Wondering about requirements for CAD 3D

[Gastje88]

Hi all, 

I'm new to the forum, familiar with the content of Linus. 

I'm an mechanical engineer in Holland and have a lot of experience with solidworks and Inventor. 

 

Looking at all the game pc builds I started to make some plans of making one for myself to replace my old crappy laptop. 

Although I also want to be able to do some 3d modeling on it. 

Now in the experience I have the most parametric cad programs can only run calculations on 1 core (because solving a model happens sequential). 

I have recently worked on a pc that was equipped with a p2000 gpu, but when the model was struggling the gpu had a 3% load. 

So I thought, is a gpu a requirement for 3d  work? 

Do I need as many cores as possible for CAD work? 

 

Hope the community can enlighten me a bit on the requirements for CAD programs as I don't see the use for many cores and gpu power ( except when rendering images of course) 

Aldo never saw a video about cad stations which I would like to see ;) 

 

Thanks in advance, 

 

[WereCatf]

1 minute ago, Gastje88 said:

So I thought, is a gpu a requirement for 3d  work? 

Do I need as many cores as possible for CAD work? 

A GPU isn't a requirement, but some CAD-software can make some use of one. That said, most of the heavy lifting is done on the CPU. As for how many cores...well, that depends on the software. I have only personal experience with Fusion 360, but it certainly can use multiple cores. If the software you use cannot take advantage of multiple cores, well, that depends on the software you use.

[TVwazhere]

4 minutes ago, Gastje88 said:

So I thought, is a gpu a requirement for 3d  work? 

Do I need as many cores as possible for CAD work? 

For Inventor and Solidworks, no you do not. I currently run a P1000 that see's single digit % usage because it's only job is to output an image. 

 

Inventor and solidworks take advantage of multi threads in Rendering, Stress analysis and flow analysis work environments, so there is a place for multi-core systems in those use cases, but in the end, your day-to-day part creation, assembly and 2D drawing will only require 1 core (as the base code is the same from the early days of the programs lifespans and would require a complete rewriting of the program)

 

More modern rendering programs like 3DSMax, Blender and Fusion360 can take advantage of multiple cores (admittedly I've never used any of these programs though)

[LogicalDrm]

-> Moved to Programs, Apps and Websites

***

 

As said, this depends heavily on software. I work mainly with GIS where fast storage and RAM are mostly at use. Lately while working on my BA on land survey engineering I had point clouds, meshing and other tools at hand (using 3DReshaper from Leica/Hexagon). That was mainly stressing CPU and RAM, CPU when calculating triangulations for point clouds and RAM when loading up models.

 

[ronasux]

Most CAD software are single threaded so get the best CPU you can afford from this list. You'll need a good GPU if you're going to start sculpting/organic modeling like in Zbrush, 3D-Coat, etc..

 

https://www.cpubenchmark.net/singleThread.html

 

Edit: 16GB of RAM are "OK". I would recommend 32GB as a starting base but it all depends on what you're modeling, poly count, etc. so you might find 64GB is a better choice for you.

[LogicalDrm]

17 hours ago, ronasux said:

Most CAD software are single threaded ...

 

That seems rather odd if its true. I don't have licenses to AutoCAD or any other more common ones to test. Microstation uses all of my cores and threads when testing with old school project.

 

I know Surpac utilizes all cores, and its currently owned by same company that owns SolidWorks. Don't know how much that plays into things, but I would find it odd that one specialized software (Surpac is for mining industry) utilizes multicore and rest of the catalog somehow doesn't.

[Gastje88]

On 5/5/2020 at 2:37 PM, LogicalDrm said:

-> Moved to Programs, Apps and Websites

***

 

As said, this depends heavily on software. I work mainly with GIS where fast storage and RAM are mostly at use. Lately while working on my BA on land survey engineering I had point clouds, meshing and other tools at hand (using 3DReshaper from Leica/Hexagon). That was mainly stressing CPU and RAM, CPU when calculating triangulations for point clouds and RAM when loading up models.

 

Thanks for all the replys, and the link is very useful in choosing the right cpu then. 

Of course more RAM is better with these kind of programs. 

but thanks to confirm my concerns. I'm a step closer to configuring my set-up. 

 

Kind regards

[ronasux]

3 hours ago, LogicalDrm said:

 

That seems rather odd if its true. I don't have licenses to AutoCAD or any other more common ones to test. Microstation uses all of my cores and threads when testing with old school project.

 

I know Surpac utilizes all cores, and its currently owned by same company that owns SolidWorks. Don't know how much that plays into things, but I would find it odd that one specialized software (Surpac is for mining industry) utilizes multicore and rest of the catalog somehow doesn't.

I've never heard of the CAD programs you mentioned as I'm in the product development space but Autocad, Solidworks, and Rhinoceros are all single threaded software so higher clock speeds + fastest RAM you can afford will be the best way to go. Every year I've had this fight on their forums advocating for multi-thread support.

[Gastje88]

53 minutes ago, ronasux said:

I've never heard of the CAD programs you mentioned as I'm in the product development space but Autocad, Solidworks, and Rhinoceros are all single threaded software so higher clock speeds + fastest RAM you can afford will be the best way to go. Every year I've had this fight on their forums advocating for multi-thread support.

The way I was told once is that the solving is done single threaded is because the solution of one solve can have consequences for the next one, so when the tasks are spread over multiple threads you can have calculations that have unknown inputs. 

In my opinion this should be possible to do for parts in an assembly that don't have any external references, but somehow they haven't figured that out yet. 

[ronasux]

1 hour ago, Gastje88 said:

The way I was told once is that the solving is done single threaded is because the solution of one solve can have consequences for the next one, so when the tasks are spread over multiple threads you can have calculations that have unknown inputs. 

In my opinion this should be possible to do for parts in an assembly that don't have any external references, but somehow they haven't figured that out yet. 

That makes sense and I'm not a programmer so this is over my head but considering how other software work, I don't see why they can't use multi-cores/threads to share the calculations. For instance, Vray rendering can spread out the rendering to other nodes to speed up the process.

[shadow_ray]

14 hours ago, ronasux said:

That makes sense and I'm not a programmer so this is over my head but considering how other software work, I don't see why they can't use multi-cores/threads to share the calculations. For instance, Vray rendering can spread out the rendering to other nodes to speed up the process.

Exactly what @Gastje88 said. Some tasks cannot be parallelized. Three wives wont make someone a grandpa 3 times faster.

V-ray or ray tracing in the other hand is embarrassingly parallel. The scene is not modified during rendering, the program only reads the data, so all the cores can work together without disturbing each other. Rendering an animation can be done in a way where each frame rendered on a different machine.

 

[ronasux]

6 hours ago, shadow_ray said:

Exactly what @Gastje88 said. Some tasks cannot be parallelized. Three wives wont make someone a grandpa 3 times faster.

V-ray or ray tracing in the other hand is embarrassingly parallel. The scene is not modified during rendering, the program only reads the data, so all the cores can work together without disturbing each other. Rendering an animation can be done in a way where each frame rendered on a different machine.

 

Maverick render is making use of GPU parallelization. I'm currently using two GTX 980 Ti's (non-SLI) to render a single frame. I'm not sure how they're doing it but it's working. Besides, for us CAD designers complaining about multi-thread support, we're looking to speed up operations that can be sped up. We get that if there's 3 intersecting objects being booleaned and one boolean affects the other that it will take time as it needs to do them one after the other but there's no reason why other cores can't solve non-intersecting or booleans unrelated to the first set. To help you visualize what I mean, imagine you're trying to create a cogwheel from a donut and you have 20 cuts to make... In my case, I have around 300 cuts to make in a single boolean op, they're being cut from the main body but they don't affect each other so why the wait? If I export the donut and the cutters to an organic modeling program like 3DCoat, the boolean op takes a fraction of a second!

[shadow_ray]

1 hour ago, ronasux said:

Maverick render is making use of GPU parallelization. I'm currently using two GTX 980 Ti's (non-SLI) to render a single frame. I'm not sure how they're doing it but it's working. Besides, for us CAD designers complaining about multi-thread support, we're looking to speed up operations that can be sped up. We get that if there's 3 intersecting objects being booleaned and one boolean affects the other that it will take time as it needs to do them one after the other but there's no reason why other cores can't solve non-intersecting or booleans unrelated to the first set. To help you visualize what I mean, imagine you're trying to create a cogwheel from a donut and you have 20 cuts to make... In my case, I have around 300 cuts to make in a single boolean op, they're being cut from the main body but they don't affect each other so why the wait? If I export the donut and the cutters to an organic modeling program like 3DCoat, the boolean op takes a fraction of a second!

There are two things i can think of.  The devs don't want to rewrite and redesign their software and / or the representation of a surface is different. CAD applications tend to use parametric surfaces (because of precision?)  while polygon based modelling is more popular in graphics focused things, actually 3ds max even crashed several times when tired too use a certain NURBS tool. If the exported model is tessellated that would explain it. I'm really bad at math so i have no idea if it's a plausible explanation or not.

If i said something stupid please do not spear me. I don't want to spread stupidity, there are already too many of us do it.

[ronasux]

3 hours ago, shadow_ray said:

There are two things i can think of.  The devs don't want to rewrite and redesign their software and / or the representation of a surface is different. CAD applications tend to use parametric surfaces (because of precision?)  while polygon based modelling is more popular in graphics focused things, actually 3ds max even crashed several times when tired too use a certain NURBS tool. If the exported model is tessellated that would explain it. I'm really bad at math so i have no idea if it's a plausible explanation or not.

If i said something stupid please do not spear me. I don't want to spread stupidity, there are already too many of us do it.

I think you're spot on as that's how I understand those things as well. With McNeel (developer of Rhinoceros) the issue is definitely them unwilling to put in the effort it takes to rewrite things. There's a huge discussion for Linux compatibility where they claimed that Linux isn't being pursued because they don't see themselves making enough money to cover the cost of development. Someone asked them how they decided to create the MacOS version and they said some guy started working on it and they picked it up from him. Their statements don't add up no matter how I looked at it but it gives you an idea of their work mentality.