[cwl]

Hi everyone

Can anyone give some recommendations/feedback on a videocard for Star-CCM?

I'm asking this - because i still don't understand which one would provide decent framerate in scene (or 3D-CAD module mode) containing assembly consisting of about a hundred of CAD objects. Currently with my GTX 650 Ti (sounds old, but ...) performance is slow and frustrating - although the same assembly can be viewed and edited without any freezes in SpaceClaim, for example.

Another part is rendering (Resampled Volumes, opacity, Advanced Rendering) - is it CPU-based (since there is a "Client Render Threads" setting) or is it GPU-based?
If it works through CPU - then is there actually any benifit in having a powerful videocard?

Any comment/advice/feedback would be appreciated since I'm pretty confused.

[ashokac7]

Originally Posted by cwl

Hi everyone

Can anyone give some recommendations/feedback on a videocard for Star-CCM?

I'm asking this - because i still don't understand which one would provide decent framerate in scene (or 3D-CAD module mode) containing assembly consisting of about a hundred of CAD objects. Currently with my GTX 650 Ti (sounds old, but ...) performance is slow and frustrating - although the same assembly can be viewed and edited without any freezes in SpaceClaim, for example.

Another part is rendering (Resampled Volumes, opacity, Advanced Rendering) - is it CPU-based (since there is a "Client Render Threads" setting) or is it GPU-based?
If it works through CPU - then is there actually any benifit in having a powerful videocard?

Any comment/advice/feedback would be appreciated since I'm pretty confused.

There is section in Steve Portal and it is mentioned that Star CCM does use GPU for rendering. Check for the setting in Star CCM if it is using dedicated GPU for rendering.
May be this can be a limitation of Star CCM to handled such a large number of CAD objects.
In CCM, get into Tools>>Options>>Visualization
Select Unmanaged and see if there is any improvement.
Unmanaged option let graphics driver allocate GPU for the task based on usage.

If you have budget, you can go for better card.

[cwl]

Originally Posted by ashokac7

There is section in Steve Portal and it is mentioned that Star CCM does use GPU for rendering.
In CCM, get into Tools>>Options>>Visualization
Select Unmanaged and see if there is any improvement.
Unmanaged option let graphics driver allocate GPU for the task based on usage.

In my case I observed no difference between Default/Unmanaged/Opportunistic.

Again - what's then the "Client Render Threads"? - Which can be set to number of cores max, so it points to CPU.

If you have budget, you can go for better card.

Sure. But the question is what is a sufficient better card? - GTX 1070, GTX 2080 or should I get Quadro?

PS I realize that hardware requirements depend on the complexity of the simulation, so I'm just trying to figure out some kind of milestone/benchmark ...

[m0j0jojo]

Originally Posted by cwl

In my case I observed no difference between Default/Unmanaged/Opportunistic.

Again - what's then the "Client Render Threads"? - Which can be set to number of cores max, so it points to CPU.



Sure. But the question is what is a sufficient better card? - GTX 1070, GTX 2080 or should I get Quadro?

PS I realize that hardware requirements depend on the complexity of the simulation, so I'm just trying to figure out some kind of milestone/benchmark ...

I'm using a GTX780 Ti paired with a i7 4770k, and its butter smooth...But for simulations i would put my focus on a decent cpu, Starccm+ isnt really graphic intensive.

[cwl]

Originally Posted by m0j0jojo

I'm using a GTX780 Ti paired with a i7 4770k, and its butter smooth...But for simulations i would put my focus on a decent cpu, Starccm+ isnt really graphic intensive.

I wonder, would it be also smooth - if you create a scene with:
- Scalar Displayer on Resampled Volume with semi-transparend ColorBar (like "blue-red highpass");
- Geometry Displayer with semi-transparent surfaces;
- set Transparency Mode of the Scene to Depth Peeling instead of default value of Alpha-Blending.

Because in my case this reduces fps almost to zero on 2 x Xeon E5-2683 v3 with Client Render Threads set to 28 (max cores), so I'm pretty confused about rendering.

[ashokac7]

Don't get RTX 2080 or GTX 1070 foe CAD workload. The Cuda architecture of Nvdia is not good CAD workload. CAD load favor OpenGL architecture. If you are on tight budget then get AMD RX 580 8GB or Vega 56 or Vega 64 if available. AMD RX 5700 or 5700XT is also good budget choice.

For more budget we can get a Quadro card. Quadro p5000 is very good or you can settle for p4000 as well. you can even get newer RTX quadro cards but I think they are pricey.

You can check your graphics performance by,

on scene section in simulation, right click and test the graphics. Post the output window here.
It should look like
"OpenGL supported, direct rendering, OpenGL hardware/Os based rendering detected."

Hope you have turned off hyper threading on your CPUs.
As you know multiple cores doesn't help in carrying volume rendering. Volume rendering needs single CPU and a GPU. (Check if you are using monitor with cable plugged into graphics card)
I too don't have an idea about client render threads. It is set to 12 on my 12 core desktop.
I do work with Resampled volume, and it works perfectly on my Quadro m5000.

[cwl]

Originally Posted by ashokac7

Don't get RTX 2080 or GTX 1070 foe CAD workload. The Cuda architecture of Nvdia is not good CAD workload. CAD load favor OpenGL architecture. If you are on tight budget then get AMD RX 580 8GB or Vega 56 or Vega 64 if available. AMD RX 5700 or 5700XT is also good budget choice.

Hm .. I might be missing something, but I do not understand how is Quadro better than RTX 2080 Ti - since according to specifications it looks more powerful:

                   Quadro P4000   RTX 2080 Ti
GPU Memory         8 Gb GDDR5     11 Gb GDDR5
Memory Interface   256 bit        256 bit
CUDA Cores	   1792           4352
OpenGL Version     4.5            4.6

You can check your graphics performance by, on scene section in simulation, right click and test the graphics. Post the output window here.
It should look like "OpenGL supported, direct rendering, OpenGL hardware/Os based rendering detected."

I did it on desktop (GTX 650 Ti), cluster node (GT 720) and laptop (Intel HD 520) - can't see any difference in output that would actually allow to tell which is fast and which is slow; except for list of OpenGL extensions of course.

Desktop:

***Water Droplets Track [View = Iso] Report***
OpenGL:  supported
rendering:  direct
library:  native
OpenGL vendor string:  NVIDIA Corporation
OpenGL renderer string:  GeForce GTX 650 Ti/PCIe/SSE2
OpenGL version string:  4.6.0 NVIDIA 391.35
OpenGL extensions:  ...
pixel format:  8
class:  TrueColor
buffer size:  32
level:  0
renderType:  rgba
double buffer:  True
stereo:  False
rgba:  redSize=8greenSize=8blueSize=8alphaSize=8
depth size:  24
stencil size:  0
aux buffers:  4
accum:  redSize=16greenSize=16blueSize=16alphaSize=16
***End Water Droplets Track [View = Iso] Report***

Cluster Node:

***Water Droplets Track [View = Iso] Report***
OpenGL:  supported
rendering:  direct
library:  native
OpenGL vendor string:  NVIDIA Corporation
OpenGL renderer string:  GeForce GT 720/PCIe/SSE2
OpenGL version string:  4.5.0 NVIDIA 385.69
OpenGL extensions:  ...
pixel format:  8
class:  TrueColor
buffer size:  32
level:  0
renderType:  rgba
double buffer:  True
stereo:  False
rgba:  redSize=8greenSize=8blueSize=8alphaSize=8
depth size:  24
stencil size:  0
aux buffers:  4
accum:  redSize=16greenSize=16blueSize=16alphaSize=16
***End Water Droplets Track [View = Iso] Report***

Laptop:

***Scene Report***
OpenGL:  supported
rendering:  direct
library:  native
OpenGL vendor string:  Intel
OpenGL renderer string:  Intel(R) HD Graphics 520
OpenGL version string:  4.6.0 - Build 26.20.100.6913
OpenGL extensions:  ...
pixel format:  5
class:  TrueColor
buffer size:  32
level:  0
renderType:  rgba
double buffer:  True
stereo:  False
rgba:  redSize=8greenSize=8blueSize=8alphaSize=8
depth size:  24
stencil size:  8
aux buffers:  0
accum:  redSize=16greenSize=16blueSize=16alphaSize=16
***End Scene Report***

Hope you have turned off hyper threading on your CPUs.
As you know multiple cores doesn't help in carrying volume rendering. Volume rendering needs single CPU and a GPU. (Check if you are using monitor with cable plugged into graphics card)

Thank you, sure I'm aware of this :>

I too don't have an idea about client render threads. It is set to 12 on my 12 core desktop.

And here we are both confused

I do work with Resampled volume, and it works perfectly on my Quadro m5000.

Can you by any chance set Transparency Mode of the Scene to Depth Peeling instead of default value of Alpha-Blending? - If that won't reduce fps - it would be a good sign.
Unless I just expect too much from hardware - like real-time rendering.

[ashokac7]

There is a benchmark video on Youtube (By Gamer Nexus) for workstation graphics cards. There are some optimizations done workstation cards which largely influence the CAD performance.

If you are confused then if there is return window period available, buy RTX 2080 Ti, try it in you system, if it works then good or return it.

Otherwise it is better to go ahead with Nvdia RTX 5000. Price is more, but you will not miss on workstation GPU optimizations.

[ashokac7]

I tried Depth peeling instead of alpha blending. There is some stutter. May be because of reduction in FPS.

Again it will depend on mesh size as well I guess.

May be you are expecting too much from the hardware.

[LuckyTran]

I would not recommend dumping a lot of $ into a graphics card if displaying a part is the issue you want resolved.

The rendering is slow because you have too many objects to keep track of. A lot of small things will add up. Not saying that better hardware will not help, but it won't be as helpful as you might like. A lot of advances are needed in hardware and libraries for rendering large assemblies.

Originally Posted by cwl

I did it on desktop (GTX 650 Ti), cluster node (GT 720) and laptop (Intel HD 520) - can't see any difference in output that would actually allow to tell which is fast and which is slow; except for list of OpenGL extensions of course.

It should say OpenGL: supported and rendering: direct (which it does). That means it is being rendered on the graphics card using OpenGL.

[cwl]

Originally Posted by LuckyTran

I would not recommend dumping a lot of $ into a graphics card if displaying a part is the issue you want resolved.

The rendering is slow because you have too many objects to keep track of. A lot of small things will add up. Not saying that better hardware will not help, but it won't be as helpful as you might like.

Thank you so much!

A lot of advances are needed in hardware and libraries for rendering large assemblies.

What exactly do you mean?

It should say OpenGL: supported and rendering: direct (which it does). That means it is being rendered on the graphics card using OpenGL.

Makes sense, but "Client Render Threads" setting is still very confusing - since it points to the CPU threads.

[LuckyTran]

The state-of-the-art is to render (mostly) on GPU using OpenGL. Rendering 1 part works nicely and you would think that it scales nicely to 10 parts (which it sort of does). But you have to keep in mind that going from 1 to 10 parts is an order of magnitude increase in number of parts and you need to raise this by some exponent for problem complexity (number surfaces, lines, vertexes that need rendering and rays that need to be traced). And 1 to 100 parts is two order of magnitude increase raised by some exponent. Any new graphics card you get, likely won't be 2 orders of magnitude increase in performance. And that basically summarizes the problem.

I've seen time and time again, people with the same frustration sink $ into quad-SLI GFX setups only for the model to still render slowly.

A lot of advances in GFX card hardware performance and libraries for rendering a large number of objects are needed before we'll be able to render large CAD assemblies in real time.

You might wonder why the better GFX cards over the past decades have not improved CAD rendering and ask how did people render large assemblies before? Well, hardware has gotten better. But a lot of the increased performance has gone into increasing resolution (i.e. from 720p, to 1080p, and soon 4K and after that 8K) and that's where it's all going.

If someone wanted to spend some money and get an RTX 2070 or 2080 (or whatever equivalent server-class GFX card you want, or whatever the ATI equivalent is) that's okay in my opinion. These are great cards. But I would not buy four 2080's and expect CAD to be rendered any faster. If you already have an RTX 2080, I'd suggest to stick with it.

[cwl]

Originally Posted by LuckyTran

The state-of-the-art is to render (mostly) on GPU using OpenGL. Rendering 1 part works nicely and you would think that it scales nicely to 10 parts (which it sort of does). But you have to keep in mind that going from 1 to 10 parts is an order of magnitude increase in number of parts and you need to raise this by some exponent for problem complexity (number surfaces, lines, vertexes that need rendering and rays that need to be traced). And 1 to 100 parts is two order of magnitude increase raised by some exponent. Any new graphics card you get, likely won't be 2 orders of magnitude increase in performance. And that basically summarizes the problem.

I've seen time and time again, people with the same frustration sink $ into quad-SLI GFX setups only for the model to still render slowly.

A lot of advances in GFX card hardware performance and libraries for rendering a large number of objects are needed before we'll be able to render large CAD assemblies in real time.

You might wonder why the better GFX cards over the past decades have not improved CAD rendering and ask how did people render large assemblies before? Well, hardware has gotten better. But a lot of the increased performance has gone into increasing resolution (i.e. from 720p, to 1080p, and soon 4K and after that 8K) and that's where it's all going.

If someone wanted to spend some money and get an RTX 2070 or 2080 (or whatever equivalent server-class GFX card you want, or whatever the ATI equivalent is) that's okay in my opinion. These are great cards. But I would not buy four 2080's and expect CAD to be rendered any faster. If you already have an RTX 2080, I'd suggest to stick with it.

I competely agree to all this - but there is quite easier fact which leads to explanation: same huge assembly works like frozen in Star 3D-CAD (even without any complicated rendering), but at the same time on the same hardware (any from the list I've mentioned) it can be manipulated smoothly in SpaceClaim or SolidWorks regardless the CAD core (Parasolid etc).
So I'd just say - that Star's graphical performance is a disaster.

[LuckyTran]

Solidworks is also super easy to break.I remember when I tried to CAD RL10 rocket nozzle not too long ago...

It rendered nicely without cooling channels. It rendered nicely with 1 cooling channel. It renderered nicely when I patterned the cooling channel 10 times. It completely broke when I changed the circular pattern setting from 10 to 100. This is a single part, not even an assembly. The only difference, was the # of repeats in a circular pattern!