[meen_far]

HI,
I am facing a pertinent problem while using Ansys 15.07 in a Dell Precision T7610 Workstation having 64GB RAM and 1 NVDIA Tesla K20 card for GPU. The following error message is is being displayed every time for running GPU based solver be it Fluent or Structural solver.
AMG on GPGPU
NVAMG version 4
Built on Aug 21 2013, 13:02:58
Thrust failure: bad allocation: out of memory
File and line number are not available for this exception.
NVAMG ERROR: file ..\..\src\amg_gpu.c line 905
NVAMG ERROR: Thrust failure

Error: There is no active application.
Error Object: (case-modified?)
MPI Application rank 0 exited before MPI_Finalize() with status 1

Error: There is no active application.
Error Object: (data-valid?)


Can anybody comment in this context?

Thnks in advance.

[flotus1]

Does the simulation run without this error on the CPU only?

[meen_far]

Originally Posted by flotus1

Does the simulation run without this error on the CPU only?

yes the simulation run if i uncheck the GPU setting, it run without this error on CPU only

but i did the simulation before with GPU checked (same pc, same specs, few days back actually i run the simulation on both GPU and CPU, it was running fine, without any error watsoever. thats y i dont understand why for a specific model, when i check the GPU option, i get the error message. i create a new geometry, new design to test, and the simulation doesnt run if enable GPU and CPU...

[flotus1]

Are you sure that it was the EXACT same simulation? Same number of cells, same models activated?
I am asking since the error message suggests that the model requires more memory than the GPU can provide.

Just out of curiosity: why are you using the GPU?
The vast majority of benchmarks I have seen so far suggest that running in parallel with the GPU actually increases simulation times for most models.

[meen_far]

Originally Posted by flotus1

Are you sure that it was the EXACT same simulation? Same number of cells, same models activated?
I am asking since the error message suggests that the model requires more memory than the GPU can provide.

Just out of curiosity: why are you using the GPU?
The vast majority of benchmarks I have seen so far suggest that running in parallel with the GPU actually increases simulation times for most models.

no no, the simulation had different setting, more cells, more nodes, the model was slightly more complicated than the previous one. But now it makes sense. i didnt understand the exact meaning of the error message. hopefull i did manage runing the sim without the GPU on.

Oh interesting. i read somewhere that if combining gpu and cpu for simulation, the simulation will take less time. thats y i was using the gpu as well. but now that u pointed it that its the other way around, i wont be bothering using the gpu for future. Thanks a lot for clarifying. You guys rock (Y)

[nikoshiya]

Originally Posted by meen_far

HI,
I am facing a pertinent problem while using Ansys 15.07 in a Dell Precision T7610 Workstation having 64GB RAM and 1 NVDIA Tesla K20 card for GPU. The following error message is is being displayed every time for running GPU based solver be it Fluent or Structural solver.
AMG on GPGPU
NVAMG version 4
Built on Aug 21 2013, 13:02:58
Thrust failure: bad allocation: out of memory
File and line number are not available for this exception.
NVAMG ERROR: file ..\..\src\amg_gpu.c line 905
NVAMG ERROR: Thrust failure

Error: There is no active application.
Error Object: (case-modified?)
MPI Application rank 0 exited before MPI_Finalize() with status 1

Error: There is no active application.
Error Object: (data-valid?)


Can anybody comment in this context?

Thnks in advance.

I take the following answer from Dell Precision T7610 Manual, if you face any issue in implementing these steps. Then check manual for detailed view.
Run Registry Editor (Regedt32.exe).
From the HKEY_LOCAL_MACHINE subtree, go to the following key: \System\CurrentControlSet\Control\Session Manager\SubSystems.
Select the Windows value.
From the Edit menu, choose String.
Increase the SharedSection parameter.

[hellsblade.91]

i am also facing the same problem the simulation runs when i dont use the double precision option but with double precision i get this error


I increased the no of nodes and elements ie made the mesh much finer

[flotus1]

Are you asking for advice or just stating a fact?
Your model does not fit into GPU memory, so it can nut be run using GPU acceleration.

[hellsblade.91]

Hi I am seeking advice do you know the solution is it due to increased no of elements due to mesh being finer

[flotus1]

Possible solutions:

• run smaller models that fit into GPU memory
• decrease memory usage by other means, e.g. by using single precision instead of double
• buy a GPU with larger memory (or several of them)
• not using GPU acceleration at all

There are no workarounds for this issue. If the model does not fit into GPU memory, you can not run it with GPU acceleration.

[hellsblade.91]

How do i run it with a GPU Acceleration do you know ?