Computer Build Advice

[Cambraceres]

Hello and thanks for any advice. I am compiling a new computer build, the primary use being Pix4D processing and manipulation of point clouds in Cloud Compare. I have noted that Pix4D claims the GPU does not figure into the bulk of the processing.

To what degree is a high end graphics card important for Cloud Compare? Right now, I have a 1080 TI with a Ryzen 7 1700X CPU, and I get some lag and on occasion CC will simply turn off.

Details for the intended build can be found here, entitled "1st Build"

https://docs.google.com/spreadsheets/d/ ... sp=sharing

Thank you

[daniel]

That's the same for CC, we only use the GPU for display. So it's more important to have a lot of GPU memory than anything else.

And it's maybe more subjective, but I still have the impression that NVidia cards are better at displaying large point clouds. I don't have a good / recent benchmark however.

[pavel.kuncl]

Good day

I need advice, we need to improve the graphics card.
Current PC:
Intel i9 10920X CPU OC
RAM 256GB DDR4 3200
MB ASUS X299
GPU RTX 3060 12GB
SSD-M2-256
SSD-SATA-2TB

LCD 2x LG UHD 27p

cloud calculation 1milon +

We need to improve the speed of graphic editing.

1 - NVIDIA RTX A4000 16GB (256) 4xDP
2 - NVIDIA RTX 3080 12GB
3 - AMD RX 6900XT 16GB

which one would work best?

Thank you Pavel

[daniel]

cloud calculation 1milon +

How many points exactly? Because with a 3060, you should be able to work with 100 million without too much issues...

[pavel.kuncl]

I'm sorry :) it's 1 billion +

[pavel.kuncl]

cloud calculation 1milon +

How many points exactly? Because with a 3060, you should be able to work with 100 million without too much issues...

I'm sorry :) it's 1 billion +

[daniel]

Ah, with a billion, I don't think any GPU can handle that directly in memory... Your only friend is the LoD mechanism (it takes some time to activate, as CC needs to build the octree first ... you can see the progress in the Console).

My personal preference would be the 3080 (or even 3090), but I have never tested or had any feedback on the A4000... If one consider the Quadro cards, they have always been slower when it's about rendering point clouds (at least with OpenGL).

[pavel.kuncl]

Ah, with a billion, I don't think any GPU can handle that directly in memory... Your only friend is the LoD mechanism (it takes some time to activate, as CC needs to build the octree first ... you can see the progress in the Console).

My personal preference would be the 3080 (or even 3090), but I have never tested or had any feedback on the A4000... If one consider the Quadro cards, they have always been slower when it's about rendering point clouds (at least with OpenGL).

power 3080 vs. 3060 will be how big? won't there be a 12GB ram size limit on a 4k monitor? that's why I chose rtx4000 with 16gb, 3090 and 24gb and the performance is elsewhere. Isn't it possible to benchmark 1 billion cards?

[daniel]

Sadly I don't own such cards ;)

[PablerasBCN]

I had to upgrade RAM ammount to handle large point clouds. Clouds with 300M+, once you start segmenting, and doing operations the mem usage ramps ups easily. I had 32GB, and while it would fit most of 300M point clouds it was allways at the edge and started to crash CC, so Upgraded to 64gb and had not have issues since. My suggestion would be to start with 32gb but have it in 2 modules so you can expand later on if needed. (I was lucky to 3 years later find exact same modules, you may need to keep this in mind)