[Paraview] Hardware suggestions for volume rendering
Aashish Chaudhary
aashish.chaudhary at kitware.com
Tue Mar 24 19:31:39 EDT 2015
Okay, I will talk to some folks here to see what we can do to address these
issues.
- Aashish
On Tue, Mar 24, 2015 at 6:34 PM, Burlen Loring <bloring at lbl.gov> wrote:
> I have the same issue with unstructured grids(not using the new opengl
> work yet). A while back I wanted to volume render a large image dataset
> after clipping it. The clip made it unstructured. This was on a few
> hundered nodes on edison, I couldn't get the unstructured volume rendering
> to work, it appeared to hang(took so long I couldn't tell) and it crashed a
> few times as well(may have ran out of memory). I think I tried a couple of
> the various mappers before I gave up. Would be nice to have it working
> better.
>
>
> On 3/24/2015 8:43 AM, Armin Wehrfritz wrote:
>
>> Hey Kevin and Aashish,
>>
>> I found this discussion rather interesting, especially since I'm
>> basically facing the same problem as Kevin.
>>
>> In almost all cases, we are dealing here with large unstructured grids
>> (e.g. 12M cells) and the volume rendering is performance is extremely bad,
>> not to say unusable, for these cases.
>>
>> I only tested it with ParaView 4.3.1 and older versions, i.e. with the
>> old OpenGL algorithms, but the performance is barely influenced by the
>> actual volume renderer (i.e. GPU vs. CPU based).
>>
>> For an unstructured grid the volume rendering seems to be CPU limited in
>> any case, at least my CPU is at 100% for quite some time and the GPU load
>> is negligible. ParaView's status bar indicates that it's busy with
>> "OpenGLProjectedTetrahedraMapper".
>>
>> This can be easily reproduced with the wavelet source (with e.g. 125k
>> points) and tessellation filter.
>>
>> So far I have resorted to resampling the data onto a structured grid.
>> This has the clear drawback that a structured grid with the same resolution
>> has about 70M grid points compared to the 12M points of the unstructured
>> grid.
>> The volume rendering performance of a 70M unstructured grid is however
>> already decent with the old OpenGL algorithms.
>>
>> So is there a way to improve the rendering of unstructured grids to avoid
>> the resampling?
>>
>> Best regards,
>> Armin
>>
>>
>>
>>
>> On 03/24/2015 04:48 PM, Aashish Chaudhary wrote:
>>
>>> Hey Kevin,
>>>
>>>
>>>
>>> On Fri, Mar 20, 2015 at 2:33 PM, Kevin Lau <kevin.d.lau at gmail.com
>>> <mailto:kevin.d.lau at gmail.com>> wrote:
>>>
>>> Dear Aashish,
>>>
>>> Thanks for your reply.
>>>
>>> (1)
>>>
>>> We tried testing it with 2 different data sets of simulation data,
>>> one medium sized and one very large:
>>>
>>> Medium
>>> Cells: 1309489
>>> Points: 247984
>>> Memory [MB]: 119.179
>>> Geometry [MB]: 8.201
>>>
>>> Very Large
>>> Cells: 51937191
>>> Points: 11019274
>>> Memory [MB]: 4709.731
>>> Geometry [MB]: 776.237
>>>
>>> These datasets are unstructured grids (tetrahedral meshes) read into
>>> ParaView using the Phasta data reader.
>>>
>>> I see. Most of the performance improvements we have achieved is with GPU
>>> RayCast Volume mapper (uses image data). If you can sample your data
>>> into a strucuted grid then you will be able to use the new mapper.
>>>
>>> Thanks,
>>>
>>> (2)
>>>
>>> The OS used was Ubuntu
>>>
>>> (3)
>>>
>>> The graphics card was a NVIDIA Quatro K6000 running the NVIDIA
>>> drivers from the Ubuntu Repos (v 331.113)
>>>
>>> (4)
>>>
>>> We configured ParaView using the default CMake options, changing
>>> VTK_RENDERING_BACKEND=OpenGL2 and disabling all the plug-ins. These
>>> tests were run in serial, as building with MPI was throwing errors.
>>>
>>> (5)
>>>
>>> We expected that the OpenGL2 version of ParaView would be able to
>>> volume render faster than the standard OpenGL version. However the
>>> reported frame rates were the same in both builds when clicking and
>>> dragging (~2 FPS).
>>>
>>> This is a problem because we're mainly interested in visualising our
>>> data and having users able to rotate the model and interact with it.
>>> At these frame rates the response is a bit sluggish.
>>>
>>> Also I've done serial test using an OpenGL2 build on Windows using
>>> an old GeForce 8800 with the same medium sized data. On this machine
>>> the frame rate was similar at around ~1.5 FPS.
>>>
>>> (6)
>>>
>>> ParaView was built by checking out the master from git. Sadly I'm
>>> not on that test machine now, so I can't check the SHA
>>>
>>> Many thanks for your help again.
>>>
>>> Kevin
>>>
>>>
>>> On 20 March 2015 at 10:33, Aashish Chaudhary
>>> <aashish.chaudhary at kitware.com
>>> <mailto:aashish.chaudhary at kitware.com>> wrote:
>>>
>>> Hey Kevin,
>>>
>>> On Thu, Mar 19, 2015 at 5:33 PM, Kevin Lau
>>> <kevin.d.lau at gmail.com <mailto:kevin.d.lau at gmail.com>> wrote:
>>>
>>> Dear Aashish,
>>>
>>> Thanks for your reply, I've now managed to build ParaView
>>> from the Master branch with VTK_RENDERING_BACKEND set to
>>> OpenGL2 and I have noticed a visible difference in the
>>> rendering.
>>>
>>> However comparison of the rendering performance in the
>>> interactive mode of builds with the standard OpenGL backend
>>> vs the OpenGL2 backend, both show similar frame rates.
>>>
>>>
>>> Depending on the card and features you are using you may not see
>>> a huge bump in performance. On some systems we have seen 300
>>> times performance improvements and on others there were close
>>> specially when you have VSync on. Said that, can you provide
>>> more infromation on
>>>
>>> 1) Data size
>>> 2) OS
>>> 3) Graphics card
>>> 4) Features you have enabled?
>>> 5) You expectation and problem
>>> 6) What SHA of ParaView you are using?
>>>
>>> Are there other build options that I'm missing? Should I be
>>> building the VTK libraries themselves?
>>>
>>>
>>> Just make sure that you built from a clean state other than that
>>> you should be good.
>>> Thanks,
>>>
>>>
>>> Many thanks,
>>>
>>> Kevin
>>>
>>>
>>> On 3 March 2015 at 14:53, Aashish Chaudhary
>>> <aashish.chaudhary at kitware.com
>>> <mailto:aashish.chaudhary at kitware.com>> wrote:
>>>
>>> On Tue, Mar 3, 2015 at 2:43 PM, Kevin Lau
>>> <kevin.d.lau at gmail.com <mailto:kevin.d.lau at gmail.com>>
>>> wrote:
>>>
>>> Dear Aashish,
>>>
>>> Thank you very much for your reply. The latest work
>>> you're doing on volume rendering is very impressive.
>>>
>>>
>>> Thanks.
>>>
>>>
>>> In order to access the latest VTK rendering in
>>> ParaView would we have to compile both the latest
>>> VTK and ParaView from source? Is it possible to link
>>> ParaView to specific VTK builds?
>>>
>>>
>>> Latest ParaView from source should be fine. Just make
>>> sure that you select VTK_RENDERING_BACKEND to OpenGL2
>>>
>>>
>>> Also do you know if it is preferential to have a
>>> single GPU (i.e. a workstation graphics card) or
>>> multiple GPU (i.e. gaming graphics card) when using
>>> ParaView?
>>>
>>>
>>> If I remember correctly, we don't explicitly assign
>>> tasks to GPU in ParaView if running a single process.
>>> You can find more information here:
>>> http://www.paraview.org/Wiki/Setting_up_a_ParaView_Server#
>>> Multiple_GPUs_Per_Node
>>>
>>> Hope it helps.
>>>
>>>
>>> Many thanks,
>>>
>>> Kevin
>>>
>>>
>>> On 27 February 2015 at 08:15, Aashish Chaudhary
>>> <aashish.chaudhary at kitware.com
>>> <mailto:aashish.chaudhary at kitware.com>> wrote:
>>>
>>> Dear Kevin,
>>>
>>> First, I would like to mention that we have been
>>> working hard to improve rendering in VTK that
>>> includes volume rendering as well. You may want
>>> to read this source article:
>>> http://www.kitware.com/source/home/post/154.
>>> Most of the changes related to this work is now
>>> available in VTK github repository in master
>>> branch. If you are targetting current VTK, I
>>> would suggest that any card that can support
>>> OpenGL2.1 or higher would be great. The higher
>>> the GPU memory the better. While these newer
>>> card will work with the previous version of
>>> volume rendering, it may not scale very well (in
>>> our testing that's what we have observed).
>>> Current the default in ParaView is OpenGL Volume
>>> Rendering but that is going to change sometime
>>> soon and once that happens you will have OpenGL2
>>> backend for everything.
>>>
>>> Hope this helps.
>>> - Aashish
>>>
>>>
>>>
>>> On Thu, Feb 26, 2015 at 5:41 PM, Kevin Lau
>>> <kevin.d.lau at gmail.com
>>> <mailto:kevin.d.lau at gmail.com>> wrote:
>>>
>>> Hi,
>>>
>>> I'm looking to putting together a
>>> workstation to perform realtime volume
>>> rendering of our simulation data.
>>>
>>> Are there any particular GPU features I
>>> should look for?
>>>
>>> If there are multiple GPUs will Paraview be
>>> able to take advantage of each GPU for
>>> volume rendering?
>>>
>>> Many thanks
>>>
>>> Kevin
>>>
>>>
>>>
>>> _______________________________________________
>>> Powered by www.kitware.com
>>> <http://www.kitware.com>
>>>
>>> Visit other Kitware open-source projects at
>>> http://www.kitware.com/opensource/opensource.html
>>>
>>> Please keep messages on-topic and check the
>>> ParaView Wiki at:
>>> http://paraview.org/Wiki/ParaView
>>>
>>> Search the list archives at:
>>> http://markmail.org/search/?q=ParaView
>>>
>>> Follow this link to subscribe/unsubscribe:
>>> http://public.kitware.com/mailman/listinfo/paraview
>>>
>>>
>>>
>>>
>>> --
>>> /| Aashish Chaudhary
>>> | Technical Leader
>>> | Kitware Inc.
>>> /
>>> /|
>>> http://www.kitware.com/company/team/chaudhary.html/
>>>
>>>
>>>
>>>
>>>
>>> --
>>> /| Aashish Chaudhary
>>> | Technical Leader
>>> | Kitware Inc.
>>> /
>>> /| http://www.kitware.com/company/team/chaudhary.html/
>>>
>>>
>>>
>>>
>>>
>>> --
>>> /| Aashish Chaudhary
>>> | Technical Leader
>>> | Kitware Inc.
>>> /
>>> /| http://www.kitware.com/company/team/chaudhary.html/
>>>
>>>
>>>
>>>
>>>
>>> --
>>> /| Aashish Chaudhary
>>> | Technical Leader
>>> | Kitware Inc.
>>> /
>>> /| http://www.kitware.com/company/team/chaudhary.html/
>>>
>>>
>>> _______________________________________________
>>> Powered by www.kitware.com
>>>
>>> Visit other Kitware open-source projects at http://www.kitware.com/
>>> opensource/opensource.html
>>>
>>> Please keep messages on-topic and check the ParaView Wiki at:
>>> http://paraview.org/Wiki/ParaView
>>>
>>> Search the list archives at: http://markmail.org/search/?q=ParaView
>>>
>>> Follow this link to subscribe/unsubscribe:
>>> http://public.kitware.com/mailman/listinfo/paraview
>>>
>>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at http://www.kitware.com/
>> opensource/opensource.html
>>
>> Please keep messages on-topic and check the ParaView Wiki at:
>> http://paraview.org/Wiki/ParaView
>>
>> Search the list archives at: http://markmail.org/search/?q=ParaView
>>
>> Follow this link to subscribe/unsubscribe:
>> http://public.kitware.com/mailman/listinfo/paraview
>>
>
>
--
*| Aashish Chaudhary | Technical Leader | Kitware Inc. *
*| http://www.kitware.com/company/team/chaudhary.html
<http://www.kitware.com/company/team/chaudhary.html>*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview/attachments/20150324/e478e19b/attachment.html>
More information about the ParaView
mailing list