Re: OctaneRender® 2026.4 for 3ds max® v26.11
Posted: Tue Sep 29, 2026 9:48 am
Hi coilbook,
The GPU that is connected to the monitor also has to handle the monitor itself, 3D applications such as 3ds Max, etc., as well as the operating system. As mentioned several times, the VRAM of multiple GPUs is not added together. Instead, the amount of VRAM available is determined by the GPU with the lowest amount of VRAM. That said, I had recommended putting the 3080s together with the 2080s. The render nodes’ available VRAM is not monitored by the main render PC. Therefore, if a scene cannot be handled with the amount of VRAM available on the render nodes, the node will crash. Out of Core allows the scene to be rendered anyway, but if, for whatever reason, the amount of memory available through Out of Core is not sufficient, the render node will still crash. A specific case that could apply to your situation, for example, would be a very large piece of geometry that fits within the 12 GB of VRAM available on the main computer but does not fit within the 11 GB available on the render node. If, for example, this piece of geometry requires 5 GB but exceeds the available VRAM by only 1 GB, all 5 GB would be moved to Out of Core, reducing the available Out of Core memory by 5 GB. As a result, the node could crash if its Out of Core memory was already full. Even if the difference were only 1 GB with the same amount of Out of Core memory available, that difference could still contribute to the problem. Furthermore, this memory movement could also involve textures stored in VRAM or in Out of Core memory. Therefore, assuming that a 1 GB difference is insignificant would not be correct. As I mentioned, my recommendation is to make full use of the VRAM available on the RTX 5090 GPU by keeping it alone in a computer. Then, I would keep the 3080s and 2080s on one or two separate computers. On those systems, I would set Out of Core to 21 GB plus the amount of Out of Core configured on the main computer with the 5090. In any case, one test you can perform right now is to set the Headroom on the main computer with the 5090 and 3080s to 1.3 GB, and thhe headroom to 0,3gb with 2080s and then set Out of Core to 16 GB on both the main computer and the render node. This way, you should have exactly the same hardware configuration and the same amount of Out of Core memory available on both computers, allowing you to verify what happens.
Regards
Paride
The GPU that is connected to the monitor also has to handle the monitor itself, 3D applications such as 3ds Max, etc., as well as the operating system. As mentioned several times, the VRAM of multiple GPUs is not added together. Instead, the amount of VRAM available is determined by the GPU with the lowest amount of VRAM. That said, I had recommended putting the 3080s together with the 2080s. The render nodes’ available VRAM is not monitored by the main render PC. Therefore, if a scene cannot be handled with the amount of VRAM available on the render nodes, the node will crash. Out of Core allows the scene to be rendered anyway, but if, for whatever reason, the amount of memory available through Out of Core is not sufficient, the render node will still crash. A specific case that could apply to your situation, for example, would be a very large piece of geometry that fits within the 12 GB of VRAM available on the main computer but does not fit within the 11 GB available on the render node. If, for example, this piece of geometry requires 5 GB but exceeds the available VRAM by only 1 GB, all 5 GB would be moved to Out of Core, reducing the available Out of Core memory by 5 GB. As a result, the node could crash if its Out of Core memory was already full. Even if the difference were only 1 GB with the same amount of Out of Core memory available, that difference could still contribute to the problem. Furthermore, this memory movement could also involve textures stored in VRAM or in Out of Core memory. Therefore, assuming that a 1 GB difference is insignificant would not be correct. As I mentioned, my recommendation is to make full use of the VRAM available on the RTX 5090 GPU by keeping it alone in a computer. Then, I would keep the 3080s and 2080s on one or two separate computers. On those systems, I would set Out of Core to 21 GB plus the amount of Out of Core configured on the main computer with the 5090. In any case, one test you can perform right now is to set the Headroom on the main computer with the 5090 and 3080s to 1.3 GB, and thhe headroom to 0,3gb with 2080s and then set Out of Core to 16 GB on both the main computer and the render node. This way, you should have exactly the same hardware configuration and the same amount of Out of Core memory available on both computers, allowing you to verify what happens.
Regards
Paride