Had enough of CFD Post - can Tecplot parallel compute over many cores (CPU and GPU)?

JuPa · January 19, 2018, 04:44

Hi,

We do a lot of CFD (with Ansys) in our organisation. Frankly CFD Post doesn't cut it.

We need the ability to make high quality animations of our results and we need to be able to do it over many cores (i.e. quickly).

Can Tecplot does this?

wsfowler · January 23, 2018, 19:03

Tecplot 360 can certainly make nice animations, and much of our processing is shared-memory parallel. Meaning that we'll use all the cores on a single machine.

Tecplot 360 cannot natively do distributed memory parallel. For time-dependent solutions you may be able to launch Tecplot 360 on different nodes of a cluster with each Tecplot 360 instance creating an image for a different time step. The instructions for which time step to write would be in a macro or Python script (see PyTecplot).

That said - you may run into disk I/O being a bottleneck. Post-processing is very "disk read" dependent - once you saturate your disk you're stuck. I've heard reports that distributed animations typically work up to about 4 nodes - anything beyond that and you become I/O bound so about the best you'll do is 4x faster. It's of course your gains will be very hardware and problem dependent.

loesung · June 25, 2022, 12:47

can we overcome that bottleneck by simply using, say 10 nvme drives?

wsfowler · June 26, 2022, 22:59

Parallel file systems certain help with this bottleneck. The scaling is also dependent on the distribution of the post-processing work. Some post-processing tasks are more I/O bound (contour plots and slices - these require mostly disk I/O and quite little CPU time) - but others (like computing Q-Criterion and plotting complex iso-surfaces) may be CPU bound. It's possible to become GPU bound as well for very complex plots, but I've found this is infrequent to be the bottleneck.

So the bottleneck can be problem dependent.

January 19, 2018, 04:44	Had enough of CFD Post - can Tecplot parallel compute over many cores (CPU and GPU)?	#1
JuPa Senior Member Mr CFD Join Date: Jun 2012 Location: Britain Posts: 361 Rep Power: 14	Hi, We do a lot of CFD (with Ansys) in our organisation. Frankly CFD Post doesn't cut it. We need the ability to make high quality animations of our results and we need to be able to do it over many cores (i.e. quickly). Can Tecplot does this?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GPU acceleration in Ansys Fluent	flotus1	Hardware	63	May 12, 2023 02:48
AMD Epyc CFD benchmarks with Ansys Fluent	flotus1	Hardware	55	November 12, 2018 05:33
[Resolved] GPU on Fluent	Daveo643	FLUENT	4	March 7, 2018 08:02
Computing platform for OpenFOAM(Acceleration with GPU vs. CPU)	cagri.metin.ege	OpenFOAM	2	August 17, 2016 02:04
Urgent help in using all cpu cores running Fluent in dual 4669v3 CPU	HyperNova	Hardware	2	July 19, 2016 20:16

January 23, 2018, 19:03		#2
wsfowler Senior Member Scott Fowler Join Date: May 2009 Posts: 112 Rep Power: 16	Tecplot 360 can certainly make nice animations, and much of our processing is shared-memory parallel. Meaning that we'll use all the cores on a single machine. Tecplot 360 cannot natively do distributed memory parallel. For time-dependent solutions you may be able to launch Tecplot 360 on different nodes of a cluster with each Tecplot 360 instance creating an image for a different time step. The instructions for which time step to write would be in a macro or Python script (see PyTecplot). That said - you may run into disk I/O being a bottleneck. Post-processing is very "disk read" dependent - once you saturate your disk you're stuck. I've heard reports that distributed animations typically work up to about 4 nodes - anything beyond that and you become I/O bound so about the best you'll do is 4x faster. It's of course your gains will be very hardware and problem dependent.

June 25, 2022, 12:47		#3
loesung New Member Join Date: Mar 2022 Posts: 18 Rep Power: 4	can we overcome that bottleneck by simply using, say 10 nvme drives?

June 26, 2022, 22:59		#4
wsfowler Senior Member Scott Fowler Join Date: May 2009 Posts: 112 Rep Power: 16	Parallel file systems certain help with this bottleneck. The scaling is also dependent on the distribution of the post-processing work. Some post-processing tasks are more I/O bound (contour plots and slices - these require mostly disk I/O and quite little CPU time) - but others (like computing Q-Criterion and plotting complex iso-surfaces) may be CPU bound. It's possible to become GPU bound as well for very complex plots, but I've found this is infrequent to be the bottleneck. So the bottleneck can be problem dependent.