Al Dean explains why Autodesk running generative design in the cloud is wrong

On Twitter,  DEVELOP3D co-founder Al Dean explains why it is a mistake for Autodesk to limit running its generative design software on a cloud service: 

So here’s the thing with this. Autodesk’s Generative Design tools run in the cloud, its rationalization being that it needs parallel computation and it needs collaboration.

The issue is that this is going to restrict experimentation -- experimentation that’s vital in both an industrial and an educational context. (It’s unlikely to be made available in Fusion Educational licenses.)

There’s also the fact that Autodesk say they need to recoup the costs of that computation. That’s entirely their decision.

Autodesk’s Generative Design tools run on GPUs, specifically CUDA. GPUs are cheap. Folks could stack out a machine and do the computation on their own hardware -- as they do now for rendering and viz[ulization].

My personal belief is that it’s a huge error on Autodesk’s part, and one that I doubt they’ll correct any time soon. </rant>

As a reference point, all other CAD vendors to offer generative design do so via desktop software, proving that no cloud is needed. Generative design from Autodesk is available through Fusion 360 Ultimate. https://www.autodesk.com/products/fusion-360/compare#generative-design

Comments

[Steve Johnson]

Al's right. It's not only going to discourage people from experimenting with designs once they've established that workflow, it's also going to discourage people from doing the experimentation required to establish that workflow in the first place.

[Orlando Sardaro]

Hi Al,

Why would computing in the cloud restrict experimentation? If you can get a fast calculation, you can only experiment more.

And have you seen http://unlimited.solidthinking.com/inspire/?

Thanks, Orlando

[Ralph Grabowski]

Why would computing in the cloud
restrict experimentation?
Autodesk restricts use of generating design to (a) people who subscribe
to Fusion 360 and (b) those who pay extra for tokens. This means very few
have access to it.

If you can get a fast
calculation, you can only experiment more.
Even though it runs on the cloud, it is generally unavailable, and so few
people can experiment with it. Even then, every run costs tokens, which
discourages experimentation.

And have you seen

http://unlimited.solidthinking.com/inspire/?
We first wrote about it in 2009. See

http://worldcadaccess.typepad.com/blog/2009/11/bonebased-architectural-design.html

Regards,

Ralph H. Grabowski
Publisher

upFront.eZine > newsletter >

www.upfrontezine.com
WorldCAD Access > daily blog >

www.worldcadaccess.com

eBooks.onLine > learning portal >

www.worldcadaccess.com/ebooksonline
Twitter > @upfrontezine >

https://twitter.com/upFronteZine

[Tom Foster]

Generative Design is precisely the kind of instant-on-demand facility that's maybe returning to the desktop according to this from http://www.upfrontezine.com/2018/04/upf-978.html:

"Not that Gartner is often right, but they feel that cloud computing is as much of the future as, oh, let's say the Pony Express, to be replaced by edge computing: "Around 10% of enterprise-generated data is created and processed outside a ...cloud. By 2022, Gartner predicts this figure will reach 50 percent.

"The market for so-called micro-modular data centers is growing rapidly" because customers can "run enterprise-grade IT in close proximity to their operational technology (OT) environments, machines and equipment to enable low-latency, secure, and reliable digital processes." gartner.com/smarterwithgartner/"

Roll on Distributed - in all its forms!

[Orlando Sardaro]

Ah I agree, Ralph! Paying per calculation makes one cautious. I too like solutions better that just make you pay per period or perpetual.

And yes solidThinking Inspire has been around for a while (actually Altair is busy developing this technology since the 80-ies) but is now available in the cloud, is why I put it up there.

Cheers, Orlando

[Darren Young]

I've always noticed that people can waste 10x as much as they spend before anyone notices. It's more difficult to see waste than cash spent. The important point here is people see spending and will make choices based on it.

With a "pay to play" business model, people will pay when then "need to" but not when they "should" or "could". This is the benefit of perpetual licensing, it's like a physical asset that you have....as long as it's there, people will make use of it rather than let it sit idle. "As long as we have it, let's use it, it'll only cost time".

With rental services, they are only used when needed, because any other use not only costs time, but additional for access. This is essentially why experimentation and access is limited.

Perpetual licenses, like physical assets, are typically cost justified with a purpose in mind. Any additional use beyond the cost justification is added value potential. In a rental economy, this added value is removed.

[Matthew Stachoni]

"Autodesk’s Generative Design tools run on GPUs, specifically CUDA. GPUs are cheap. Folks could stack out a machine and do the computation on their own hardware -- as they do now for rendering and viz[ulization]."

GPUs are NOT that cheap. Perhaps they are slightly less expensive than CPUs when you compare them on a transistor-by-transistor basis (an AMD 1950X has 9.6B transistors at $780, whereas a GTX 1080Ti has 12B transistors for $800). However, the idea that it's trivial to stuff half a dozen high-end GPUs in a workstation is ludicrous, as is that such a system would compare at all to the hundreds or thousands of GPU resources in the cloud.

The number of GPUs you can put in a system is tied directly to the CPU you have. A typical Intel desktop CPU has 16 PCIe lanes meaning you can have one GPU running at 16x, or at most at 8x speed. High-End Desktop CPUs like the Core i9 and Threadripper provide 24, 44, or even up to 64 PCIe lanes, but some of those are dedicated to PCIex4 lanes for M.2 SSDs and the like. And those Core i9s aren't cheap either, with the price of admission around $1,000 for the CPU alone. All told you might get up to 3 or 4 GPUs in a high-end box.

Above that, you could go all out and get 8 Quadro P6000 GPUs in an Nvidia Visual Computing Appliance (VCA), which at $6,500 a pop would set you back at least around $60,000.

Even if you build out a GPU farm on your network using a Backburner like technology you need more than a few high-end PCs that aren't doing anything else.

Compare that huge outlay to get suboptimum performance to the vastly better performance available in cloud computing, where you have just the rather petty inconvenience of paying for such software and services (horrors!)

[Rahul Kalakuntla]

Siemens Solid Edge is actually very friendly in terms of generative design. I currently use it as a student for generative design as opposed to fusion 360. The UI is very similar to Inventor and Solidworks, and it is very easy to import and convert parts and assemblies from other software while retaining all modifiable functionality. Based on models I have examined, the quality between Fusion generative and Solid Edge generative is very similar. Solid Edge is also run locally for all computation, making it all the better.

[Christopher LeFay]

"Solid Edge is also run locally for all computation, making it all the better."

I'm skeptical. Provide an example with computation times, Rahul Kalakuntla. If Siemens has some how managed to radically lower the processing requirements, they've got a bigger product than just generative design.