Will generative design ever get a local compute feature?

[rkalak]

I'm really contemplating purchasing a commercial license, solely for generative design capabilities, and didn't want to have to pay for cloud credits IN ADDITION to paying for the license. Will generative ever get a local compute feature so that I can utilize my workstation and not have to pay for cloud credits? I currently don't have commercial because I'm a student.

[mavigogun]

No.    A search for Generative Design on this forum will yield detailed explanations, as this question has been confronted repeatedly.

[rkalak]

Well, I guess I'm switching over to Siemens Solid Edge in terms of generative. Free for students and support of local compute. 

[mavigogun]

---

@rkalak wrote:

Free for students and support of local compute. 

---

"Local compute" on what hardware?    Practically, aren't only institutions or those with the means to justify their own processor array going to be able exploit that local computation?   That doesn't sound like an Acceptable solution to me.

[mvassilevJSN2W]

---

@mavigogun wrote:

---

@rkalak wrote:

Free for students and support of local compute. 

---

"Local compute" on what hardware?    Practically, aren't only institutions or those with the means to justify their own processor array going to be able exploit that local computation?   That doesn't sound like an Acceptable solution to me.

---

Not necessarily - the solution is sound. I can see people using local machines despite not being optimized for Generative Design and waiting the many hours if there is no rush. I think the users should have the option to choose rather than being forced. Also Amazon Cloud Services has serious compute options that are cheaper than the $25/25cr per study. Perhaps charging for compute time seems more reasonable rather than a flat fee - some studies will be cheaper and others maybe more expensive depending on the complexity.

[mavigogun]

---

@mvassilevJSN2W wrote:

---

@mavigogun wrote:

Not necessarily - the solution is sound. I can see people using local machines despite not being optimized for Generative Design and waiting the many hours if there is no rush. I think the users should have the option to choose rather than being forced. Also Amazon Cloud Services has serious compute options that are cheaper than the $25/25cr per study. Perhaps charging for compute time seems more reasonable rather than a flat fee - some studies will be cheaper and others maybe more expensive depending on the complexity.

---

How have you quantified ANY of that?    When you say "hours", what is that- tens, hundreds, thousands?   And what is your basis?    The server farm being used for Fusion's generative design IS Amazon's- how can you say how much retaining that service would be without knowing the processing time required?    From whence  has the proclamation sprang?

[rkalak]

@mavigogun If local compute wasn't feasible, why would the addition to the software exist. From what I know, It solely utilizes the processor, so Threadripper cpus would be ideal. There are a couple videos out there showing the process, and calculation timings are completely acceptable considering the final result. Various quality levels can also be set, making compute times shorter for those with less processing power. Oh, and it isn't $100 to download a model; very ideal for startups or experimenters with workstations. 

[mvassilevJSN2W]

---

@mavigogun wrote:

---

@mvassilevJSN2W wrote:

---

@mavigogun wrote:

Not necessarily - the solution is sound. I can see people using local machines despite not being optimized for Generative Design and waiting the many hours if there is no rush. I think the users should have the option to choose rather than being forced. Also Amazon Cloud Services has serious compute options that are cheaper than the $25/25cr per study. Perhaps charging for compute time seems more reasonable rather than a flat fee - some studies will be cheaper and others maybe more expensive depending on the complexity.

---

How have you quantified ANY of that?    When you say "hours", what is that- tens, hundreds, thousands?   And what is your basis?    The server farm being used for Fusion's generative design IS Amazon's- how can you say how much retaining that service would be without knowing the processing time required?    From whence  has the proclamation sprang?

---

This is where I'm trying to go with this, and I'm glad you're asking the right questions. We the users don't know what the processing time is. Many of us would like some type of a benchmark for a particular study case study under different criteria to see what it actually takes to run the simulations rather than " just take our word, it has to be done this way". Any information on this I think would bring the community closer. I also think if this proves to be in the hundreds or thousands of hours than it's obviously a no go, but we don't know and would like to. Do you know where we can get more information about the compute necessary or any other details? Or point us in the right direction.

[mavigogun]

---

@rkalak wrote:

@mavigogun If local compute wasn't feasible, why would the addition to the software exist.

Feasible for who using what- exactly? Have you read the build cost for machines that could crunch these numbers in a reasonable amount of time? How many tens-of-thousands are you willing to spend? Not enough, I'll hazard to guess, based on the discomfort caused by Autodesk's fee structure. I'm sure EVERYBODY would love to hear the specs and time taken by a test case- have you ever seen any such data? Or are these statements more along the lines of "Godzilla could totally take Mecha Godzilla"?

[mavigogun]

There's no point in speculating: I'll just download the trial and measure the time required.

[mvassilevJSN2W]

---

@mavigogun wrote:

---

@rkalak wrote:

@mavigogun If local compute wasn't feasible, why would the addition to the software exist.

Feasible for who using what- exactly? Have you read the build cost for machines that could crunch these numbers in a reasonable amount of time? How many tens-of-thousands are you willing to spend? Not enough, I'll hazard to guess, based on the discomfort caused by Autodesk's fee structure. I'm sure EVERYBODY would love to hear the specs and time taken by a test case- have you ever seen any such data? Or are these statements more along the lines of "Godzilla could totally take Mecha Godzilla"?

---

Where can we read about the build cost of these machines? And can you define what reasonable time means? I ran two simulations and they solved fast - less than 2 min - which was awesome using the fancy servers on the cloud. I certainly wouldn't be disappointed if a case study took 12 hours to finish on one of my workstations at work. 

Another glossed over argument many have is that 100cr starting might not be enough to learn the software well enough in order justify using it in the future. Not sure how that aspect can be made better.

[mavigogun]

---

@mvassilevJSN2W wrote:

Another glossed over argument many have is that 100cr starting might not be enough to learn the software well enough in order justify using it in the future. Not sure how that aspect can be made better.

---

Most all of the area knowledge necessary is encompassed by the Simulation Workspace; once you understand and are able to produce a useful simulation of the forces your part will be subjected to, the set up for the generative process is relatively straight forward- this coming from my limited experience with simulation and a day-workshop on GD at Autodesk's Dallas campus.

I reckon an actionable, high quality simulation requires a mind equipped with a solid Engineering background.

[mvassilevJSN2W]

---

@mavigogun wrote:

---

@mvassilevJSN2W wrote:

Another glossed over argument many have is that 100cr starting might not be enough to learn the software well enough in order justify using it in the future. Not sure how that aspect can be made better.

---

Most all of the area knowledge necessary is encompassed by the Simulation Workspace; once you understand and are able to produce a useful simulation of the forces your part will be subjected to, the set up for the generative process is relatively straight forward- this coming from my limited experience with simulation and a day-workshop on GD at Autodesk's Dallas campus.

I reckon an actionably, high quality simulation requires a mind equipped with a solid Engineering background.

---

I agree. There is a lot of useful information, and well done videos with Lars about setting up environments. I'm a ME with 10 yrs experience in design and simulation and found myself wanting to play and try things because that's what engineers do I guess. After setting up my Generative Design and simulation I was dissatisfied with the results I received due in part of being too ambiguous with some of the constraints. It doesn't leave room for experimentation. I haven't convinced myself yet that I would produce something useful in 4 tries after burning my first 2 immediately.

[rkalak]

---

@mavigogun wrote:

There's no point in speculating: I'll just download the trial and measure the time required.

---

If you are going to run a test, could you let me know if Solid Edge utilizes all CPU cores? As of now, I don't think Solid Edge has a GPU compute option. I want to finalize my purchase of a Ryzen TR 2950x or i9 9900k. The 9900k is faster in terms of single core and quad-core performance, but the TR is faster in multi-core.

[mavigogun]

Solid Edge Pro- the version required to unlock fully functional GD -is $329.... per month.   For $3550 per year, you could run quite a few simulations on the cloud using Fusion.

[mvassilevJSN2W]

---

@mavigogun wrote:

Solid Edge Pro- the version required to unlock fully functional GD -is $329.... per month.   For $3550 per year, you could run quite a few simulations on the cloud using Fusion.

---

Around ~10 simulations including model export. It's a wash

[rkalak]

But, I don't need it for the whole year. And Fusion commercial is 500 per year. With Solid Edge I could always run as many studies as I want in the one-month time frame because of local compute. And the extra cost of processing power in my to be built PC shant count since I'll be using it for several other tasks. Preference of fusion over solid edge solely depends on the user.

[rkalak]

@mvassilevJSN2W

A "wash" in which direction?

---

@mvassilevJSN2W wrote:

---

@mavigogun wrote:

Solid Edge Pro- the version required to unlock fully functional GD -is $329.... per month.   For $3550 per year, you could run quite a few simulations on the cloud using Fusion.

---

Around ~10 simulations including model export. It's a wash

---

[mvassilevJSN2W]

---

@rkalak wrote:

@mvassilevJSN2W

A "wash" in which direction?

---

@mvassilevJSN2W wrote:

---

@mavigogun wrote:

Solid Edge Pro- the version required to unlock fully functional GD -is $329.... per month.   For $3550 per year, you could run quite a few simulations on the cloud using Fusion.

---

Around ~10 simulations including model export. It's a wash

---

---

I was trying to point out that Fusion's Generative Design is not superior based on the monthly cost of Solid Edge. Personally I'd much rather spend $350 to have as many simulations of generative design as I want in 1 month assuming it's not many tens of hours/simulation on a decent workstation. This weekend I'll download the free trial and play with it and see how long something takes to simulate. I get frustrated when I keep seeing autodesk people post how long simulations take and how optimized their servers are for generative design forcing you to use their cloud service etc etc, without any actual evidence. If it does take a tremendous amount of time so be it, no big deal. I wish they'd prove it in some meaningful way.

[Anonymous · Accepted solution]

There's not a single simulation that fusion 360 offers that would require a computer in the tens of thousands. You dont really even need to spend more than 3K for a decent simulation machine now that hardware has gotten so cost effective. 

I'm a huge simulation user, and a big fan of the simulation module in fusion 360. But the cloud credit thing is a business model, not a hardware or software limitation. And those of us in the simulation world dont normally blink an eye at a few hour long solve times. I've had several cfd simulations run into the 24 to 48 hour range for a complex enough analysis.

I do wish fusion would offer a bank of cloud credits to test and train. I have been doing a lot of topology optimization in ANSYS recently and it's taken several trys to truly understand how I want to approach different problems. And I've been using that software for a decade. I would not be happy if I had to spend $100 just to figure out the approach to solving a problem.