Jump to content

Optimizing workstation upgrades


Recommended Posts

I work for a small consulting and production firm which works in the field of lighting and video for architecture, and the arts. We have three multipurpose workstations which are coming up due for an upgrade. About half of their time is spent using Vectorworks 2018, and the other half is dominated by Adobe products, and running the Disguise (formerly D3) designer application. There are a several key points where we would like to improve performance in VW, and I would like to understand better the hardware factors which will help us achieve that.

 

The largest issue that we have on a daily basis is navigation performance where very large PDFs are used as backgrounds. Our most common task is to extensively mark these up, or draw out 3D geometry based on these plans. The time taken to redraw the images on screen is painfully long, and makes fluid, efficient navigation difficult. Related to this, we often experience long waits before redraw in 3D views, when new symbols are inserted, especially ones with other nested symbols. Fluidly navigating through IFC files, often exported from Revit and Archicad, is also pretty painful. It performs better when we turn off portions of these models, as one would expect, but our work often involves the evaluation of facade lighting conditions, so this this not as practical. Our primary 3D navigation tools are 3D connexion mice, so long redraw times become especially jarring, as the fluidity these devices provide is broken.

 

So as I see it, there are two dimensions of choice in both CPU, and GPU choices for these machines, which is what I am seeking advice on, given these particular goals.

 

CPU:

Scale out or scale up?

Are these issues related to deficits in single, lightly multi-threaded, or heavily multi-threaded performance? Where does the VW roadmap point? Are high clocks, and high IPC going to be the biggest factors over the next several years, or is the plan to break up these draw tasks across many cores?

 

AMD or Intel?

Are their specific instruction sets being used which favor one architecture, or is VW largely platform agnostic?

 

GPU

Gaming performance or general compute?

Should we be looking at gaming benchmarks for determining suitability? How well do these DX11/Vulkan tests translate to OpenGL performance? Should we be concerned about raw FLOPS for future releases over looking at current gaming results?

 

AMD or Nvidia?

Based on the other workloads, especially Disguise, we are generally leaning towards AMD (As GCN based GPUs are also used on the actual media servers), however if there is a significant performance differential in VW, its needs would take precedence.

 

General Questions

Is there a limit, imposed by how VW is written, to how smooth we can make this kind of navigation? Is this a situation where no matter what kind of hardware we throw at the problem, we are not going to achieve consistently smooth navigation with this kind of content? What other hardware factors may be at play?

 

For reference, our current systems.

 

1. i7-5930K@4.3ghz, GTX-1080, 8x8GB@2666 DDR4, Samsung 850 Boot, and storage, 240mm AIO

2. i7-5930K@4.3ghz, GTX-980, 8x8GB@2666 DDR4, Samsung 850 Boot, and storage, 240mm AIO

3. i7-6800K, GTX-980, 2x16GB@2666 DDR4, Samsung 850 Boot, and storage, Hyper 212

 

I look forward to hearing the communities thoughts. Thanks!

Edited by Peter A. Milo
Link to comment

CPU

VW CAD :

fastest single core / Turbo Boost as possible (often means less cores)

CPU Rendering (VW RW, C4D, Modo, VRAY, ...) :

as many and fast cores/hyperthreading as possible (often means less GHz)

Both :

find a sweet spot according to priority

 

AMD or Intel?

As the bang for the buck, AMD (Ryzen or Threadripper)

Overall CPU Rendering Performance (currently Threadripper 32 core)

Overall CAD single core performance - still Intel

Look at Cinebench results.

 

GPU

For VW, mostly Gaming performance

look at gaming benchmarks.

How well do these DX11/Vulkan tests translate to OpenGL performance?

Important for now is OpenGL only. But faster DX/Vulkan cards may be faster in OpenGL too.

Pro cards are more expensive, hardware may be cherry picked and more stable but mostly

offer more VRAM. (which is important as your Models, now and in the future need to fit in)

But VW does not make use of special Pro driver's features and often runs better with the

standard gaming drivers.

So for VW, NVidia gaming cards may be the easiest choice.

 

AMD or Nvidia?

I think currently NVidia still offers a bit more Performance and needs less power.

Also it offers CUDA, often used in GPU Renderers (VRAY, Octane, ...) and most times

more VRAM, which may make it for now the better choice.

That may change in the future with AMD's Vega roadmap and smaller transistors when

they get power consumption down.

 

General Questions

how smooth we can make this kind of navigation?

View Navigation in VW will profit from a fast OpenGL GPU with enough VRAM and a

fast single core CPU for loading Views into GPU.

In VW there are still some bottlenecks or certain things not GPU hardware accelerated

as they could. You should wait these 4 weeks to see if VW 2019 changes may influence

hardware decisions in any way.

 

 

 

But I think your current main problems are these complex PDFs :

 

1. Try to deactivate PDF Snapping option.

This should eliminate one bottleneck in View Navigation.

(PDF aren't as accurate as a CAD File though, like 124.0 m may be 123.999973652 m)

You can use a Grid Snap as a workaround.

 

2.

PDFs get converted into a pixel image for the GPU at a certain screen resolution

so they need some time to display. Then you often can rotate or drag fluidly.

But at least as soon as you zoom your View, it will need to recalculate,

which results in lags. (Other Apps may offer PDF caching to work around)

 

So you can try to export your PDFs to PNG raster images at a suitable resolution

and import these instead of the PDF. (Of course this means no snap option)

It worked very well for me in an other App without PDF caching,

although the image was ridiculous high res. I have 6 GB of VRAM here.

So basically there is no more need to dynamically recalculate the PDF for View

changes, it is just that the Image and Geometry need to fit into your GPU.

 

I converted my PDFs in Apple's Preview App so I don't know if Acrobat Reader will

do this on Windows but I think there are even options to convert from VW itself (?)

You have to test a bit with export resolutions to find something suitable,

to get the smallest possible file size.

 

So that's worth a try. Maybe that even avoids all View refresh lags and makes

viewports fluid again.

 

 

3.

For the IFCs there isn't much to do :

They often lost their CAD Symbol instancing and store every part separately.

(While instances of repeating elements would also accelerate OpenGL Views

and geometry calculation for RW renders)

What you can do is to ungroup all IFC Entities to get native VW elements

with less nesting. In the best case there come in Extrudes or Generic Solids

which speed things up, in the worts case there will be Meshes which aren't

very optimized in any CAD.

 

4.

Generally avoid Meshes and prefer VW (Solids) geometry.

Especially Library items often contain complicated Mesh geometry

at a detail level you may not need. It may be worth to rebuild (parts of)

them by better VW geometry.

 

5.

Symbols :

Not nested may be better in general than nested Symbols,

but I think it is better to have nested Symbols instead of more data by redundant

content copies because of not nesting and reusing.

Generally the use of Symbols in VW is a very good idea for any objects that

appear a few times in the file.

But I have no clue about how complex Spotlight (Light) Symbols behave

and how to best deal with those kind of objects.

 

 

Edited by zoomer
  • Like 2
Link to comment

Thanks for this thoughtful and detailed information. A couple of thoughts and questions.

 

CPU:

We may have to specialize our machines a little bit. Maybe move two of the machines over to 8086K's, and move the 3rd to a TR4 based system for our non-VW rendering needs. That being said, with the the way VW is written, are we going to see the 20% gains that the clock speed increase would suggest between Zen and Coffee Lake, or are there significant diminishing returns?

 

GPU:

We had not been considering pro GPUs, as they provide no advantages in our other workloads. The trouble we have been having with assessing GPU performance is the availability of Windows based OpenGL benchmarks. I am very familiar with the Linux based benchmarks, as I refer to these frequently when building video playback systems (which use HAP-Q, so GPU performance is relevant), however I do know that AMD's drivers have a disadvantage in certain kernel's which is not reflected in windows. I have seen conflicting information about how well there cards perform in Windows OpenGL tests. Some software tests show Nivida cards handily defeating AMD cards, other show the opposite. What are some specific benchmarking applications, which use OpenGL, and whose results are ranked separately from DX11 tests, whose test regime is similar to the VW workload?

 

Other notes:

 

1, 2

 

2 hours ago, zoomer said:

Try to deactivate PDF Snapping option. 

 

2 hours ago, zoomer said:

So you can try to export your PDFs to PNG raster images at a suitable resolution

and import these instead of the PDF. (Of course this means no snap option

 

I think a little more information about our use case is needed. The summary of it is, that because of who generates these PDFs, how they are generated, and what we need to do with them, we need to find the hardware to handle the workload, rather than try to cause other firms to change their processes, or disable the tools we need. There are political complexities that I would rather not get into here related to when and how we get brought into projects. The main way we receive drawings to be annotated is being given pre-formatted PDFs, which do contain vector information, rather than being rasterized, and that we are expected to return for direct attachment to a drawing set, rather than being incorporated into the source drawings and models. This is why we don't often ask for CAD files anymore, as we often can not style these to look exactly like the directly exported PDFs of BIM applications. So ultimately, if the answer is, "you need to buy a phase change cooler and overclock to 5.5ghz", then this is a better option politically and financially for all parties involved. Snapping and doing work on top of giant PDFs is the main thing we do.

 

EDIT: The above largely relates to our architectural work. For our arts work, the PDF's we receive are often the only remaining drawings of a space or site, produced in a long dead version of autocad, whose source files were created by an entity which no longer exists, or who is no longer under contract with the client.

 

3, 4

Our main use of IFC's is for internal design discussions, and client conceptual flythoughs, and as such these suggestions are greatly appreciated. As noted above, these models are not useful for us as a method to output work product, so we can absolutely afford to lose any contained information. This is also why we are primarily concerned with display fluidity above all else. Again, thanks for the tips.

 

5

Understood. We don't have many issues with spotlight. I guess we'll re-evaluate our processes when after we perform the hardware upgrade, and after VW2019.

 

Thanks

 

 

Edited by Peter A. Milo
Link to comment
2 hours ago, csmilo said:

and move the 3rd to a TR4 based system for our non-VW rendering needs.

 

CPU :

I think there was a misunderstanding.

CAD work (single core, or a few at best) will run best with Intel so far.

Rendering (multi core), a AMD TR4 based CPU may be faster (for the same prize)

 

So a TR4 board better for Rendering.

 

GPU :

Not much choice here on the Mac side.

Currently AMD "Pro" GPU sonly anyway.

 

When I build a PC for Windows, or more likely Linux (sorry VW),

I think NVidia would be better overall (faster, less energy, CUDA option, ....)

But I prefer AMD as a company, their Open Source usage (for Linux ?) and

their cards itself look much better, so I would go AMD nevertheless.

Maybe even a Pro GPU like WX 9100 or similar to get more VRAM

 

But look for Benchmarks in the www for your specific OS, API and App.

I saw large differences for different Graphic APIs vs different OS vs different Apps.

Like C4D seems to always run 20% better on AMD, while Modo seems to always have

been 5x faster on Nvidia opposed to AMD.

I read that for AMD on Linux, some Apps run better with Mesa, some with AMD drivers.

 

Benchmarks :

For VW (+Modo + Blender) CPU + GPU I think the best Benchmark is still Cinebench.

CPU single + CPU multi + OpenGL.

It worked well for me.

If a new machine has a double CineMultiBench I knew my Modo or C4D Renders will

need only half the time.

Also I trust the Blender Rendertimes used everywhere as a comparison.

(Blender has just released an own Benchmark in Beta)

 

 

2 hours ago, csmilo said:

I think a little more information about our use case is needed.

 

I may still not really understand your work and workflows.

I didn't meant you to ask your clients for better sources.

 

But if you get any PDFs, as an alternative, try to save them as raster images before

inserting into VW. Just for VW, to see it work is more fluid.

Maybe their is a way to combine your work at the end with the PDF sources again.

No option though, if you want or need to use PDF Snapping option.

 

You can also convert PDFs to DWG like from Illustrator (?) or explode them in VW.

But I think I would not be happy with that mess but maybe worth a try too.

Would allow snapping and doesn't need recalculation for view changes.

Maybe explode in an extra VW file and reference only.

 

But if you look for an option to keep the current workflow by throwing better hardware

on it, I fear the bottleneck is VW and benefits won't be linear with hardware speed.

 

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...