Mac Pro vs. iMac with VW a few test results

[steve s]

I ran a few render tests with the new Mac Pro (6core 32GB ram D500) and iMac 2012 i7 24 Gb ram ,on the same file, to see if the MP is faster.

Fast renderworks 1st render iMac = 1 min. 25sec MP= 1min. 21sec

re-render = 30 sec MP= 30 sec

Final Q 1st render iMac= 6min. 45 sec MP= 5min 36 sec

re-render = 2min 57 sec MP= 2min 43 sec

Full Q , High Q detailed 3d, OpenGL very smooth with flyover tool on MP, just a little smoother than on the iMac.

Not much difference, hope they optimise VW for the MP soon.

Final Cut Pro screams though on the new MP.

[CipesDesign]

I assume the speed difference can be attributed to 64 bit processing?? Or???

[M5d]

Doesn't seem right, Renderworks should already see about a 50% speed improvement from the two additional cores.

Do you have hidden line foregrounds in your test renders?

[billtheia]

That's really disappointing. I would have expected the MP to be MUCH quicker. What gives?

[steve s]

no hidden lines, its a DTM, trees, high polygon playground equipment, a few other surfaces

[Christiaan]

I'd be interested to know what's holding it back too. Jim, do you have any insights?

[steve s]

I'd be happy to run some more tests if anyone has a particular file they want me to try.

[PVA - Admin]

All the test renders were done with the same VWX file in Vectorworks 2014 SP1. It was a complex file sometimes reaching 6+ hour renders for the higher qualities, so some of the cells are blank due to incomplete tests. However all of the fast and hidden line test results give you an idea of what you could expect as a difference between certain CPUs.

Im a little surprised by the results in the OP of this thread, however since its all under 10 minutes, (coupled with the fact that re-renders were significantly faster than initial renders) it could be that the majority of the render time in that test was spent on geometry, which is still single-core and wouldnt show improvement on a CPU with more cores.

Attached is a sample of some of the speed results I have been getting with specific hardware, part of a larger project to help users buying/assembling machines to use with Vectorworks. They are not official in any way, just my preliminary results, but they reflect what I would expect performance differences to be between slower and faster CPUs:

Edited January 29, 2014 by JimW

[PVA - Admin]

The "cb" score listed for each CPU is a score assigned by Cinebench R15, by the way. This is a snippet of a much larger chart.

[Christiaan]

All the test renders were done with the same VWX file in Vectorworks 2014 SP1. It was a complex file sometimes reaching 6+ hour renders for the higher qualities, so some of the cells are blank due to incomplete tests. However all of the fast and hidden line test results give you an idea of what you could expect as a difference between certain CPUs.

Jim, what tests are you talking about here? With a new Mac Pro? I don't see a new Mac Pro in your attachment.

[PVA - Admin]

The times listed in the image, Im listing the various CPU types and various OSes of machines I have here. If/when I get access to one of the 2013 MacPros ill be adding them specifically to the list.

This was mainly to show that even a moderate increase in CPU power can dramatically affect rendering times.

For instance the i7 at 2.4GHz 4c/8t rendered the test in 14 minutes, but the i7 at 3.4GHz 4c/8t rendered the test in 6 minutes.

As for why the Mac Pro in the OPs example wasnt dramatically faster than the iMac, I suspect the test file used in OP had a lot more geometry (which is single threaded) to work through than indirect lighting (which is multi threaded) etc.

[steve s]

I think that is true , the file had mostly geometry not much lighting.

[Christiaan]

steve, perhaps you run some tests again with a bigger rendering to geometry ratio?

[steve s]

would adding lights be a good way to do that?

[PVA - Admin]

You (or anyone reading) can feel free to use the same file I used in my tests, its linked at the bottom of this post. Final Quality will likely take entirely too long, but a test in Fast Renderworks should be reasonable, under 25 mins I suspect for your hardware.

The viewport on the sheet layer marked as "Fast RW Test" is what you'll want to use, all of the times on the table were done from those. Just select the viewport and hit update.

Rendering in the design layer or with different DPI settings will change the results.

https://vectorworks.groupdropbox.com/pickup/file/7bab7789b5f78044

[steve s]

got it, I'll give it a go.

[bcd]

I know it's not approved but the mac mini (worth it's weight in silver)

renders this out at under 8 mins! (just sayin')

(7:29)

[PVA - Admin]

I know it's not approved but the mac mini (worth it's weight in silver)

renders this out at under 8 mins! (just sayin')

Which Mac Mini are you currently using? We only have the 2,1 version with 1.8GHz Core 2 Duos here, so they perform poorly and I didn't include them in my tests.

http://en.wikipedia.org/wiki/Mac_Mini

From what I am hearing from you the latest ones may be worth another look performance-wise.

[bcd]

and 7:42 for HLR

[bcd]

I would Jim, especially now with a refresh on the horizon.

Granted - the machine can be treacle when the rendering is running - and HLR may/maynot complete depending on how many nuts the VW memory squirrels have stashed in their cheeks.

I'm running the Late 2012 - a different beast to the C2D

Macmini6,2

2.6 GHz

i7 processor

4 Core

L2 Cache (per Core): 256 KB

L3 Cache: 6 MB

Memory: 16 GB

and 1 TB Fusion Drive

[steve s]

OK some results of the Render Test file

Fast Q iMac 7 min:13 2nd time 6 min 50

Mac pro 5 min 00 2nd time 4 min 13

Final Q iMac failed to finish, seems to get stuck, tried 1st time 38 hours with no progress on the blue render bar; 2nd try currently 10 hours made it to about 80% at 7 hours but seems stuck now

Mac Pro final Q Average 2 runs 3hours 8 min

[PVA - Admin]

That is a little more in line with the render times I would expect.

Don't worry too much about the FQRW one for now, I haven't finished tweaking the model yet for a good bench in that mode, it takes many many hours here when it does finish successfully which is just too long for an assessment tool. (I totally acknowledge it should NEVER fail to render, but most of its failings are my fault in this case.)

[steve s]

Also the hidden line test fails on the Mpro, got this error

[/img]

[dtheory]

Jim,

Has the table from earlier in this posting been updated to include any recent iMac 5K or 2013 Mac Pro??

[PVA - Admin]

Somewhat, no iMac 5K yet but I did a bit on the two Mac Pro models. I haven't yet gotten the time to fully sit down and retest everything with enough rigor for my taste, I have a project proposal pending that would allow me to do this and publish a more formal chart.

( The machine named Ceres with the crazy numbers is a rig I'm working on at home, so far it seems like I was able to trounce a Mac Pro at less than half the cost, but more on that later maybe )

However, from the Cinebench CPU scores, you can fairly well extrapolate the expected improvements in render speed, they scale almost exactly together in full tests done with enough rigor to confirm results. But as of right now some of these are only one-runs and aren't enough data points for a full comparison:

Edited April 16, 2015 by JimW