display driver stopped responding and has recovered - error

[DiuS]

Sometimes when I run the simulation, I see this error. And simulation hangs. Although the storage or memory (RAM) is still free to approximately 10% percent.
Sometimes an error occurs on frame 1, and sometimes in the middle of the animation. Download latest Nvidia driver but it did not help. System Windows 8. Are there any solutions?
This error occurs in different scenes.

[Ivaylo Katev]

i suppose the simulation hangs in the preview part, try to disable all the preview channels.

[Svetlin.Nikolov]

Do you have a GPU preview enabled? Sometimes we get this problem but were unable to properly reproduce it and fix it, suppose it happens on specific GPUs. What are your specs?

[DiuS]

Yes, I divide the viewport into 2 parts. Left to PhoenixFD GPU Preview, and on the right side Perspective projection.
I understood I would try to use only the perspective projection for the simulations.

And one more question. Can you do like Grid Gizmo to conserve RAM memory in a future release?
Or adaptive simulation. Something like this picture http://rendermansite.pixar.com/ca_tw...42_2766_me.gif but only for simulation.
I'll try to explain. Comes with 2 calculation: the first rough in low resolution. PhoenixFD analyzes in which the cells of the grid will be the simulation and in what will not.
Next comes the second step of simulation. In which cell without simulation "empty" are combined / are not divided / or deleted. And cells in which there was a simulation increases the resolution.
You can also to make an "empty" cell is recorded on HDD to release the RAM. They will always have a value of "0", ie empty. Therefore it is not necessary that they would to waste RAM.

Or you can add the ability to build not only a rectangular grid. A more complex form. For example unites grid in the pyramid. Thus it will be possible to calculate the complex forms. For example rocket launch. Where need a broad base (very wide grid) on the ground and a very narrow and high grid in the air.

Maybe you can to make so that it became possible to combine the grid using the boolean



Or write what methods you use that would save RAM without reducing the resolution of the grid.

Sorry for bad english and thanks for your attention. I hope you understand my idea.

[DiuS]

Forgot to add I am working on:
Nvidia GTX 660 2gb
16 gb RAM
Intel Xeon 1230 v2
Windows 8
Maybe it will help you.

Also I found in the internet that this error occurs in other people who do not use PhoenixFD. Ie perhaps this is a bug drivers Nvidia + new Windows. But the solution is not found.

[Ivaylo Katev]

your suggestion will not lead to optimization, because one very important property of the grid structure - having the coordinates, we can easily calculate where in the memory the cell is placed, and the most important - where the neighbors are placed.
if we exclude the empty cells from the simulation and ram, we lose this very very important advantage, and we have to keep some kind of address book that gives us the connection coordinates-memory.
in general, this address book is bigger than the ram that you spare. and this is not all. actually the "empty" space is not empty, it contains air that is moving. the movement of this air contributes significantly to the shape of the visible areas. that's why the adaptive grid gives worse result than the static one.

[Svetlin.Nikolov]

Yes, the only possible optimization will be a post-process like in FumeFX, where they optimize the already done caches to save HDD space and eventually render time. But if you want this to happen during simulation, it will be very hard to do.

[DiuS]

your suggestion will not lead to optimization, because one very important property of the grid structure - having the coordinates, we can easily calculate where in the memory the cell is placed, and the most important - where the neighbors are placed.
if we exclude the empty cells from the simulation and ram, we lose this very very important advantage, and we have to keep some kind of address book that gives us the connection coordinates-memory.
in general, this address book is bigger than the ram that you spare. and this is not all. actually the "empty" space is not empty, it contains air that is moving. the movement of this air contributes significantly to the shape of the visible areas. that's why the adaptive grid gives worse result than the static one.

Understand, and can I combine 2 or more grid in a pyramid? Or coordinate system is also broken?

[Ivaylo Katev]

we tried that at the very beginning, in general it works, but there are visible artifacts on the contact between the grids

[Svetlin.Nikolov]

In your case, how much RAM does the simulation use?

In general, to save memory you should get rid of any channels in the Output panel that you don't use, and also you should be very careful if you use adaptive grid and the 'Preallocate' option, because it will reserve memory that the simulator will not actually use.

[DiuS]

In your case, how much RAM does the simulation use?

In general, to save memory you should get rid of any channels in the Output panel that you don't use, and also you should be very careful if you use adaptive grid and the 'Preallocate' option, because it will reserve memory that the simulator will not actually use.

Sometimes spends 12GB memory. Then the computer starts using HDD and simulation is very retarded.
I do not understand the meaning Preallocate memory
How it helps to save memory? If I turn on this option then it consumes more memory.
Or memory savings can be observed at the end of the simulation? When adaptive grids grow strongly? Please explain to me more.

[Svetlin.Nikolov]

You should just be careful not to use preallocate by accident, because some people do that Check the help file for more info http://docs.chaosgroup.com/display/P220N/Grid.

Otherwise - keep unused grid and particle channels disabled. Currently that's all that can be done unfortunately.