Zoo seems to be a new CAD company that built their own software stack. In this blog post they give a technical overview of their approach. From the article: … …
Zoo seems to be a new CAD company that built their own software stack. In this blog post they give a technical overview of their approach. From the article:
It turns out, many of the geometric operations needed by CAD algorithms lend themselves naturally to the massively parallel nature of GPU computing. Furthermore, it turns out that the techniques needed to render the CAD display are identical to the techniques used by 3D video game engines to render the game world. Additionally, the low-level mechanics of managing compute shaders to do complex work on the GPU are fundamentally the same between CAD software, 3D modeling software, and game engine software. While over the last several decades, gaming has pushed the state of the art in 3D graphics rendering alongside GPU computing, CAD software has not.
Most existing CAD programs today allow you to quickly create more content than they can efficiently render in a few steps. […]
…
We aim to address this problem by maintaining a minimal set of geometric primitives. Rather than implementing specialized representations for features like fillets, we represent these as B-splines with specific constraints. This way, operations only need to handle B-splines, allowing us to develop highly optimized, GPU-friendly algorithms like our surface-surface intersection method. By avoiding the combinatorial explosion of primitive interactions, we can achieve significant performance improvements over existing CAD systems.
…
Traditional SSI [Surface to Surface Intersection] methods are CPU based, ie, they are recursive, sequential and conditional. We aim to develop an approach that makes use of the power of modern GPUs. These tend to be brute force with simple instructions and massive data. Benefits of using GPU for SSI, include speed, increased efficiency and handling higher levels of detail.
Sounds great. My exposure to CAD software has been limited and mostly comes from sketching out models to 3D print and toying with furniture design ideas, but even at that level the lack of...
Sounds great. My exposure to CAD software has been limited and mostly comes from sketching out models to 3D print and toying with furniture design ideas, but even at that level the lack of competition, poor performance, and platform-locked nature of currently available options is strongly felt. Anything with the potential to shake up the market has my applause.
It seems likely that their approach will work well with commodity GPUs which is great too, though I’m sure Nvidia and AMD won’t be happy to lose those workstation card sales if this takes off.
I think Nvidia will do just fine supplying hardware to people using AI. :-) Incidentally, this new CAD system seems to be AI-enabled. They are giving the product away for free with a limited...
I think Nvidia will do just fine supplying hardware to people using AI. :-)
Incidentally, this new CAD system seems to be AI-enabled. They are giving the product away for free with a limited number of queries.
Free with the caveat that they will use your data/designs to train their AI. I've known AI driven CAD was coming, but was under the impression that we were further away due to the computational...
Free with the caveat that they will use your data/designs to train their AI. I've known AI driven CAD was coming, but was under the impression that we were further away due to the computational power required (and also that much of the more complex designs/training opportunities will be under proprietary lock and key). It sounds like this was either designed specifically with AI in mind or someone realized that their direction (optimization) could maximize investment/profit by integrating AI/data harvesting.
Not completely in the loop on whether any other big players in the CAD world have started to integrate AI, but this engine level optimization may be what's limiting them at the moment.
Yeah, business users will want to pay for a plan. But I've previously used OnShape with their free plan, which has the limitation that all your designs are public. As a hobbyist, it seemed like a...
Yeah, business users will want to pay for a plan. But I've previously used OnShape with their free plan, which has the limitation that all your designs are public. As a hobbyist, it seemed like a good deal, since I would probably release them as open source anyway.
I tried out the tutorial. Interestingly, it's text-based like OpenScad. There is a main.kcl file that includes the others in the project directory. Software developers will feel at home. (They...
I tried out the tutorial. Interestingly, it's text-based like OpenScad. There is a main.kcl file that includes the others in the project directory. Software developers will feel at home. (They claim you can do everything in the UI and don't need to touch the code.) Much like OpenScad, it's parametric, but doesn't directly support constraints (yet).
Since AI is text-based, I guess that makes sense.
Switching between text files and watching it re-render, it didn't seem all that snappy on my Mac Mini M2. There also isn't a progress indication telling you when it's done. Looks like it actually runs in the cloud:
Zoo Design Studio requires an internet connection because processing and our geometry engine run in the cloud. This lets you handle complex CAD workflows smoothly, even on low-powered machines.
They have a test version that runs in a browser, but don't recommend it.
It doesn't seem as polished as OnShape, but perhaps it will improve.
Zoo seems to be a new CAD company that built their own software stack. In this blog post they give a technical overview of their approach. From the article:
…
…
Sounds great. My exposure to CAD software has been limited and mostly comes from sketching out models to 3D print and toying with furniture design ideas, but even at that level the lack of competition, poor performance, and platform-locked nature of currently available options is strongly felt. Anything with the potential to shake up the market has my applause.
It seems likely that their approach will work well with commodity GPUs which is great too, though I’m sure Nvidia and AMD won’t be happy to lose those workstation card sales if this takes off.
I think Nvidia will do just fine supplying hardware to people using AI. :-)
Incidentally, this new CAD system seems to be AI-enabled. They are giving the product away for free with a limited number of queries.
Free with the caveat that they will use your data/designs to train their AI. I've known AI driven CAD was coming, but was under the impression that we were further away due to the computational power required (and also that much of the more complex designs/training opportunities will be under proprietary lock and key). It sounds like this was either designed specifically with AI in mind or someone realized that their direction (optimization) could maximize investment/profit by integrating AI/data harvesting.
Not completely in the loop on whether any other big players in the CAD world have started to integrate AI, but this engine level optimization may be what's limiting them at the moment.
Yeah, business users will want to pay for a plan. But I've previously used OnShape with their free plan, which has the limitation that all your designs are public. As a hobbyist, it seemed like a good deal, since I would probably release them as open source anyway.
I tried out the tutorial. Interestingly, it's text-based like OpenScad. There is a main.kcl file that includes the others in the project directory. Software developers will feel at home. (They claim you can do everything in the UI and don't need to touch the code.) Much like OpenScad, it's parametric, but doesn't directly support constraints (yet).
Since AI is text-based, I guess that makes sense.
Switching between text files and watching it re-render, it didn't seem all that snappy on my Mac Mini M2. There also isn't a progress indication telling you when it's done. Looks like it actually runs in the cloud:
They have a test version that runs in a browser, but don't recommend it.
It doesn't seem as polished as OnShape, but perhaps it will improve.