[visit-users] Local rendering, and how do I know it's using the graphics card? (and other silly questions)

Burlen Loring bloring at lbl.gov
Thu Jul 10 13:04:26 EDT 2014


> Trying to manage multiple GPU's on the same node was virtually 
> impossible as I recall! 
I didn't find it very difficult with ParaView. Given a multi device / 
screen x11 config, one simply needs to use DISPLAY to select the GPU. I 
had used some metadata provided by batch system to figure out the MPI 
rank in the batch script and set the DISPLAY so that GPU are mapped to 
local processes in a round robin manner. on some systems I've had to 
start/stop x,  a pain but not too hard. The hardest part for me was 
writing the xorg conf files. that was a pain!

On 07/10/2014 09:52 AM, Angelini, Richard C (Rick) CIV USARMY ARL (US) 
wrote:
> Conceptually you'd think (and expect) better performance from GPU and 
> Mesa -- but in either case, I agree with Eric that it's very difficult 
> to manage from a script standpoint -- at lease a  couple of years ago 
> when I last tried this. Trying to manage multiple GPU's on the same 
> node was virtually impossible as I recall!   8-)   So, by default Mesa 
> is "better" because it can be easily managed and scripted.
>
>
>
> ________________________________
> Rick Angelini
> USArmy Research Laboratory
> CISD/HPC Architectures Team
> Phone:  410-278-6266
>
> From: Burlen Loring <burlen.loring at gmail.com 
> <mailto:burlen.loring at gmail.com>>
> Reply-To: VisIt software users community <visit-users at elist.ornl.gov 
> <mailto:visit-users at elist.ornl.gov>>
> Date: Thursday, July 10, 2014 at 12:40 PM
> To: VisIt software users community <visit-users at elist.ornl.gov 
> <mailto:visit-users at elist.ornl.gov>>
> Subject: Re: [visit-users] Local rendering, and how do I know it's 
> using the graphics card? (and other silly questions)
>
> Don't disagree with you, but on the ParaView comment: My experience is 
> that GPU's can be significantly faster, with the caveat that it's 
> depends on: rendering algorithm in use, on the number of processes 
> sharing the GPU, and screen size and vertex count. Obviously if you 
> have 16 processes hitting the same GPU it can be a problem. In those 
> situations CPU is a better choice. Here are some benchmarks 
> <http://www.paraview.org/Wiki/ParaView_And_Mesa_3D#A_comparison_of_OSMesa_Gallium_llvmpipe.2C_OSMesa_classic_and.2C_GPU_Accelerated_Rendering_Performance> 
> made recently using VTK comparing GPU to recent OSMesa. You can see 
> that GPU offers big speed up in some cases and not so much in others. 
> Note the red bars show the OSMesa classic which is what VisIt is 
> currently using if you build with build_visit.
>
> One potential advantage of having GPU's is that in ParaView and VTK 
> this enables more rendering techniques. OSMesa especially the older 
> versions is quite limited in capabilities it provides.
>
> Actually I'm curious if using the GPU enables more features in VisIt?
>
>
> On 07/10/2014 08:08 AM, Brugger, Eric wrote:
>> Paul,
>>
>> There is a distributed GPU-based rendering in VisIt, we just don't generally recommend it. It may be tricky to set up and it generally doesn't improve performance. When running in parallel the amount of geometry to render per process is fairly small and modern CPUs can render small amounts of geometry very quickly and you don't have the cost of doing the frame buffer readback. The rendering rate is usually limited by the compositing, which requires shipping frame buffer fragments over the interconnect. I have heard of similar experiences from ParaView users at Sandia.
>>
>> Eric
>>
>> -----Original Message-----
>> From: Paul Melis [mailto:paul.melis at surfsara.nl]
>> Sent: Thursday, July 10, 2014 12:32 AM
>> To: VisIt software users community
>> Subject: Re: [visit-users] Local rendering, and how do I know it's using the graphics card? (and other silly questions)
>>
>> Hi Eric,
>>
>> On 09-07-14 17:18, Brugger, Eric wrote:
>>> If you are running completely locally you should be
>>   > using the graphics card. You can change the rendering  > options with the "Rendering" window, which can be  > brought up by going to "Options->Rendering".
>>   > You would need to go to the "Advanced" tab and  > change "Use scalable rendering" to "Always" to force  > it to always render in software. If you changed it to  > "Never" then it would always use the graphics card.
>>
>> I happened to be following this thread, a question: the above seems to imply that scalable rendering is always rendering in software (i.e. not using GPUs). Is that really true? Is there no distributed GPU-based rendering in VisIt like ParaView provides?
>>
>> Regards,
>> Paul
>>
>> --
>> ** SURFsara heeft een nieuw algemeen telefoonnummer: 020 800 1300 **
>>
>> Paul Melis
>> | Groepsleider & Adviseur Visualisatie | SURFsara | Science Park 140 |
>> | 1098 XG Amsterdam | T 020 592 30 59 |paul.melis at surfsara.nl  |
>> |www.surfsara.nl  |
>> --
>> VisIt Users Wiki:http://visitusers.org/  Frequently Asked Questions for VisIt:http://visit.llnl.gov/FAQ.html  To Unsubscribe: send a blank email tovisit-users-unsubscribe at elist.ornl.gov
>> More Options:https://elist.ornl.gov/mailman/listinfo/visit-users
>> --
>> VisIt Users Wiki:http://visitusers.org/
>> Frequently Asked Questions for VisIt:http://visit.llnl.gov/FAQ.html
>> To Unsubscribe: send a blank email tovisit-users-unsubscribe at elist.ornl.gov
>> More Options:https://elist.ornl.gov/mailman/listinfo/visit-users
>
>
>
> --
> VisIt Users Wiki: http://visitusers.org/
> Frequently Asked Questions for VisIt: http://visit.llnl.gov/FAQ.html
> To Unsubscribe: send a blank email to visit-users-unsubscribe at elist.ornl.gov
> More Options: https://elist.ornl.gov/mailman/listinfo/visit-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://elist.ornl.gov/pipermail/visit-users/attachments/20140710/9ecdfc24/attachment.html>


More information about the visit-users mailing list