Hi,
I'm reading several topics that discuss about hardware performance here (and there).
Everybody seems to agree that i7 architecture are better for pointcloud treatment than Xeon or else.
Every body seems also to agree that Nvidia CUDA graphic cards are better than AMD one's.
So what ?
I'm "self"-learning about pointcloud and their use, by testing every software I can find on big data set point cloud (one is about 270 station the other is over 1500). And I'm searching for good hardware solution.
At home, I have a brand new overclocked GTX 1070 with 8GB DDR5. At works I have a Quadro K2200 with 4GB Ram.
I'm testing :
Sketchup (for obj and collada import)
Meshmixer
Cloudcompare
3D Reshaper
Realworks
Revit
Recap
Autocad
My home config is based on i7 4700 16GB ram SSD disks
My Work config is on Xeon 1220 16GB ram Old Hard drive and one SSD
Now on I'm looking something that can mesure the effective collaboration of Cuda and graphical hardware in software calculation.
It seem that none of the soft I've tested really use the graphical hardware to increase their speed.
For me, in fact, the only difference I've seen is about graphic ram, the more quantity and the more speed the better ...
And sorry for Nvidia fans but previously I have a RADEON 7870 with 2GB, and when my home CPU was 100% load, I was still capable of watching full hd video that is no more affordable with the Nvidia 1070, it freeze, it cluster, I really think that there is a Driver or directx issue with Nvidia cards.
So what are your experience at measuring graphic board performance with your CAD or what else software ?
CUDA or not CUDA, how GPU helps ?
-
- I have made 10-20 posts
- Posts: 19
- Joined: Tue Nov 08, 2016 10:12 am
- 7
- Full Name: Christophe
- Company Details: AB VD
- Company Position Title: Technician
- Country: France
- Linkedin Profile: No
-
- V.I.P Member
- Posts: 846
- Joined: Tue Feb 26, 2013 9:32 pm
- 11
- Full Name: Massimo De Marchi
- Company Details: Massimo De Marchi
- Company Position Title: freelancer
- Country: Italy
- Skype Name: massimo_de_marchi
- Has thanked: 15 times
- Been thanked: 53 times
- Contact:
Re: CUDA or not CUDA, how GPU helps ?
In my experience CUDA boosts performance during the first steps of photogrammetry reconstruction (point matching) using Pix4D. The time saving is huge.
You have not advantage with generation of point cloud, mesh, orthophoto.
With Pix4D you can enable and disable CUDA.
I timed the same project twice, so in this case it was easy.
Massimo
You have not advantage with generation of point cloud, mesh, orthophoto.
With Pix4D you can enable and disable CUDA.
I timed the same project twice, so in this case it was easy.
Massimo
ing. Massimo De Marchi - +39 347 32 17 049 - www.studiodemarchi.net
-
- V.I.P Member
- Posts: 252
- Joined: Sat Apr 09, 2016 3:10 pm
- 7
- Full Name: Bashar Dawood
- Company Details: Canada West Drafting Inc.
- Company Position Title: Owner
- Country: Calgary Canada
- Linkedin Profile: Yes
- Has thanked: 24 times
- Been thanked: 15 times
- Contact:
Re: CUDA or not CUDA, how GPU helps ?
Did any one tried the Xeon PHI coprocessor or the Nvidia Tesla, or have any info on how good are they for point cloud processing and generating
- gilles_3DR
- V.I.P Member
- Posts: 235
- Joined: Mon Dec 08, 2014 11:41 am
- 9
- Full Name: Gilles Monnier
- Company Details: Hexagon
- Company Position Title: 3DR Technical Manager
- Country: France
- Linkedin Profile: Yes
- Has thanked: 2 times
- Been thanked: 28 times
Re: CUDA or not CUDA, how GPU helps ?
Hi,
I'm not surprised with what you have found.
For sure, using the gpu when processing data would be very nice.
Most of the SW use gpu when rendering the data of course, but running data processing on the gpu is another story...
Some algorithm are more easy to implement on gpu than others. Photogrammetry algorithms are good candidate as far as I know...
I'm not surprised with what you have found.
For sure, using the gpu when processing data would be very nice.
Most of the SW use gpu when rendering the data of course, but running data processing on the gpu is another story...
Some algorithm are more easy to implement on gpu than others. Photogrammetry algorithms are good candidate as far as I know...
-
- I have made 10-20 posts
- Posts: 19
- Joined: Tue Nov 08, 2016 10:12 am
- 7
- Full Name: Christophe
- Company Details: AB VD
- Company Position Title: Technician
- Country: France
- Linkedin Profile: No
Re: CUDA or not CUDA, how GPU helps ?
Really I'm a bit disappointed by the low use of the GPU.
Cuda exist from a long time now, and it seems to be really poorly implemented in those software that need so much calculation power. (or I maybe wrong)
However I'm questioning myself about what would be more efficient in these. Calculation power vs Data bit rate. Assuming the biggest problem is about the quantity of data to be treated maybe not the calculs complexity by itself.
And now the real question is about commercial policy of Nvidia and difference between Quadro and GTX cards.
As CUDA is not so well implemented in those software, what is the purpose of using professionnal quadro cards in our application ? can't see the real "cost effective" investment.
Seems AMD RADEON card should be great with this, because I'm really convincing myself that the real speed factor is about memory quantity and processor frequency. (even multi threading doesn't seems to be so "powerfull")
Maybe I'm not using the correct softwares to works also. Does it exist kind of a "software benchmarking" platform ?
(in fact laserscanningforum is a bit of a one but it's a forum so may not be as "clinical" as it should be needed to make real scientific comparison and conclusion)
Cuda exist from a long time now, and it seems to be really poorly implemented in those software that need so much calculation power. (or I maybe wrong)
However I'm questioning myself about what would be more efficient in these. Calculation power vs Data bit rate. Assuming the biggest problem is about the quantity of data to be treated maybe not the calculs complexity by itself.
And now the real question is about commercial policy of Nvidia and difference between Quadro and GTX cards.
As CUDA is not so well implemented in those software, what is the purpose of using professionnal quadro cards in our application ? can't see the real "cost effective" investment.
Seems AMD RADEON card should be great with this, because I'm really convincing myself that the real speed factor is about memory quantity and processor frequency. (even multi threading doesn't seems to be so "powerfull")
Maybe I'm not using the correct softwares to works also. Does it exist kind of a "software benchmarking" platform ?
(in fact laserscanningforum is a bit of a one but it's a forum so may not be as "clinical" as it should be needed to make real scientific comparison and conclusion)