CA Stress Testing Results

[architect]

Looks like all that is needed is a 10ghz single core machine (and a fire extinguisher)……

My guess is a similar type model tested on Revit/Archicad would have similar delay times, etc ?.

 

Eric

[TheKitchenAbode]

22 minutes ago, architect said:

My guess is a similar type model tested on Revit/Archicad would have similar delay times, etc ?.

 

I'm certain this would happen, as with any software at some point things will start to visibly slow down if you give it enough to do, it's the nature of the beast.

 

Though the relevance of these stress tests may be disputed, their results and what they reveal concerning the varied processes appear to be clear. They may also go against conventional generalizations but I don't really have any control over that, the facts speak for themselves. So far the stress tests have primarily revealed CPU related processes, will now try to develop tests that will hopefully provide a means to better quantify video card and disk drive relevance.

 

 

[HumbleChief]

Just a little test to add to the confusion....

 

 

[TheKitchenAbode]

16 minutes ago, HumbleChief said:

Just a little test to add to the confusion....

 

 

 

Good example Larry. If you are PBR'ing without a lot of active lights your GPU is not going to be stressed. Each added light requires additional computations to calculate the effect on textures/surfaces, reflections, etc. Also, as you add lights your GPU memory usage will increase, I have ones where one scene eats 4 GB or more. If you were to throw in say 100 active lights and have all camera options checked you will notice a significant difference in your GPU usage as you move the scene. So the take on this is if PBR'ing with complicated lit scenes is not really part of your workflow then there's no real benefit in spending crazy amounts of money on the latest and greatest video card.

[HumbleChief]

35 minutes ago, TheKitchenAbode said:

...So the take on this is if PBR'ing without complicated lit scenes is not really part of your workflow then there's no real benefit in spending crazy amounts of money on the latest and greatest video card.

Seeing that pretty clearly with the benefit of using the 1080 if I upgrade. I'll test a couple more plans at some point if I decide to upgrade but the video card doesn't seem to be a weak link on this system as well it shouldn't - the 1080 is very capable...

 

 

[TheKitchenAbode]

2 minutes ago, HumbleChief said:

Seeing that pretty clearly with the benefit of using the 1080 if I upgrade. I'll test a couple more plans at some point if I decide to upgrade but the video card doesn't seem to be a weak link on this system as well it shouldn't - the 1080 is very capable...

 

 

 

It is accurate, that's the reality of what's going on. Another great example.

 

CA only needs what it needs to do the job. For example, in your vid you rotated/moved around the model. How fast you moved the mouse dictated the number of frames needed to reflect that degree of motion and for each frame the lights and their effect had to computed. The rate is being dictated by your mouse movement. Though simplified, each move of your mouse is equal to some distance, let's say 12", if you scroll 10 feet then 10 frames need to be generated to reflect that movement, not much work for a 1080 card that can easily crank out 60 or more frames per second.

[Michael_Gia]

Is this why my crappy MacBook seems to behave the same as some of the other guys with monster machines?

 

Other than raytrace I don’t see why I would even need to upgrade. 

 

[HumbleChief]

2 minutes ago, Michael_Gia said:

Is this why my crappy MacBook seems to behave the same as some of the other guys with monster machines?

 

Other than raytrace I don’t see why I would even need to upgrade. 

 

I dunno but if you don't experience any slow downs why would you upgrade? I have quite a few plans that 'hit the wall' and Chief just starts to crawl and it blows. Very time consuming and painful as my mind is going X miles an hour and Chief is going X minus 3 - 5 seconds slower.

[TheKitchenAbode]

2 minutes ago, HumbleChief said:

I dunno but if you don't experience any slow downs why would you upgrade? I have quite a few plans that 'hit the wall' and Chief just starts to crawl and it blows. Very time consuming and painful as my mind is going X miles an hour and Chief is going X minus 3 - 5 seconds slower.

 

This is when upgrading is warranted when plans increase in complexity to the point that lag starts creeping into the system. The stress tests are designed to help identify the different causes of lag and hopefully provide some guidance as to the most effective approach to resolve this. I believe it is evident that for your situation the video card has nothing at all to do with the lag you are experiencing, you have 24 GB or system memory so there should be no issue there. Everything tells me that you are just running into certain limitations of those Xeon CPU's. Seeing a specific plan that you struggle with might provide some further insight into this.

 

[TheKitchenAbode]

Library Analysis - I have just finished an extensive assessment of the processes involved during library use.

 

When the library is being accessed there are 3 processes involved, the first being accessing the disk drive to retrieve the symbol/texture, the second process involves the CPU to process the symbol/texture and the third process involves the GPU to display the model/texture.

 

Disk Drive Accessing - disk access times, regardless of symbol/texture complexity it was always under 1 millisecond. Also, under most circumstances once a symbol/texture had been accessed for the first time it remains in memory so subsequent accessing of the symbol does not require accessing the disk drive.

 

CPU Processing - this is where the bulk of the time is spent. Essentially the CPU processes are no different than those involved when generating a camera view. CA must generate the images of the symbol/texture for display in the selection panel and the preview panel, which are basically mini camera views. As symbol/texture complexity increased the CPU processing time increased accordingly.

 

GPU Processing - Though involved as it would be in any camera view the time to process was relatively minor. The GPU really only comes into play once the model has been generated/displayed and it then takes over when the symbol/model is being rotated or zoomed in on, no different than what happens in a standard camera view.

 

As revealed in prior stress tests, the CPU plays the most significant role in overall processing time and it's importance becomes even greater as complexity increases.
 

 

 

[Kbird1]

On 5/27/2019 at 5:59 AM, TheKitchenAbode said:

Library Analysis - I have just finished an extensive assessment of the processes involved during library use.

 

When the library is being accessed there are 3 processes involved, the first being accessing the disk drive to retrieve the symbol/texture, the second process involves the CPU to process the symbol/texture and the third process involves the GPU to display the model/texture.

 

Disk Drive Accessing - disk access times, regardless of symbol/texture complexity it was always under 1 millisecond. Also, under most circumstances once a symbol/texture had been accessed for the first time it remains in memory so subsequent accessing of the symbol does not require accessing the disk drive.

 

CPU Processing - this is where the bulk of the time is spent. Essentially the CPU processes are no different than those involved when generating a camera view. CA must generate the images of the symbol/texture for display in the selection panel and the preview panel, which are basically mini camera views. As symbol/texture complexity increased the CPU processing time increased accordingly.

 

GPU Processing - Though involved as it would be in any camera view the time to process was relatively minor. The GPU really only comes into play once the model has been generated/displayed and it then takes over when the symbol/model is being rotated or zoomed in on, no different than what happens in a standard camera view.

 

As revealed in prior stress tests, the CPU plays the most significant role in overall processing time and it's importance becomes even greater as complexity increases.
 

 

 

 

 

Thanks Graham , I'm still following along , I just don't have much to add to the thread myself..... 

 

Mick.

 

 

[HumbleChief]

On 5/11/2019 at 8:44 AM, TheKitchenAbode said:

Here's a different stress test model. This one allows you to explore more typical CA functions related to walls, roofs, windows and doors. It should not bog down anyone's system but will run slower so you can see what's going on. Suggest having Task Manger open on the process tab so you can see what component does what at what time. At the bottom of task manger there is a digital clock so you can time your operations.

 

1.) Open Plan.

2.) Open Task Manger, Select Process Tab, Select Options "Always on Top".

2.) Open Standard Camera View, Record Time to Open.

3.) Zoom in on one of the Houses. Should be no problem.

4.) Click on a Wall Surface and Drag up Top Edge up. Record Time to Complete Task.

5.) Hit Undo, Record time to complete task.

6.) Zoom out to see all houses.

7.) Select Build, Roof, Build Roof, Roof, Build Roof Planes, Record Time to Complete Task.

 

What to watch for.

 

1.) The "3D Rebuild" pop up.

2.) In Task Manager Observe the activity levels of the CPU, GPU and Disk Drive, When They are Active and When are Not During the Processing.

3.) Depending on What you do Notice That the full 3D Rebuild only Occurs for Certain Changes.

 

My Results (Alien X51 R3)

 

1.) Open Standard Camera View = 18 seconds.

2.) Drag Wall Surface up = 22 seconds.

3.) Undo Drag Wall Surface up = 22 seconds.

4.) Build Roof Planes = 35 seconds.

 

Parade of Homes 400.plan

Results Ryzen 3900x signature system

 

1.) Open Standard Camera View = 16 seconds.

2.) Drag Wall Surface up = 18 seconds.

3.) Undo Drag Wall Surface up = 17 seconds.

4.) Build Roof Planes = 32 seconds.

 

Add as much time as you feel I missed the timer start by.

 

 

Undo Wall Drag, forgot to record the wall drag up but got 18 seconds again, adding in the time lag to get to the timer.

 

 

Build roof planes

 

 

[HumbleChief]

On 5/11/2019 at 1:34 PM, TheKitchenAbode said:

Here is another stress test model. This one looks at the effect of very high surface counts. The model contains 5,000 BBQ symbols for a total surface count of just under 19 million. The test here is a simple one.

 

1.) Open Plan.

2.) Open Camera View, Record Time to Display.

 

The results may surprise you.

 

This one should be of greatest interest to those with high core count processors as the processing here is fully threaded.

 

My 4 core/8 thread 6700K displays in 34 seconds.

 

The rule in respect to core/thread performance is in general, to reduce the time by half you need double the cores. Theoretically to drop my 38 seconds to 19 seconds will require 16 cores that can run at my 4.1 Ghz rate. To get this down to 8 seconds would require 32 cores, 4 seconds 64 cores, 2 seconds 128 cores and to 1 second 256 cores.

 

BBQ Parade.plan

 

 

3900X

 

 

[HumbleChief]

BBQ'S WITH CPU USEAGE. This might interest you Mark. The hardware monitor is with Ryzen's PBO single thread settings. You can see the CPU hit a high of 4591 and it uses 100% CPU to render that camera view. Interesting.

 

 

[TheKitchenAbode]

11 hours ago, HumbleChief said:

BBQ'S WITH CPU USEAGE. This might interest you Mark. The hardware monitor is with Ryzen's PBO single thread settings. You can see the CPU hit a high of 4591 and it uses 100% CPU to render that camera view. Interesting.

 

The high CPU core usage is due to the fact that the BBQ's are symbols and the majority of work is related to the textures, this is a multi threaded process so your high core count 3900X and it's ability to run those cores at a high frequency provides a considerable benefit. On the parade of homes you can also see this if you watch task manger, however due to the model rebuilding which is primarily single threaded the benefit is not as great as texture processing here only represents maybe 10% of the overall processing time.

 

IMO you've done a great job in putting together your new system and doing so without breaking the bank. Personally, though you are considering other possible future upgrades I would be very cautious not to set to high of expectations as to how great a performance benefit will be gained over what you currently have. Certainly switching to a MVNe M.2 drive will help and given their low cost they are very good value for the money. Before coughing up the big bucks for one of those newer drives I would do as Rene suggest and just set-up a Ram Drive, you have enough memory, it costs nothing to do and it will be faster than any other drive system. Also, setting up a Ram Drive is a good way to test/predict the potential performance improvement before spending any more money. If you set one up and can't really tell the difference then it's telling you that the type of work you do will not really benefit from these new drives.

 

[HumbleChief]

Thanks Graham and I've learned over the years to keep my expectations low and am under no illusion that a new 4.0 SSD will have a noticeable effect. The challenge now is that my 1 TB 2.5" SSD is almost full and need more storage. Not today, but soon and am watching NVMe drive prices for a deal. both 3.0 and 4.0 PCIe. Will do nothing for as long as possible as that always been smart in PC world.

 

I'm only mildly curious about a RAM drive and the first thing I read said be wary of them for various reasons. May explore but usually find my time better spent  at other tasks. Do you have a RAM drive? They seem risky but am quite ignorant about such things...will research.

[TheKitchenAbode]

1 minute ago, HumbleChief said:

Thanks Graham and I've learned over the years to keep my expectations low and am under no illusion that a new 4.0 SSD will have a noticeable effect. The challenge now is that my 1 TB 2.5" SSD is almost full and need more storage. Not today, but soon and am watching NVMe drive prices for a deal. both 3.0 and 4.0 PCIe. Will do nothing for as long as possible as that always been smart in PC world.

 

I'm only mildly curious about a RAM drive and the first thing I read said be wary of them for various reasons. May explore but usually find my time better spent  at other tasks. Do you have a RAM drive? They seem risky but am quite ignorant about such things...will research.

 

I don't run with a Ram Drive but have set one up just to see. I myself could not detect any difference however that may be related to the type of work that I do. I believe Rene derives a true benefit as he seems to be using numerous custom high resolution textures and some very complex models/symbols, due to their size their load times will be much greater than CA's included textures and symbols so a Ram Drive will get this done as fast as possible.

 

What you need really depends on what you do, and that varies for every user.

[HumbleChief]

Thanks again Graham always appreciate your input/feedback.

[Renerabbitt]

1 hour ago, TheKitchenAbode said:

Rene derives

 

1 hour ago, HumbleChief said:

Thanks again Graham always appreciate your input/feedback.

I love the way the pros on the site speak to one another, such respect!

About the RAMdisk, Graham is correct in that I see the biggest benefit in Library, and though it's not quantifiable I had noticed a speed-up across a few of my active plans at the time of installing CA to RAM.

an aside, the BIGGEST benefit I see from using a RAMdisk is outside of CA with the pagefile in my Adobe Suite and the temp directory for Google Chrome...adobe loading times re a fraction of what they were and Chrome feels supercharged.

I have yet to have a problem with AMD's RAM drive as it backs up at close of session and automatically loads from previous state...with as much RAM as you have, changing your temp drives to a RAMdisk shouldn't effect anything, fun t play around with when your time frees up 

[HumbleChief]

7 minutes ago, Renerabbitt said:

 

I love the way the pros on the site speak to one another, such respect!

About the RAMdisk, Graham is correct in that I see the biggest benefit in Library, and though it's not quantifiable I had noticed a speed-up across a few of my active plans at the time of installing CA to RAM.

an aside, the BIGGEST benefit I see from using a RAMdisk is outside of CA with the pagefile in my Adobe Suite and the temp directory for Google Chrome...adobe loading times re a fraction of what they were and Chrome feels supercharged.

I have yet to have a problem with AMD's RAM drive as it backs up at close of session and automatically loads from previous state...with as much RAM as you have, changing your temp drives to a RAMdisk shouldn't effect anything, fun t play around with when your time frees up 

Thanks as always Rene,

 

My system is so much faster everywhere right now that I don't see a problem that needs fixing, which is a really refreshing thing to post and a sentiment I could not share just a few days ago. Still curious about the RAMdisk and found a great video you did explaining same. When my curiosity gets the better of me I might explore but for now it might only solve a problem I do not have. 

[RobUSMC]

Great testing report Graham.   Thank you

[Renerabbitt]

On 5/13/2019 at 11:48 AM, TheKitchenAbode said:

 

I'll send you my address so you can have one dropped off in my driveway for a test run. Results won't be very good though, with our traffic conditions about 60 miles an hour is tops and there are traffic lights about every 2 blocks.

Hey Graham found something interesting on the rebuild of my computer...While I was waiting for pci x16 extension cables I had to use 1x riser cards to temporarily keep my machine up and running. With the GFX cards limited to 1x lanes, the mouse in the CA model-space would jitter all over the place and was extremely slow to manipulate the plan...same as 3d views. When the cursor made it back to hovering over toolbars it was restored to normal responsiveness but as asoon as you hovered over plan it was almost unusable...gfx cards definitely in play here.

Upon installing the 16x lane extension cables the problem immediately dissapeared.

Just interesting.

[TheKitchenAbode]

12 minutes ago, Renerabbitt said:

Hey Graham found something interesting on the rebuild of my computer...While I was waiting for pci x16 extension cables I had to use 1x riser cards to temporarily keep my machine up and running. With the GFX cards limited to 1x lanes, the mouse in the CA model-space would jitter all over the place and was extremely slow to manipulate the plan...same as 3d views. When the cursor made it back to hovering over toolbars it was restored to normal responsiveness but as asoon as you hovered over plan it was almost unusable...gfx cards definitely in play here.

Upon installing the 16x lane extension cables the problem immediately dissapeared.

Just interesting.

 

Sounds like what happens when they do road maintenance on the highway and narrow it down from 6 lanes to one. If I can recall from past playing around I did notice that behavior, I believe that when the cursor is in a plan or camera view window the graphics processing is different than when the cursor is over the menu window area. I envisioned it as a window within a window, each are independent of the other and depending on the content type it can take more time for the GPU to regenerate that window.

[Renerabbitt]

3 minutes ago, TheKitchenAbode said:

 

Sounds like what happens when they do road maintenance on the highway and narrow it down from 6 lanes to one. If I can recall from past playing around I did notice that behavior, I believe that when the cursor is in a plan or camera view window the graphics processing is different than when the cursor is over the menu window area. I envisioned it as a window within a window, each are independent of the other and depending on the content type it can take more time for the GPU to regenerate that window.

Sounds right to me, I was surprised at how crippled the software was, it was the only modelling software that was effected in such a way.

[TheKitchenAbode]

It also made sense to me so I'm sticking with it until proven wrong.

 

I think there are a lot of things in CA for whatever reason perform different than what one would expect. PDF importing is another one, I can do this in OneNote and everything is smooth, import into CA and things slow down real fast, I can also load Tab after Tab of PDF files in my browser and nothing slows down. If others can do it why can't CA.