BruceX benchmark with RX 580 slower in High Sierra 10.13.3

[h9826790]

I do also have a problem with my BruceX performance in Final Cut Pro.

Yesterday I got a pretty cheap Mac Pro 4,1 (flashed it to 5,1) with 8 Cores and tested it against my 3770K Hackintosh. Both machines were equipped with the same GTX1080, Final Cut 10.4 Test Library was stored both times on an internal SSD. So I ran the BruceX test an noticed a huge difference in the results (see below). And I have no idea, where this difference is coming from, because BruceX is completely GPU-related? So maybe someone else has.

3770K@4,3GHz, 16GB DDR3-1333, macOS 10.12.4 = 32 Seconds
MacPro4,1@ 8x 2,26GHz, 20GB DDR3-1066, macOS 10.12.6 = 60 Seconds

The only difference these two machines have is the single core CPU performance, but CPU idled about 99% during the run. First I ran 10.13.3 on the Mac Pro, got also 60 Seconds and thought the difference is because of nVidia messed up the new drivers. So I installed Sierra and got the same slow result.

Any thoughts? :/

It's possible that the Nvidia web driver's performance (in FCPX) is somehow related to the CPU single core performance. You are not the 1st one to report that your Nvidia card has good (relatively) performance on a Hackintosh.

In fact, your OC 3770 has roughtly 2x E5520's single core performance.

And coincidentally, your Hackintosh also provide roughly 2x 4,1's performance in BruceX.

 

[hajwer]

Yeah, that would make sense obviously. I will start further testing in the next days and try to underclock the 3770K to check if it's really single core related.

Also it will be interesting, if the change to 2x Xeon X5680 will improve the performance.

 

[hajwer]

So I was able to underclock my 3770K to 2.4GHz and the BruceX benchmark duration increased significantly. From the previous 32 seconds to about 44 seconds. Geekbench score was about 600 points higher than on the E5520 (2600 vs 2000 in single core).

I tried to clock down as far as they both were nearly equal, but somehow the macOS on the hackintosh seems to ignore lower clock speeds, because the Geekbench scores where near the stock speed numbers with 2.00 or 1.80 GHz.

 

[h9826790]

So I was able to underclock my 3770K to 2.4GHz and the BruceX benchmark duration increased significantly. From the previous 32 seconds to about 44 seconds. Geekbench score was about 600 points higher than on the E5520 (2600 vs 2000 in single core).

I tried to clock down as far as they both were nearly equal, but somehow the macOS on the hackintosh seems to ignore lower clock speeds, because the Geekbench scores where near the stock speed numbers with 2.00 or 1.80 GHz.

Thanks for the report. So, it seems BruceX is very similar to CineBench OpenGL test. It tax the GPU, but very CPU single thread performance limiting.

In fact, 44 x (2600/2000) = 57.2

It's very close to 60s (your E5520 cMP's BruceX score)

 

[hajwer]

But considering that a RX580 for example is much than the GTX1080 I think it’s the entire NVIDIA driver which is limited by the single core performance.

 

[h9826790]

But considering that a RX580 for example is much than the GTX1080 I think it’s the entire NVIDIA driver which is limited by the single core performance.

I believe that you are very correct indeed.

CineBench OpenGL test is a well known very CPU single thread limiting test. And my 1080Ti is 15% slower than my HD7950 in this particular benchmark. Which indicates that web driver is more CPU single thread limiting then the AMD driver in MacOS

 

[mghamms]

I have run boinc setti project on several Mac pro's and Windows machines with Nvidia gtx 680, 780,980 and 1080 cards.
Runs opencl and I have not seen any noticeable speed difference using the same card in the two platforms.

One problem that I just ran into though on a MacPro 5,1 OS X 10.13.3 only one of two gpu's will run opencl.
Cuda is fine on both.
Anyone have problems running two nvidia or Ati gpu's?

 

[bsbeamer]

Anyone have problems running two nvidia or Ati gpu's?

There are some issues with 10.13.3 with eGPU at the moment that Apple and Sonnet are apparently working out, some which have been resolved in the 10.13.4 betas. Could this be impacting dual GPU as well?