if there is any possibility to make the alembic rendering faster or a little bit closer to the proxy mesh render time. So I decided to forward this case to our developers and now waiting for their feedback.

[asen.atanasov]

Hi Manuel,

we may need to do some profiling, but here is what you can try. First, the reading of alembic files from the disk is slower than vrmesh.
You can try to export the vrmesh with "one mesh per voxel" option and see if this is faster. Than you can try to render with "Maya mesh" preview type. This should be identical to Maya geometry.

We will do some testing here as well.

Best regards,
Asen

[MFlemming]

Hi,

thanks for getting back to me. I did some more tests with what you've suggested. Switching geometry to load to "Maya mesh" did produce identical rendertimes as if it was set to "preview". If I export the proxy with "one mesh per voxel" the rendertime gets slower... something inbetween the rendertime of the original proxy and the rendertime of the alembic. So the order is still this (from fast to slow):

1. standard maya geometry
2. vray proxy (no difference whether it's set to preview or maya mesh)
3. vray proxy (exported with "one mesh per voxel")
4. alembic (no difference whether it's set to preview or maya mesh)

Would be interesting to know, if you can validate this in your tests.

Best regards,
Manuel

[Svetlozar Draganov]

Hi,

I just did a render time comparison with a geo heavy asset. (VRAY 3.40.01)

1. Maya geo
2. Vray proxy
3. alembic (loaded through vray proxy)

I figured out, that in my case a single vray proxy takes 1.2 to 1.3 times longer than regular maya geometry. Alembic was even longer with about 1.3 to 1.4 times of the original render time. Now, while I know that the primary purpose of vray proxies is to reduce memory consumption, an increase in rendertime by a factor of 1.4 is really an argument against an alembic based pipeline with vray in maya. Switching the default geometry type to static didn't help.

Am I missing something or is this a known fact?

Best regards,
Manuel

As far as I know this is a known fact. The main goal of V-Ray Proxies is to save memory, they are dynamically loaded/unloaded during the rendering and these operation requires additional rendertime.
Original geometry will be always rendered faster than Proxies and Alembics.

[asen.atanasov]

Yes, this is a known fact. Maya geometry is fastest, because the geometry is already in memory. Proxies need additional disk operations during render time, to load the geometry on demand. Alembic is open source library and it is slower than our vrmeshes. Still, we will make some test and see if something can be done.

Best regards,
Asen

[MFlemming]

ok, good to know that it's a known fact. Thanks for investigating the topic!

Cheers,
Manuel