RealFlow Official Forum

Hi

I spend some time on OpenCL & HyFlip-Domains. I attached my very simple test scene as well. Its about 16 Mio. particles. The domain has a cell size of about 0.1. The core particle sampling rate is 8, so in total 16 Mio particles. Creation mode of the domain is "Dense". Because in the doc is mentioned, that this mode is best for GPU. I start the tests via commando line ( to my experience the best way, because the GUI is quite unstable by big amount of particles). I simulated the first 23 Frames.

I´m on Mac . My machine is a MacPro 2010, 12 cores ( 2 x 6 Xeon 2.93) and i have a Nvidia Quadro 4000 installed ( 2 GB VRAM). I installed the newst driver for OS from Nvidia webpage (313.01.01f03) .

I also have an old PC ( 4 years old or more , Win7 but with a Quadro FX5800 equipped, 4 cores 3 GHz, 8 GB Ram in total old system), i tested the same scene here as well. Here i installed also the newest driver.

All results based on starting via commando line:
MacPro - no GPU, OpenCL totally deactivated - only CPU, 24 Threads : 17 min, 11 sec.
MacPro - with activated GPU / OpenCL on GPU and CPU : 34 min , 14 sec (!!!) ((
Old Pc - 4 cores, no Hypherthreading, OpenCL etc. activated, FX5800 : 17 min, 09 sec

The old PC runs really well with FX5800 compared with the big MacPro. The Quadro 4000 on Mac seems to be terrible , slows down the sim dramatically.

I spend some time on new Nvidia Cards for MacPro and got the following link http://forums.macrumors.com/showthread. ... ht=gtx+680. Take some time there in the forum, if you are on Mac, worthful information there. ( I will go to a GTX 680 4GB soon).

I would like to see some recommendations directly from NL about preferred graphic cards.

Perhaps someone could run the scene as well, would like to hear your results, in best case from Mac users .

Best,
Tom
(posted on official RF forum as well )

https://dl.dropboxusercontent.com/u/15698846/OpenCL.flw

Pinochio wrote:Hi

I spend some time on OpenCL & HyFlip-Domains. I attached my very simple test scene as well. Its about 16 Mio. particles. The domain has a cell size of about 0.1. The core particle sampling rate is 8, so in total 16 Mio particles. Creation mode of the domain is "Dense". Because in the doc is mentioned, that this mode is best for GPU. I start the tests via commando line ( to my experience the best way, because the GUI is quite unstable by big amount of particles). I simulated the first 23 Frames.

I´m on Mac . My machine is a MacPro 2010, 12 cores ( 2 x 6 Xeon 2.93) and i have a Nvidia Quadro 4000 installed ( 2 GB VRAM). I installed the newst driver for OS from Nvidia webpage (313.01.01f03) .

I also have an old PC ( 4 years old or more , Win7 but with a Quadro FX5800 equipped, 4 cores 3 GHz, 8 GB Ram in total old system), i tested the same scene here as well. Here i installed also the newest driver.

All results based on starting via commando line:
MacPro - no GPU, OpenCL totally deactivated - only CPU, 24 Threads : 17 min, 11 sec.
MacPro - with activated GPU / OpenCL on GPU and CPU : 34 min , 14 sec (!!!) ((
Old Pc - 4 cores, no Hypherthreading, OpenCL etc. activated, FX5800 : 17 min, 09 sec

The old PC runs really well with FX5800 compared with the big MacPro. The Quadro 4000 on Mac seems to be terrible , slows down the sim dramatically.

I spend some time on new Nvidia Cards for MacPro and got the following link http://forums.macrumors.com/showthread. ... ht=gtx+680. Take some time there in the forum, if you are on Mac, worthful information there. ( I will go to a GTX 680 4GB soon).

I would like to see some recommendations directly from NL about preferred graphic cards.

Perhaps someone could run the scene as well, would like to hear your results, in best case from Mac users .

Best,
Tom
(posted on official RF forum as well )

https://dl.dropboxusercontent.com/u/15698846/OpenCL.flw

Hi Tom,

Can you please post the OpenCL information about both machines? Mainly for checking which OpenCL versions are both running.

I can try to run your benchmark on this Mac, but it has the cheap GeForce GT120 from the 2009 Mac Pro. I'll let you know what comes out.

Finally, for benchmarking I would definitively disable any data export.

Best regards