Learning Forum

[Amir]

Hi everyone,

For postprocessing and making animation in Fluent, I wonder if somebody can help with this questions and if possible shares some benchmark and information

PART A: GPU

1)Which hardware is the key for such a case that includes more than 12000 frames for each animation,

checking the performance, I see on one workstation with Quadro GPU only, up to %20 CPU and %60 GPU are in use, although it has an old K2000 GPU!

In another computer with a newer GPU but not Quadro (GTX), the GPU not in use at all!

2)Does this process (animation making) need a high single-precision floating-point (FP32) performance or a double-precision (FP64) is applied?

3)if FP64 matters, should my GPU also be in the supported GPU list? even for post-processing not for solving in parallel? as my GTX GPU FP64 performance is better than the FP64 performance of the old Quadro

PART B: SSD

4)About the performance contribution of SSD, is the random access (read/write) important or the sequential one? I hardly can make 3 animations per 8 hours! Although I am using a really fast workstation with NVME SSD in one case and for another one with raid 0 SSDs, 

trying to understand what is the reason, the possible bottleneck and how can we speed up the process.

Regards,

Amir

[Rob]

12,000 frames may be a little excessive and is likely to require a fair bit of RAM. The cpu solver (Fluent) generally handles the processing on cpu, and only uses gpu for images, not processing. The GPU solver will use the gpu for solving, and not all post processing features (as per the cpu solver) are available. 

SSD is good where you're reading/writing a lot and also for laptops: we've broken far fewer drives since switching. 

[Amir]

I don't have a RAM issue, well let me paraphrase the questions maybe better

1) For the process of image-to-animation generation(PPMs to a video conversion), can a GPU assist and speed up the process?

2) If the GPU can help, which element of the spec of the GPU is more important (if the VRAM is enough)

CUDA/pipeline? memory bandwidth?....

3) FP32 or FP64 which one is important? the nature of this conversion is dealing with double floating-point information or single floating point?

4) The GPU for such post-processing should be in the supported GPUs for the GPU solver.

5) As the ppm files are small but in many for making the animation, for an SSD (NVME or SATA) the processing speed is affected by the random access or sequential access performance of the SSD?

 

[Rob]

GPU won't do much, if anything for post processing on the cpu solver: it's an offload approach for certain models. 

Regarding the rest, I suspect you're better off asking your IT, how busy is the disc when you're generating the movies?