[Jiuan]

I came across a very strange problem using parallel computing in Fluent. I want to simulate the flow and temperature in a CVD chamber, the substrate is heated by the chamber wall. When I run the case with 6 processors, the temperature in several points on the substrate surface is too high, 500K high than the temperature of chamber wall. Then if I use a single processor, the result seems ok. I cannot figure out the reason. Anybody has the same experience?

Thanks.

[-mAx-]

Just for information, you have to write case and data each time you change your hardware architecture.
I mean if you load the case file from serial, and the data from parallel, you become wrong results.

[Jiuan]

Thanks.
I don't understand your meaning very well.
Each time when I want to check the result, I load the case file and data file together.

Originally Posted by -mAx-

Just for information, you have to write case and data each time you change your hardware architecture.
I mean if you load the case file from serial, and the data from parallel, you become wrong results.

[-mAx-]

Write the case file from the serial solver.
Then go to the parallel architecture, iterates, and write the data file.
Go back to serial, load the case file (from serial), but load the data file from parallel.
You will see what I mean.

[Jiuan]

Thanks.
This time I almost understand.
"Go back to serial, load the case file (from serial), but load the data file from parallel." How to do that? I mean I can load the case from serial, but how can I load the data file from parallel in the same time?

Originally Posted by -mAx-

Write the case file from the serial solver.
Then go to the parallel architecture, iterates, and write the data file.
Go back to serial, load the case file (from serial), but load the data file from parallel.
You will see what I mean.

[-mAx-]

no: load the data file previously saved from parallel

[Jiuan]

Thanks so much.

[behest]

Hi my friends,
I have a strange problem with parallel computing too. Actually, my problem is about writting force convergence history. I want to simulate a 3D wing in transient flow and lift and drag convergence history files are very usefull for me. when I applied the parallel computing, the files will be overwritten and the last results are saved in them; but the saving process is work very well by unsing serial computing.
would you like to make any comment? Have you any suggestion for me?

[BigMc]

Originally Posted by -mAx-

Write the case file from the serial solver.
Then go to the parallel architecture, iterates, and write the data file.
Go back to serial, load the case file (from serial), but load the data file from parallel.
You will see what I mean.

this solution works perfectly for fluent, but when I export the cdat file to CFD Post from Fluent there is discontinuities in the flow fields, has anyone a solution for this?

Thanks