A question about Iray and Hardware

hominicchiolo_ba2ad446f6

Hello. I am writing to solve a problem and some doubts about iray rendering. the hardware consists of AMD Rysen 9 3900x processor, 2x GPU nvidia 1080 8gb, Ram 32gb.
In practice ... some rendering of heavy scenes (Ram PC occupied 20Gb max, 750MiB / 1.700GiB geometry mem consumption + Texture etc ..) start processed by the CPU regardless of the choices made in render settings / advanced.
First question.
I seem to have understood that once the Ram capacity of the video cards is exceeded, the CPU takes over. Quite right?
Second question.
Why, and this also occurs when the rendering starts fast with video cards, Task Manager indicates the GPU at 0/2%. This thing puzzles me.
Third question.
SLI mode not good for Iray?
I hope the questions are not too confusing. I don't have much technical knowledge ... :)

JamesJAB

1. If the scene does not fit completely into the VRAM of a GPU, That device will drop from the render job. (Each GPU will work as a self contained unit during rendering)

2. The task manager does not show Cuda based work loads by default. (iray uses Cuda)

3. SLI is not used for Iray rendering. (it is a method of syncing 2 cards so they can share the load in real time 3D graphics.)

 

 

hominicchiolo_ba2ad446f6

I am very grateful to you. I think I understand correctly
I just wonder why I didn't ask this question before

dawnblade

Deleted. Not relevant to SLI.

 

hominicchiolo_ba2ad446f6

OK thanks. I didn't know NVLINK. I understand that it is a technology for the generation of 20 series cards and that in any case it is not yet optimal, as you wrote. You cleared my mind, thanks. I understood that at least for the moment I will have to economize my scenes better ... Sic!
Thanks again

 

kenshaw011267

dawnblade said:

Looks like SLI mode is possible in Iray, according to this Iray dev post, but it has to be enabled in Daz Studio by the DS devs.

Also, there is a limitation that currently allows only certain types of scene information to be shared:

"Devices in an NVLINK peer group can efficiently share device memory. Iray shares scene memory among all devices in the peer group. This allows Iray to render scenes exceeding the device memory of a single device. Note that devices in the peer group will not share all their scene and rendering related memory (at the moment only bitmap/texture data is shared)."

 

NVLink isn't SLI. Pascal and earlier cards, like the 1080, do not have the NVLink connector but the older SLI one.

On Volta and Turing cards that have a connector only some features of NVLink work on consumer cards. DS is apparently trying to implement VRAM pooling for 4.13 but that is likely a ways off. For DS/iRay the vast majority of VRAM is occupied by textures so it should work out to nearly the same thing assuming you don't go crazy with subD which multiplioes the complexity of the geometries.

Since it is likely going to be another few months at least until 4.13 rolls out I would hold off making any purchasing decision involving multi GPU's for iRay until it launches.

fastbike1

I would suggest not using the term SLI, since Iray memory pooling is limited to NVlink and either the 2080TI, 2080Super, 2080, and 2070 Super.

dawnblade said:

Looks like SLI mode is possible in Iray, according to this Iray dev post, but it has to be enabled in Daz Studio by the DS devs.

Also, there is a limitation that currently allows only certain types of scene information to be shared:

"Devices in an NVLINK peer group can efficiently share device memory. Iray shares scene memory among all devices in the peer group. This allows Iray to render scenes exceeding the device memory of a single device. Note that devices in the peer group will not share all their scene and rendering related memory (at the moment only bitmap/texture data is shared)."

 

 

dawnblade

kenshaw011267 said:

dawnblade said:

Looks like SLI mode is possible in Iray, according to this Iray dev post, but it has to be enabled in Daz Studio by the DS devs.

Also, there is a limitation that currently allows only certain types of scene information to be shared:

"Devices in an NVLINK peer group can efficiently share device memory. Iray shares scene memory among all devices in the peer group. This allows Iray to render scenes exceeding the device memory of a single device. Note that devices in the peer group will not share all their scene and rendering related memory (at the moment only bitmap/texture data is shared)."

 

NVLink isn't SLI. Pascal and earlier cards, like the 1080, do not have the NVLink connector but the older SLI one.

On Volta and Turing cards that have a connector only some features of NVLink work on consumer cards. DS is apparently trying to implement VRAM pooling for 4.13 but that is likely a ways off. For DS/iRay the vast majority of VRAM is occupied by textures so it should work out to nearly the same thing assuming you don't go crazy with subD which multiplioes the complexity of the geometries.

Since it is likely going to be another few months at least until 4.13 rolls out I would hold off making any purchasing decision involving multi GPU's for iRay until it launches.

 

fastbike1 said:

I would suggest not using the term SLI, since Iray memory pooling is limited to NVlink and either the 2080TI, 2080Super, 2080, and 2070 Super.

dawnblade said:

Looks like SLI mode is possible in Iray, according to this Iray dev post, but it has to be enabled in Daz Studio by the DS devs.

Also, there is a limitation that currently allows only certain types of scene information to be shared:

"Devices in an NVLINK peer group can efficiently share device memory. Iray shares scene memory among all devices in the peer group. This allows Iray to render scenes exceeding the device memory of a single device. Note that devices in the peer group will not share all their scene and rendering related memory (at the moment only bitmap/texture data is shared)."

 

 

Oops! When I saw the descriptions of NVLink products, they included "SLI" which is why I thought it pertained to SLI. I now see that my reply has no relevance to this thread, so I'll delete it. Sorry! Thanks for letting me know.

 

kenshaw011267

Let me make clear that people should be wary of rushing out to get second cards or 2 cards to use VRAM pooling. I'm unaware of anyone getting it working in the beta. Also NVLink requires matched cards. For SLI it was even finicky about using 2 cards of the same model but different brands (sometimes it worked and sometimes it didn't).