GPU simulation Fluent out of memory

xab · January 10, 2017, 10:01

Hi all,
I am trying to perform my first simulations with Fluent 16 with gpu acceleration and I'm facing some problems. My cpu has 8Gb Ram and the gpu is Quadro k600 with 4Gb.

My problem is that with small problems (200.000 cell) the simulation is slower with graphic card due to the partition, which is normal, but when trying bigger simulations (>500.000 cell) my graphic card runs out of memory however I can perform the simulation without the gpu acceleration.

Is this normal? That means that it isn't worthy to make gpu acceleration with this kind of cards?

Any advice or comment would be well appreciated
Best regards
Xabi

flotus1 · January 11, 2017, 11:43

The results you get are not surprising. Ansys recommends high-end compute cards for GPU acceleration. High amounts of memory and DP performance are keys to success here. A Quadro K600 lacks both factors. I doubt that you have a version with 4GB of VRAM, to my knowledge such a video card does not exist.

Here is presentation that helps estimating the memory requirements for GPU acceleration: https://www.nvidia.cn/content/tesla/...-userguide.pdf

My advice: stick to CPU-only without GPU "acceleration".
I would not subscribe your statement "the simulation is slower with graphic card due to the partition". It is probably slower because your GPU is rather low-end and GPU acceleration is still a feature under development.

xab · January 12, 2017, 05:21

Thank you very much for your quick reply, I didn't know that this kind of simulation was so much consuming. After reading the link you posted, I will stick to cpu calculation.

Thank you again for your help

January 10, 2017, 10:01	GPU simulation Fluent out of memory	#1
xab New Member xabi Join Date: Apr 2015 Posts: 13 Rep Power: 11	Hi all, I am trying to perform my first simulations with Fluent 16 with gpu acceleration and I'm facing some problems. My cpu has 8Gb Ram and the gpu is Quadro k600 with 4Gb. My problem is that with small problems (200.000 cell) the simulation is slower with graphic card due to the partition, which is normal, but when trying bigger simulations (>500.000 cell) my graphic card runs out of memory however I can perform the simulation without the gpu acceleration. Is this normal? That means that it isn't worthy to make gpu acceleration with this kind of cards? Any advice or comment would be well appreciated Best regards Xabi

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Some questions about flow boiling simulation in Fluent	beastieboys6	FLUENT	8	November 20, 2017 23:47
fluent: Received signal SIGIOT	puneet336	FLUENT	1	May 12, 2016 02:51
Lenovo C30 memory configuration and discussions with Lenovo	matthewe	Hardware	3	October 17, 2013 10:23
Simulation on fluent 14	NAD	FLUENT	0	September 11, 2013 02:05
LSR (Land speed record) car simulation on FLUENT	Maxime31850	FLUENT	2	May 1, 2013 11:15

January 11, 2017, 11:43		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	The results you get are not surprising. Ansys recommends high-end compute cards for GPU acceleration. High amounts of memory and DP performance are keys to success here. A Quadro K600 lacks both factors. I doubt that you have a version with 4GB of VRAM, to my knowledge such a video card does not exist. Here is presentation that helps estimating the memory requirements for GPU acceleration: https://www.nvidia.cn/content/tesla/...-userguide.pdf My advice: stick to CPU-only without GPU "acceleration". I would not subscribe your statement "the simulation is slower with graphic card due to the partition". It is probably slower because your GPU is rather low-end and GPU acceleration is still a feature under development.

January 12, 2017, 05:21		#3
xab New Member xabi Join Date: Apr 2015 Posts: 13 Rep Power: 11	Thank you very much for your quick reply, I didn't know that this kind of simulation was so much consuming. After reading the link you posted, I will stick to cpu calculation. Thank you again for your help