dForce multi GPU ?
Does anybody know a way to make sure DAZ Studio use mutilple GPUs for dForce (of course, without waiting for DAZ to eventually maybe implement it in DAZ Studio in future) ?
Notice, I know dForce (simulation) is not Iray (rendering), and uses OpenCL (Iray, coming from NVIDIA, uses directly CUDA) to talk to the CUDA driver, but as far as I can see, OpenCL in itself does support the possibility of spreading the computational work across multiple CPUs and/or GPUs.
I activated the NVLink, but I am not sure it helps (in theory not, because apparently for rendering it is automatically used, for simulation computing it is the software that must spread the workload between the GPUs).
Or can someone suggest another physics engine that I can use with DAZ Studio integrated as simulation engine (I noticed dForce is the only one that I can select, but in the "Advanced" tab of the simulation is in theory possible to select another one), that might be able to leverage the two GPUs ?
Just to be clear, the goal is to use it for simulating movement of dForce body parts and eventually clothes.
Or at least, does anybody know if in doing the simulations, dForce uses single precision FP16 (I assume it uses floating numbers, that would be normal for physics simulations) or double precision FP32 ? That can be helpful anyway, because I have a card supposedly older but that has bigger FP32 performances, so I could try using that one when I want to do simulations, and the other one for rendering.
Thanks,

Comments
dForce uses only one GPU, I am not aware of any way to change that.
While the framework is, apparently, there for adding other engines I am not aware of anything that has been integrated in that way - nor do I know whether the necessary details have been published.
Thanks, well, I think the problem is one of programming complexity not being worth it for DAZ, because from what I see in fora around, most people have only one graphic card they use for everything - OpenCL does allow to share the workload between multiple GPUs and CPUs, even combining CPUs and GPUs, but the division in threads is not managed in a transparent way by the libraries (there is one such library, but was something created as part of a university research and I think only on linux), it is the developer that has do it by hand.
I don't use only one card in my workstation, and actually it is great for performance, since Iray uses all the cards available (and yes, I can see it really spread the work, depending on the complexity of the rendering), and it allows to have one card for visualisation and computation/rendering while others do only rendering or computation, so, despite not being latest generation (more expensive) cards, I get better performances overall that if I had one single newer card.
About this, one thing that I noticed in my simulation for dForce, is that an older card (NVIDIA Volta generation) and classified as Geforce (consumer) has much better performance in dForce than a card two generations after and classified as professional (Quadro Ampére) in doing dForce simulations - though the card in question is a TitanV, it was a Geforce, but a bit special (first card with Tensorcores, and it FP64 peformances leave in the dust all the cards of the generations affter, even the Quadro/PRO, not just the Geforce).. So, newer is always better is not forcibly always true.
Pity that by end of next year probably NVIDIA new NVIDIA drivers will stop supporing Volta, though since for dForce Daz uses OpenCL I could simply stil continue to use the latest compatible driver, and anyway I plan to stay on Daz 4.x for a veeeery long time (since I am sure I wil anyway continue to buy daz content, if I do not ask them to support me, DAZ should not try to force or "force" me to change version).