25x faster through bootcamp than native MacOS on 2017 iMac with Radion Pro 570 for triangles per sec

Murray_2

Hi all, 

My triangle count using the compatibility check is very odd. Is there a reason why the mac is performing so negatively?

iMac Mojave = 32 Million triangles
Windows 10 using a virtualised bootcamp through Parallels 15 = 822,8 Million+ triangles

both are using the Chrome web browser.

thanks

Murray

brian_brady

Have you tried using Safari instead of Chrome on the Mac? I have found that Chrome always underperforms versus Safari on my MacBook Pro. Often by quite a lot.

Murray_2

thanks for the suggestion, but the result is the same (30,8 million).

john_mcclary

Both are using the radeon? Those look like intel integrated numbers you're getting on your macos

Murray_2

Yep, they are both Radeon. I was actually getting slightly better numbers from my 2012 i7 Mac mini with integrated graphics :-(

my line numbers are the same on both the mac and Windows. 

john_mcclary

Can you give a full screen shot of the system checks?

I donno what to say yet.  Could be webgl not active or something, browser version, drivers...

jakeramsley

Hi Murray_2,

Boiling down graphical performance to a single (or a small set of) numbers can be difficult and it is something we are looking to improve.  As part of this, I hope we can disclose some of the science behind how the information is calculated a bit more when we expand on this.  So while I don't have an exact answer, here are some of the factors that go into graphical performance:

• Browser
• Operating system
• GPU
• GPU driver

When we try to do our performance tests, we use a representative model and try to load up as much detail as possible, do some actions, and measure the results.  The detail includes faces, edges, lighting, ambient occlusion, and other factors.  Unfortunately all of these algorithms aren't computed and rendered with the same efficiency when talking to the GPU.  As such, when we detect a slow algorithm we try to switch to another one based on the factors listed above.

Something we noticed back in 2017 is that Radeon cards using the OSX driver had a ~20x performance degradation with ambient occlusion, which is roughly comparable to the numbers you've posted here.  My guess is two things:

1. We have a bug detecting Radeon cards for OSX properly and running the wrong algorithm in this case.  Please send in a support ticket so that we can track and triage this issue.
2. When running through a VM like Bootcamp, we are detecting which algorithms are available differently and running the check through different shaders giving a higher number.

It is also possible that the driver that Bootcamp is using is somehow more efficient.  I am no expert on VMs, so I can't give any definitive statement on it.

The product does dynamically switch algorithms based on performance, so I would take these differences with a bit of grain of salt and determine if it is a hardware/driver issue more through trial and error than just these numbers listed.