Thinkbox Releases Sequoia 1.0!
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
Got my mesh computer finally running today.
Intel Server MB, S2600CW, 16 RAM slots
8 Crucial 32GB DDR4 2133 LRDIMM ECC = 256GB RAM
2 Xeon E5 V4 2637 3.5 GHz CPU
2 10GbE ports
500 GB Samsung 850 SSD Boot disk
1 M.2 SSD port
10 SAS/SATA ports
5 PCI-e 3.0 slots
GTX 980 video waiting for GTX 1080 price to drop
Maybe this is not the thread for this discussion, but it took a lot of research to be sure that all the parts matched and could be housed in a non-Intel case. Intel tech support was very helpful once you get the right phone number.
If things work out and I need more RAM, I can add another 8 sticks of 32GB, to equal 512GB for $2K more. I can also add 16 sticks of 64GB to generate 1TB of RAM for $15K
Have not figured out what kind of storage will interface with this and no additional time to run any compute examples since I am leaving for HXGN in Anaheim, CA in the next 24 hours.
Intel Server MB, S2600CW, 16 RAM slots
8 Crucial 32GB DDR4 2133 LRDIMM ECC = 256GB RAM
2 Xeon E5 V4 2637 3.5 GHz CPU
2 10GbE ports
500 GB Samsung 850 SSD Boot disk
1 M.2 SSD port
10 SAS/SATA ports
5 PCI-e 3.0 slots
GTX 980 video waiting for GTX 1080 price to drop
Maybe this is not the thread for this discussion, but it took a lot of research to be sure that all the parts matched and could be housed in a non-Intel case. Intel tech support was very helpful once you get the right phone number.
If things work out and I need more RAM, I can add another 8 sticks of 32GB, to equal 512GB for $2K more. I can also add 16 sticks of 64GB to generate 1TB of RAM for $15K
Have not figured out what kind of storage will interface with this and no additional time to run any compute examples since I am leaving for HXGN in Anaheim, CA in the next 24 hours.
- LPaulCook
- V.I.P Member
- Posts: 485
- Joined: Sat Jul 18, 2015 9:42 pm
- 8
- Full Name: L Paul Cook
- Company Details: Land surveyor doing LiDAR
- Company Position Title: President
- Country: United States
- Linkedin Profile: Yes
- Location: Santa Barbara CA
- Has thanked: 47 times
- Been thanked: 44 times
- Contact:
Re: Thinkbox Releases Sequoia 1.0!
Dennis,
That should move things along quite nicely.
Do you really need that much of a PC for your meshing?
That should move things along quite nicely.
Do you really need that much of a PC for your meshing?
L. Paul Cook, PLS
www.LPC3D.com
www.LPC3D.com
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
I did a quick processing job using my standard APT Z+F 6 scan project illustrated in this thread. I ran it with 99% removal of the of the meshes using Sequoia with a radius of 5mm. I used a single cell for the entire 6 scans, rather than the 7X7X1 in the previous examples. In the test today, the CPU fluctuated between 60 to 90% for 30 minutes, rather than pegged at 100% the majority of the run for many hours. Ram was never more than 70 to 80GB or 30% of the 256GB, as opposed to crashing at 99% of the 64GB on our other system runs.
The only problem was the ram available tried to load over 2B faces, which could not be addressed using 32bit integer variables. They tell me they will be increasing that to 4B faces soon.
Having spoken to a few people about this mesh computer build, I am surprised how many people are building these type of computers according to the Intel technical support person that I spoke to several days ago.
The only problem was the ram available tried to load over 2B faces, which could not be addressed using 32bit integer variables. They tell me they will be increasing that to 4B faces soon.
Having spoken to a few people about this mesh computer build, I am surprised how many people are building these type of computers according to the Intel technical support person that I spoke to several days ago.
- LPaulCook
- V.I.P Member
- Posts: 485
- Joined: Sat Jul 18, 2015 9:42 pm
- 8
- Full Name: L Paul Cook
- Company Details: Land surveyor doing LiDAR
- Company Position Title: President
- Country: United States
- Linkedin Profile: Yes
- Location: Santa Barbara CA
- Has thanked: 47 times
- Been thanked: 44 times
- Contact:
Re: Thinkbox Releases Sequoia 1.0!
Interesting Dennis. Thanks.
One thing I noticed when I bumped my RAM to 128GB was I soon was missing alot of capacity in my SSD boot drive. I finally discovered that the hybranation system had to make a space on that drive equal to the amount of RAM I had. This was about half of my boot drive! I only had a 256GB SSD at the time. I eventually determined how to eliminate the hybranation space on my boot drive which I never used anyway and gained all that space back.
So, you my want to look into doing the same thing unless you use the hybranation function.
Just a thought...
One thing I noticed when I bumped my RAM to 128GB was I soon was missing alot of capacity in my SSD boot drive. I finally discovered that the hybranation system had to make a space on that drive equal to the amount of RAM I had. This was about half of my boot drive! I only had a 256GB SSD at the time. I eventually determined how to eliminate the hybranation space on my boot drive which I never used anyway and gained all that space back.
So, you my want to look into doing the same thing unless you use the hybranation function.
Just a thought...
L. Paul Cook, PLS
www.LPC3D.com
www.LPC3D.com
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
Paul
This is why you need lots of RAM in some meshing instances.
Depending on what meshing software you are using, the work flow will be different to accomplish the end result.
I am attempting to mesh point clouds consisting of many dense scans with billions of points. One can use Sequoia to accomplish this goal with computer systems with moderate CPU and RAM. The use of Hacksaw and Deadline in the workflow will accomplish this using 16GB of RAM on a laptop and additional Servers.
My goal is to get it accomplished faster and take as an experiment, more than 250M points to see if one can generate meshes of 500M to 1B faces and then reduce the meshes to less than 2M faces. The good thing about Sequoia, is I can load our raw projects/scans directly from Z+F 5010X and Riegl VZ400 scanners, without converting to an intermediate format like LAS or E57 (although we could).
Original point cloud = 264,345,799 points. Using Sequoia, produced a mesh using Hacksaw in a 7X7X3=147 matrix before using my current system. This produced a series of 147 matrices, the following mesh, which I "welded" into a single 24M face mesh seen below.
faces=24,412,134
points=12,065,641
Sequoia has the ability to take the reduced single mesh and reduce it again, in a uniform manner. So I reduced the mesh from 24M faces to 2M faces, which looked OK, but decided that 5M faces looked better. This is still pretty good considering that I started at a point of at least 500M faces.
Now I know why I might need the $2K to increase my RAM to 512GB, or try to figure out if it is cheaper on AWS.
This is why you need lots of RAM in some meshing instances.
Depending on what meshing software you are using, the work flow will be different to accomplish the end result.
I am attempting to mesh point clouds consisting of many dense scans with billions of points. One can use Sequoia to accomplish this goal with computer systems with moderate CPU and RAM. The use of Hacksaw and Deadline in the workflow will accomplish this using 16GB of RAM on a laptop and additional Servers.
My goal is to get it accomplished faster and take as an experiment, more than 250M points to see if one can generate meshes of 500M to 1B faces and then reduce the meshes to less than 2M faces. The good thing about Sequoia, is I can load our raw projects/scans directly from Z+F 5010X and Riegl VZ400 scanners, without converting to an intermediate format like LAS or E57 (although we could).
Original point cloud = 264,345,799 points. Using Sequoia, produced a mesh using Hacksaw in a 7X7X3=147 matrix before using my current system. This produced a series of 147 matrices, the following mesh, which I "welded" into a single 24M face mesh seen below.
faces=24,412,134
points=12,065,641
Sequoia has the ability to take the reduced single mesh and reduce it again, in a uniform manner. So I reduced the mesh from 24M faces to 2M faces, which looked OK, but decided that 5M faces looked better. This is still pretty good considering that I started at a point of at least 500M faces.
Now I know why I might need the $2K to increase my RAM to 512GB, or try to figure out if it is cheaper on AWS.
You do not have the required permissions to view the files attached to this post.
Last edited by dhirota on Tue Jun 21, 2016 8:37 am, edited 1 time in total.
- arie
- I have made 40-50 posts
- Posts: 47
- Joined: Wed Jun 05, 2013 11:43 am
- 10
- Full Name: Arie Kai-Browne
- Company Details: University of Applied Sciences Berlin
- Company Position Title: Scientific staff member
- Country: Germany
- Skype Name: aaaarie
- Linkedin Profile: No
- Location: Berlin
- Been thanked: 11 times
Re: Thinkbox Releases Sequoia 1.0!
Dennis,
without knowing much about Sequoias meshing capabilities, this might be a naive question but why try to create billions of faces just for decimating it to a couple million? Would it not be much easier to subsample the pointcloud and mesh it afterwards?
Or do you use the high-res mesh for calculating depth- and/or normalmaps to use tessellation or similiar on the reduced mesh?
I could understand than one might use the full-res pointcloud if the meshing software uses the information of an ordered pointcloud to select the points with the highest quality (based on incident angle and distance) for creating the final mesh. But does Sequoia?
Meshing is quite a interesting topic with so many possible workflows, always nice to hear about the different ones. Seems like I need to get a trial version of Sequoia.
Cheers.
without knowing much about Sequoias meshing capabilities, this might be a naive question but why try to create billions of faces just for decimating it to a couple million? Would it not be much easier to subsample the pointcloud and mesh it afterwards?
Or do you use the high-res mesh for calculating depth- and/or normalmaps to use tessellation or similiar on the reduced mesh?
I could understand than one might use the full-res pointcloud if the meshing software uses the information of an ordered pointcloud to select the points with the highest quality (based on incident angle and distance) for creating the final mesh. But does Sequoia?
Meshing is quite a interesting topic with so many possible workflows, always nice to hear about the different ones. Seems like I need to get a trial version of Sequoia.
Cheers.
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
Arie
What you described maybe the best answer, if the decimation is automatic and you get the answer one is looking for the solution.
Whether Sequoia is using the selection of the ordered point clouds to select the points with the highest quality, I am not sure.
What you described maybe the best answer, if the decimation is automatic and you get the answer one is looking for the solution.
Whether Sequoia is using the selection of the ordered point clouds to select the points with the highest quality, I am not sure.
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
In the last post a few days ago, I should have included some closer screen images to show the quality of the meshes. These are screen shots, reduced to 1000 pixels to fit on the LSF format. Total project faces = 7.5Million, as compared to 24M and 5M faces in the images shown earlier in the thread and approximately 500M total in the original point cloud.
We are still trying to reduce large point clouds to the size that will allow direct 3D color printing with less than 2M faces. If some one has an easier workflow, we would be interested in trying it.
We are still trying to reduce large point clouds to the size that will allow direct 3D color printing with less than 2M faces. If some one has an easier workflow, we would be interested in trying it.
You do not have the required permissions to view the files attached to this post.
- Aaron_Hazen
- V.I.P Member
- Posts: 146
- Joined: Wed Oct 29, 2014 6:31 pm
- 9
- Full Name: Aaron J.R. Hazen
- Company Details: Coast 2 Coast Surveys
- Company Position Title: Senior Project Manager - 3d Architecture
- Country: United States
- Linkedin Profile: Yes
- Has thanked: 1 time
-
- V.I.P Member
- Posts: 958
- Joined: Sun Nov 01, 2009 11:18 pm
- 14
- Full Name: Dennis Hirota
- Company Details: Sam O Hirota Inc
- Company Position Title: President
- Country: USA
- Linkedin Profile: Yes
- Location: Hawaii, USA
- Has thanked: 87 times
- Been thanked: 379 times
Re: Thinkbox Releases Sequoia 1.0!
Aaron
We have Geomagic and Meshlab, which will not accomplish what we are trying to do with our point clouds at this time. We have tried many others, but prefer not to name them since we have not purchased them or not currently using the software.
If you have other meshing apps or workflows to suggest, I would be interesting in learning of them.
We have Geomagic and Meshlab, which will not accomplish what we are trying to do with our point clouds at this time. We have tried many others, but prefer not to name them since we have not purchased them or not currently using the software.
If you have other meshing apps or workflows to suggest, I would be interesting in learning of them.