[LuckyTran]

Originally Posted by kimireggie

Thanks for your quick reply.

Did you mean that once GPGPU acceleration is enabled, Fluent always adopts coupled scheme for P-V-Coupling, even if SIMPLE/SIMPLEC has been set in the case file? (Assumed no more parameters regarding GPGPU is set in command line or journal file except '-gpgpu=1')

No, sorry if I was unclear or mislead anyone. If GPU acceleration is enabled and simple scheme is selected, then nothing happens on the GPU and it proceeds like a regular calculation. The GPU is used only if the Coupled scheme is being used. However, with additional settings, you can use GPU with simple scheme.

[Jonathan]

Hi Guys,

Sorry for the slow reply. To clarify my understanding (based on what was said on the ANSYS webinars), actually you can use SIMPLE with the GPU, and the calculations WILL take place on the GPU.

However, the issue is:

1) GPU are designed for computing LARGE matrices quickly, so dealing with (solving) the large pressure + velocity coupled matrix will give the best speedups.

2) The major bottle neck with GPU computing at this stage, is the speed at which information gets onto and off of the GPU. i.e. how fast data gets through the PCI-Express bus into the GPU from the CPU. If you solve in a segregated fashion, i.e. using SIMPLE, the time overhead of getting info on and off of the GPU overcomes the time saving from the matrix solution.

So its better to move a big lot of data onto the GPU, solve and move it off with GPU computing, rather than move lots of smaller packets of data on and off the gpu (i.e. x3 times min for SIMPLE vs x1 for COUPLED).

This is my understanding,
hope it helps,
cheers
jonathan

Originally Posted by LuckyTran

No, sorry if I was unclear or mislead anyone. If GPU acceleration is enabled and simple scheme is selected, then nothing happens on the GPU and it proceeds like a regular calculation. The GPU is used only if the Coupled scheme is being used. However, with additional settings, you can use GPU with simple scheme.

[kimireggie]

Originally Posted by LuckyTran

No, sorry if I was unclear or mislead anyone. If GPU acceleration is enabled and simple scheme is selected, then nothing happens on the GPU and it proceeds like a regular calculation. The GPU is used only if the Coupled scheme is being used. However, with additional settings, you can use GPU with simple scheme.

Thanks & got it, though it's a little bit different from Jonathan's point

[kimireggie]

Originally Posted by Jonathan

So its better to move a big lot of data onto the GPU, solve and move it off with GPU computing, rather than move lots of smaller packets of data on and off the gpu (i.e. x3 times min for SIMPLE vs x1 for COUPLED).

Hi Jonathan,

During my experiment using GPU calculation with SIMPLE/COUPLED scheme in Fluent, once '-gpgpu=1' is called, the fluent process can be found on GPU card (by monitor command provided by nvidia). This seems to be agreed with your point.

On the other hand, speedup gained by gpgpu with COUPLED scheme compared to that with SIMPLE scheme goes the other way. Five cases for test are prepared, and here is what I tested with/without gpu for each case:
a) use SIMPLE scheme
b) use COUPLED scheme

The results? Solving time with GPU is almost equivalent to that without GPU under scenario a), i.e with SIMPLE scheme (which is the case as you described). But with COUPLED scheme, time with GPU increases in an unexpected way which is fairly longer than solving time with CPU only (which is the case as you mentioned at the very start of this post).

Is there anything inappropriate in my test procedure? Have you found way to solve this? Thanks a lot.

[HyperNova]

Hi Experts, i am going to model a propeller in Fluent, half of it is under water and half of it in air so the problem is multiphase. i am going to solve the problem using GPU to show how GPU can accelerate solving. The workstation i am going to use has 4*Xeon E5-4669 v3 + Nvidia Tesla K80. i am asking is it possible to model such a problem using GPU or not ? and advice will be appreciated.