Jump to content
Jim Wilson

Hardware Upgrade Information

Recommended Posts

A regular line of questioning I see here; users who want to upgrade their hardware, but want to make sure they get the best bang for the buck. Since hardware changes so rapidly, any specific recommendations I make will quickly become outdated and unhelpful.

Because of that, instead of recommending specific hardware, in the below article I outline various components of a computer and what those components do as far as affecting Vectorworks performance directly. The intention is to give you an idea of which bits will get you the kind of improvements you are looking for.

Since the explanations can be a bit complex (and I can be a bit long-winged ;) ) I have included a short answer in green for each category for those who just want a brief answer, and then the long explanation beneath it. If there is anything that I have left out or if I have contradicted anything I may have said in the past, please let me know and I will modify the article accordingly. I hope this helps!
 

 

Share this post


Link to post

Good article Jim!

 

You might consider adding "Do not buy a computer with an integrated graphics chip, period." :D This seems to be one of the largest sources of confusion/graphics issues on the forums.

 

Kevin

 

Share this post


Link to post
13 minutes ago, Kevin McAllister said:

You might consider adding "Do not buy a computer with an integrated graphics chip, period." :D This seems to be one of the largest sources of confusion/graphics issues on the forums.


Very good point, doing so now.

Share this post


Link to post

Great info! I was just wondering if I should upgrade.

 

But (you knew that was coming) I'm a bit confused by the terminology in your post vs that in Cinebench that I just downloaded and ran.  No GPU number unless you mean the OGL fps

From my system:

image.thumb.png.4629d912a90aabf4dd7c293ad2534b65.png

From your post

image.thumb.png.be3281015e99f5b033efbcc0b805bd8b.png

If this is all right then my system is 7ish times faster rendering but about twice as fast in basic navigation. Therefore it would rank low in your "good" range unless I was doing mostly rendering work.

 

I also have had to turn my graphics compatibility preferences down to get certain functions to show properly. From your discussion I'm thinking this is VRAM related rather than speed related.

 

Is that how to use this info?

Share this post


Link to post

Very useful topic. Thanks a lot, Jim!

 

But sometimes I'm a little bit confused about the ongoing process and the dedicated hardware part.

 

For Example some days ago I tried an entry Level iMac Pro (the 8 core model, with 8 GB graphic card and 1 TB SSD). I tried to open an 850 MB file (an image is attached) to try the speed working improvement rendering etc. Actually, i work with the iMac you can see on firm, and trust me the improvement is absolutely not relevant (except for rendering).

 

To create a set of tree: I waited for 3 minutes the process's ending but at the end, i did a force software quit. I noticed the CPU charge was always at 100% level (on a maximum 1600%, like a Cinema Render process).

 

So my question is: having understood what improvements are to be considered in the hardware, are all software processes capable of exploiting its potential? Thanks a lot for your advice

 

Zeno

 

 

Tavola-1-Assonometria_A.jpg

Schermata 2018-01-08 alle 10.40.25.png

Schermata 2018-01-08 alle 10.40.29.png

Share this post


Link to post
Posted (edited)

Hoo,

your untitled-2 needs a lot of RAM.

If there are just the plants in showing on screen, there goes something definitely wrong !

 

EDIT :

Sorry that can be one of the other files opened of course.

Edited by zoomer

Share this post


Link to post
20 minutes ago, zoomer said:

Hoo,

your untitled-2 needs a lot of RAM.

If there are just the plants in showing on screen, there goes something definitely wrong !

 

EDIT :

Sorry that can be one of the other files opened of course.

 

I confirm that the only active process was just the plants. The other files were only opened.

Share this post


Link to post
10 hours ago, RickR said:

No GPU number unless you mean the OGL fps

 Correct!

 

4 hours ago, Zeno said:

So my question is: having understood what improvements are to be considered in the hardware, are all software processes capable of exploiting its potential?

Not all of them currently, no. Object duplication specifically is single core and wont be made faster with more CPU cores or with a faster GPU. In many cases, even a faster core can not even improve the speed, as the delays are software based and not a limitation of your hardware. If the limitation is the number of objects Vectorworks can handle at once, and not just that your VRAM is maxed out, then additional VRAM will not help. More of these limitations are removed with each Vectorworks revision however, so this type of advice becomes more accurate as versions progress.

Share this post


Link to post
3 minutes ago, JimW said:

Not all of them currently, no. Object duplication specifically is single core and wont be made faster with more CPU cores or with a faster GPU. In many cases, even a faster core can not even improve the speed, as the delays are software based and not a limitation of your hardware. If the limitation is the number of objects Vectorworks can handle at once, and not just that your VRAM is maxed out, then additional VRAM will not help. More of these limitations are removed with each Vectorworks revision however, so this type of advice becomes more accurate as versions progress.

 

Thank you very much, Jim. 

 

I think that it could be a very important issue for the next versions-release. Every CPU core is virtually doubled (correct?), and if those limitations persist, peoples can use only a half core on a machine with 4, 8 until 18 core. I can't imagine the global improvement if all commands will be built for multi-core processes (that means up to 36x more faster for a duplication? Really?).  Is it a list where we can know which commands are "single-virtual-core" programmed? That could be very useful for me, i will change my iMac in a few months, and actually, i use VW 24/7 for my work.

 

Thanks a  lot, one more time. 

Share this post


Link to post
13 minutes ago, Zeno said:

Every CPU core is virtually doubled (correct?), and if those limitations persist, peoples can use only a half core on a machine with 4, 8 until 18 core. I can't imagine the global improvement if all commands will be built for multi-core processes (that means up to 36x more faster for a duplication? Really?).


Agreed, it is one of the areas that takes the greatest amount of time to change but also yields the greatest result. It will keep happening gradually over future versions, it isn't something that will be done all at once. There are a lot of threads on this forum where it has been discussed in depth.

Share this post


Link to post
9 minutes ago, JimW said:

There are a lot of threads on this forum where it has been discussed in depth.

 

I'm so sorry. Could you direct me on one of the latest?

Share this post


Link to post
33 minutes ago, Zeno said:

Is it a list where we can know which commands are "single-virtual-core" programmed?


No official list unfortunately. However the general list at the moment:

Actions that include duplicating, moving, calculating solids geometry, cut/add/intersect surface, and tool loading are all single core activities. Meaning that having more cores will not make them any faster. While TECHNICALLY a faster CPU core will make them faster, even comparing a 2Ghz core to a 5Ghz core yeilds hardly any difference (sub 1-2% in my tests), so the difference is academic.

Rendering in Renderworks modes or Hidden Line is multi core, other than the "Geometry" phase at the beginning, which is basically taking inventory of everything to be included in the render before the multi core processes begin.

Rather than make processes like drawing plan graphics multi core on the CPU, they were moved over to more specialized GPU processes instead. This effectively made them multi core, but not in a way you would see by looking at a CPU performance manager, since theyd show up on your GPU now instead.

Share this post


Link to post

Hi Jim

Brilliant information in this feed for understanding what Vectorworks takes use of from the machine, will Vectorworks be ever able to use things more along the lines of a combined hybrid CPU and GPU for processing even basic things to speed up the process or have you ever thought about using specific things on each platform like Metal 2 (probably be Metal 3 before long) on Mac OS to speed up Vectorworks using direct access to GPU (take a look at affinity designer and affinity photo they are brilliant apps designed using Metal to speed up certain processes with live previews throughout) Overall Vectorworks is fantastic and I don't think I know many other packages that can work with a 10 floor building as easy but I just feel sometimes that Vectorworks needs that bit of a speed increase when working on large projects.

Thank you,

Share this post


Link to post
On 2/3/2018 at 9:00 AM, TSG-Sim said:

will Vectorworks be ever able to use things more along the lines of a combined hybrid CPU and GPU for processing even basic things to speed up the process or have you ever thought about using specific things on each platform like Metal 2 (probably be Metal 3 before long) on Mac OS to speed up Vectorworks using direct access to GPU (take a look at affinity designer and affinity photo they are brilliant apps designed using Metal to speed up certain processes with live previews throughout)


Engineering evaluates things like hybrid CPU/GPUs and Metal as they come, making the call on whether it makes sense to transition to them or to stick with an existing tech, or to utilize a tech that works cross platform. I do not see any immediate plans for a large technology shift, but work is ongoing to increase use of multiple CPU cores and dedicated GPUs for more elements of Vectorworks' underlying tech and its UI.

Share this post


Link to post

How predictive of Vectorworks performance are the Cinebench OpenGL numbers? The ancient 2GB HD6950 I have at work scores 82 while at home my 8GB RX480 scores 113. This is a 38% increase which seems like a pretty modest difference for two cards that have nearly six years between them. Does Cinebench simply not have a use for the full 8GB on my 480?

Share this post


Link to post
On 3/20/2018 at 9:03 PM, WesR said:

How predictive of Vectorworks performance are the Cinebench OpenGL numbers? The ancient 2GB HD6950 I have at work scores 82 while at home my 8GB RX480 scores 113. This is a 38% increase which seems like a pretty modest difference for two cards that have nearly six years between them. Does Cinebench simply not have a use for the full 8GB on my 480?

They're mostly predictive for OpenGL performance (yes, I know this is stating the obvious) and some non-rendering things do depend on the graphics card but the majority of things depend on a single core of the CPU and non-openGL renders also depend on the CPU. So it will give an idea of display performance but not necessarily of overall Vectorworks performance because some of the initial display performance also depends on the CPU single core speed.

 

Cinebench also has tests for the CPU (single and multicore performance) and you should look at those as well for the single core performance as that may give you a better idea of overall performance for non-rendering tasks and for the multicore performance for non-openGL rendering tasks.

 

Then it becomes a challenge of choosing the best combination of CPU, GPU and some other things for the best performance/cost ratio if you don't have an unlimited budget.

Share this post


Link to post
On 3/21/2018 at 6:29 PM, Art V said:

 the majority of things depend on a single core of the CPU 

 

 

 

I am not sure this is the case anymore.  @JimW has to lead us to believe that most of the processing is being done by the GPU.   Renderworks is still CPU - but I think almost everything else - openGL and top/plan is GPU

Share this post


Link to post
2 minutes ago, Tom Klaber said:

 

I am not sure this is the case anymore.  @JimW has to lead us to believe that most of the processing is being done by the GPU.   Renderworks is still CPU - but I think almost everything else - openGL and top/plan is GPU

 

The single core speed is the bottleneck for math/geometry calculations and some other items. For example Jim has specifically said some of the selection/highlighting is still done by a single CPU core. Also some of us had assumed that the upgrade to Top/Plan included working on sheet layers which it doesn't.

 

Kevin

Share this post


Link to post
On 3/24/2018 at 10:45 AM, Kevin McAllister said:

The single core speed is the bottleneck for math/geometry calculations and some other items. For example Jim has specifically said some of the selection/highlighting is still done by a single CPU core.


^ The above is still true for the majority of "operations" type actions, like duplications, additions, subtractions and move actions. More of these types of actions will become multi core in the future, but this is the slowest and most work-heavy part of switching everything from single threaded operations to multiple ones.
 

On 3/24/2018 at 10:36 AM, Tom Klaber said:

I am not sure this is the case anymore.  @JimW has to lead us to believe that most of the processing is being done by the GPU.   Renderworks is still CPU - but I think almost everything else - openGL and top/plan is GPU


^ This is true for viewing objects and zooming/panning around a file, which was a significant portion of how slow Vectorworks actually FEELS for most users even when they aren't doing the types of actions listed above. 

The statements I have made in the past will start to become less true or more muddled as time moves on and various portions of Vectorworks are upgraded (For instance, there may come a version where all Move actions have become multicore but Add Solids or Subtract Surface are not yet), however an absolute truth that can be taken to heart: Vectorworks will continue to rely more and more on the GPU as time goes on, and upgrades where we split calculations into separate threads on the CPU will appear to get faster and faster, even on much older CPUs.

Share this post


Link to post

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

 

  • 7150 Riverwood Drive
  • Columbia, Maryland 21046, USA
  • Contact Us | 410.290.5114

 

  • © 2018 Vectorworks, Inc. All rights reserved.
  • Vectorworks, Inc. is part of the Nemetschek Group.
  • ×