How fast is your Inventor PC really?

[Raider_71]

Hi guys,

We have had to do some testing on a bunch of Inventor PC's recently to determine which of the PC's needs to be replaced. Obviously we needed to find out which of the PC's are the worst eprformers as there was only budget to replace 50% of the design PC's. So we thought the Darwin theory will come in handy right... 🙂

Anyways I started searching on the net for toppics on how to benchmark an Inventor PC. Then I thought whats the point of using gaming benchmark tools because Inventor is not a game and there are more aspects than just graphics performance when it comes to percieved performance on an Inventor PC right.

So we decided to create our own Inventor benchmark tool which tests various aspects of an Inventor PC to give us an overview of our PC's performance. This then helped us make a decision as to whcih pc's to replace.

We have made the tool available free of charge to anyone interested in checking how their PC stacks up to their peers or friends. 🙂

Please download it here and post your results here as well if you want. Would be interesting to see what beast workstations are out there.

I would like to say thanks to Kirk #karthur1, for helping in testing the app.

Please feel free to send any suggestions our way. There is an email link in the app.

Download and Install

The application will work with Inventor 2014 to 2016 only.

IMPORTANT: After installation there will be an Inventor Bench icon on your desktop that looks like this: 

My resluts:

HP Elitebook 8560w with an SSD upgrade.

[brotherkennyh]

The assembly at the link below might be a good test candidate;

http://brotherkennyh.co.uk/content/knex-big-ball-factory

It only uses a small number of unique components so the package size isn't huge. There is a decent amount of geometry on the individual parts.

I know this model does get laggy on decent systems, although I'm not sure it will push the best systems. I can make a larger version if it would help.

An assembly like this would help with performance testing. The main assembly is broken into a number of sub assemblies, which could be unsuppressed one after the other by the bench tool, allowing you to produce a performance curve. You could also use a model like this to find your systems bottleneck.

[Neil_Cross]

@brotherkennyh no no the testing is done with a single card in the system.  So I install the Quadro M4000 for example on it's own as a normal person doing normal work would have, boot it up, install the driver, run the FPS tests.  Power off, take that out, put the next card in the same slot obviously, boot up, install the drivers, repeat.

My tests have been done on a both a single part and an assembly with 1100 unique parts containing 3000+ occurences, and to be honest the story has been consistent across anything I've thrown at it.  In my mind that's a good balance of testing for visuals.

To be perfectly honest though I'm 99% satisfied that a line has been drawn under this now, I'll probably be going through this all over again when my Ryzen 1700X CPU comes but I have no reason to believe I won't see the same thing happen with that... unless AMD architecture somehow handles things differently to Intel.

No matter what tests are done there'll always be doubt in someones mind, but after having banged 6 massively different GPU's all at different performance and price points into a single system and proving no graphical gains or losses, what more can you do.

[mmaes]

@Neil_Crossso your theory is that if a guy say has a great CPU and daily works in assemblies with say 60,000 parts he will see no gains in using a GTX1080 over a $150 4gb GTX1050?  I want to be sure I am understanding your theory.

[Neil_Cross]

---

@mmaes wrote:

@Neil_Crossso your theory is that if a guy say has a great CPU and daily works in assemblies with say 60,000 parts he will see no gains in using a GTX1080 over a $150 4gb GTX1050?  I want to be sure I am understanding your theory.

---

Well a theory is something based off ideas and opinions, what I'm saying is fact based off evidence presented conclusively in my video.  It's right there for everybody to see.

60,000 parts isn't the same story though, with that many parts you're not only going to hit other hardware based bottlenecks but probably be hit by poorly optimised areas of the software too.

I don't have a 60,000 part assembly to test all my cards on though, happy to receive one if someone is able to share and show it publicly

[machiel.veldkamp]

Well... I have an assembly that I'd love to see running on a heavy PC like yours @Neil_Cross

I'm gonna go and have a chat with my boss tomorrow and see if we can work something out. 

It's all intelectual property however so don't expect much. (The model I'm talking about is about 20gb's worth of CAD data including a full 3D scan of the building. )

[mmaes]

@Neil_CrossI guess I say theory because one part or one small assembly isn't really a thorough test in my mind.  I would need to see how the different cards perform in 10,000 to 50,000 part assemblies (in increments) to be 100% sold on the idea that all* cards are made equal when it comes to Inventor.

I was recently working on the below model (130,000+) parts and originally had the GTX960 installed.  I took the 960 out and replaced with a GTX1080 and saw a huge increase in performance.  Initially (GTX960) half the parts would drop out any time I tried to pan or zoom...after switching to the GTX1080 I had a lot better ability to see the model as I would pan around.  The only thing that changed was the card.

I'm not arguing or attempting to discredit anything, I just feel like a more thorough test with large assemblies would tell the tale and possibly help determine which card is best for a particular users application.

[Neil_Cross]

Where do I stop with that though, in my mind a single part up to 1100 parts was a good range, if I then test 100,000 parts then someone with a 200,000 part assembly could then pipe up and argue the tests aren't conclusive enough.  There's no way to prove it but in my experience travelling around in a VAR, the majority of Inventor users tended to work on datasets within the range of 1-1000 parts and anything around 1000 parts is generally considered to be a large assembly, what you've got is colossal!

Just speculating but with a dataset that large I would imagine your 960 was capping out immediately on VRAM.  If you look at that car assembly I have, look at the GPU-Z window, with only that open my system is using around 1.3-1.5GB of VRAM, and your GTX960 only has 2GB, so it'll have been paging off to system RAM causing a massive bottleneck.  That is definitely a case of a new GPU making a difference but for my tests I was trying to keep within the limits of available VRAM to let actual GPU acceleration be a thing if it was going to be a thing.

If I can get my hands on something in the hundreds of thousands of parts range I'll look to test that, I just have a feeling the GPU is going to play second fiddle to almost everything else with something of that size but I'll be definitely keen to look into it.

[mmaes]

@Neil_Cross, Agreed!  I don't believe 100,000+ parts is realistic for day to day use but in my work place each engineer is daily working in assemblies that consist of 20-30,000 parts. 

I guess its hard to say what a good size assembly is for testing.  For me personally, 1,000 is not a good test...but to each their own.

Also, the GTX960 I have is the 4gb version.

[Neil_Cross]

Did you ever monitor the VRAM whilst you had the 960? I've managed to cap out 3.5GB whilst using 3 monitors with a few applications open and a couple of large assemblies, but I was trying to cap it out at the time.  

[Anonymous]

@Neil_Cross @brotherkennyh

Due to Solidworks, I have to use pro card anyway. Thoughts?

Here are my specs:

CPU: Intel 7700K

Mem: 32GB DDR4; 2400MHZ

GPU: AMD FirePro W7100

Drive: Samsung 960 PRO 512GB 

Thanks,
Arjun

[mmaes]

@Neil_Cross, I've never really payed much attention to the VRAM but here is a short video of that large model with the 6700k and GTX960

[video]

[Neil_Cross]

You need to set the video to unlisted instead of private

[mmaes]

sorry, I don't use youtube.  It should be fixed

[Neil_Cross]

Yea it looks like you were having VRAM issues there, your GPU load was always under 30% but the VRAM was capped at around 3800+MB which was probably as much as it would go on a 4GB card.

Here's me right now, 2k parts and 10k occurences, using 4.5GB of VRAM if you can even see it, the quality of images on this website is dreadful.

I don't even have much open, Outlook Chrome and and Inventor are all that is going on here and it's at 4.5GB.

Sounds like a good test for me though, if I open this up on a Quadro 2000 with it's 1GB of VRAM... how much degradation is there and how unstable does Inventor become.

[brotherkennyh]

Can I ask for those of you with insanely large assemblies, are you working with shrink wraps? Or are all 130,000 parts active in the graphics window?

@mmaes Nice assembly by the way.

@Anonymous I am not as familiar with Solidworks, been a while since I have used it. Sound like you need the more expensive card more for the features than the performance. I would be interested to hear about how SW performs with large assemblies like the ones mentioned here. Nothing like a bit of competition to motivate Autodesk.

Your W7100 beats my K2200 hands down  and I am not able to overwork that yet, with 3000+ parts. No idea if SW will experience the same. Of course it will depend on each use case. Number of parts is only a rough measure. 3000 boxes is easy. 3000 detailed engine blocks is not.

I have been begging our IT department to get my a PC like yours and am keeping my fingers crossed I will be upgrading soon.

For our use I know the CPU needs to be looked at, so I will be looking for a new CPU in the top 10 of the single thread charts. Hopefully a 7700K, the graphics card will be determined by the remaining budget.

I have to say though that I will be going for a card that is unnessesary for our Inventor needs, because I will also be doing some VR visualisations

[machiel.veldkamp]

We don't work with shrink wraps anymore. The function is too buggy and slow to use in real life. 

[Anonymous]

---

@brotherkennyh wrote:

 I would be interested to hear about how SW performs with large assemblies like the ones mentioned here. Nothing like a bit of competition to motivate Autodesk.

This i can help with! give me a couple of days though.

[brotherkennyh]

Hi @machiel.veldkamp,

Slightly off topic, but...

I disagree. Yes the shrink wraps are buggy and slow. However, they are worth the time to set up. One of our own engineers told me they were a waste of time, I proved him wrong with his own assembly.

You cannot simply hit a button to shrink-wrap an assembly. You need to consider sub assembly structure and shrink wrap lower level sub assemblies, then have level of detail reps to activate the shrink-wraps. It takes some time to setup effectively. On an assembly the size of mmaes' I would probably end up spending a day just applying shrink-wraps. Still, if you have to work much in the top level assemblies, the performance improvement is worth the time.

With our own assemblies the shrink wrap also often creates errors with the shape of the part, but unless you are working directly with those components you can ignore it, enjoy the performance benefit and return to the master LOD later. I have a LOD in most master assemblies which switches the sub assemblies between master and shrink wrap state.

[machiel.veldkamp]

I agree that it functions on a dead assembly and for viewing after working on the parts it's great I believe. 

However, working on an assembly while maintaining it's Substitute is just a pain in the ass. 

The substitute just wont update if you make changes to the original part or assembly. With some tricks it will but in general... it wont update.

For that reason alone, I will not implement it in my workflow unless Autodesk does a complete overhaul on the whole substitute thing. 

[mmaes]

I agree that shrink wrap can be a good option but for what we do it just doesn't work for us.

The model above was used for rendering purposes and will be on our website, in our brochure, and displayed at our trade show.  I don't know if you have ever noticed but the colors of parts seem to get jumbled when shrink wrapping.  For that reason alone it wasn't an option for this particular project.