OS X Metal vs OpenGL benchmark

[Fuchal]

Here's my sad old Mac Pro. Great gains in Metal.

Mac Pro 3,1
2.66 8 Core, 24GB ram
GTX 970, Nvidia Web Drivers
10.11.3

All Offscreen 1080p

Metal - GL

Manhattan 3.1 265.1 - n/a

Manhattan 373.3 - 101.32

T-Rex 719.8 - 441.47

ALU 2 742.8 - 664.19

Driver Overhead 2 123.9 - 15.64

Texturing 82457 - 93403

 

[jeanlain]

The 10.11.4 update increased the frame rate of the Metal T. rex test by 30% on my Mac.
The Manhattan tests got +10%
Not bad.

 

[imacken]

The 10.11.4 update increased the frame rate of the Metal T. rex test by 30% on my Mac.
The Manhattan tests got +10%
Not bad.

10% down for me on 10.11.4!

 

[MacBH928]

So Metal is actually worse than OpenGL?
can someone explain?

 

[imacken]

Don't understand where you get that from! Look at my results on post #4

 

[MacBH928]

Don't understand where you get that from! Look at my results on post #4

from the first post stating this:

"So a clear advantage for OpenGL across the board, except for the driver overhead test."

 

[imacken]

The OP results are misleading as he has a limited Intel GPU. Look at the results from others on this thread.

 

[antonis]

However, Intel GPUs is - sadly - what the majority of Macs are equipped with. If metal performs worse than opengl on these gpus...

 

[pedrom]

So Metal is actually worse than OpenGL?
can someone explain?

No. The thing is that Metal is new, Apple only, and takes time to update and bring in the new features.

OpenGL, despite inferior as a technology, has a lot of features and companies behind it. The same happened with html5 vs flash, Vulcan Vs OpenGL, directx 12 Vs 11, etc.

 

[h9826790]

Have an interest found on my GFXBench Metal test. I have 2 GPU in my Mac Pro (dual 7950), for OpenGL, it always use the GPU that connected to the monitor, but Metal use the GPU that not connected to the monitor.

I am not sure if Metal always use the 2nd GPU regardless if it connect to the monitor or not. I haven't test it yet. But at this moment, there is no way to choose GPU in GFXBench Metal.

Anyway, my 2nd card only installed in a x4 slot (for better cooling purpose). In other tests (Luxmark, Unigine Valley, etc), the 2nd GPU is 2.4% behind the 1st GPU that installed in a x16 slot. However, I still get some crazy result in the Metal benchmark like this.

 

[antonis]

Have an interest found on my GFXBench Metal test. I have 2 GPU in my Mac Pro (dual 7950), for OpenGL, it always use the GPU that connected to the monitor, but Metal use the GPU that not connected to the monitor.

I am not sure if Metal always use the 2nd GPU regardless if it connect to the monitor or not. I haven't test it yet. But at this moment, there is no way to choose GPU in GFXBench Metal.

Anyway, my 2nd card only installed in a x4 slot (for better cooling purpose). In other tests (Luxmark, Unigine Valley, etc), the 2nd GPU is 2.4% behind the 1st GPU that installed in a x16 slot. However, I still get some crazy result in the Metal benchmark like this.
View attachment 623515

Have you ever happened to test them in crossfire, on the windows side ? I am curious if they can function under crossfire while they're running on different bus speeds.

 

[h9826790]

Have you ever happened to test them in crossfire, on the windows side ? I am curious if they can function under crossfire while they're running on different bus speeds.

Yes, I have the crossfire bridge installed, and crossfire works well on the Windows side.

There is no requirement to use identical card / same bus speed etc in Crossfire. As long as the cards belongs to the same family, they can be crossfired. Which is an advantage over SLI. However, the weaker card of course will drag the faster card. So, use 2 identical sard still preferable.

Anyway, all I need to do in Windows is OC the 2nd GPU a bit (10MHz), which can effectively compensate the lower bus speed, and make both card able to work to 99%, rather then 99%+97% under full load in Corssfire.

 

[wubsylol]

Some testing I did last night suggests the stock nVidia drivers are currently outperforming the web driver for Metal rendering (on my 780M at least). Assuming this trend remains, it would be unfortunate for Mac Pro owners running Maxwell cards that require the web driver.

 

[koyoot]

Those are exactly the same observations of behaviour of my MBP mid 2012 with GT650M.

 

[wubsylol]

Interesting. I'm hoping nVidia can resolve this. Metal, although promising, is in its infancy now and it's going to be a while before we can safely forget about OpenGL and I don't really want to have to sacrifice performance of one for the other.

 

[jeanlain]

Some testing I did last night suggests the stock nVidia drivers are currently outperforming the web driver for Metal rendering (on my 780M at least). Assuming this trend remains, it would be unfortunate for Mac Pro owners running Maxwell cards that require the web driver.

If you look at the results on the first post of this page, Metal is still a huge benefit on Maxwell cards.

 

[antonis]

Yes, I have the crossfire bridge installed, and crossfire works well on the Windows side.

There is no requirement to use identical card / same bus speed etc in Crossfire. As long as the cards belongs to the same family, they can be crossfired. Which is an advantage over SLI. However, the weaker card of course will drag the faster card. So, use 2 identical sard still preferable.

Anyway, all I need to do in Windows is OC the 2nd GPU a bit (10MHz), which can effectively compensate the lower bus speed, and make both card able to work to 99%, rather then 99%+97% under full load in Corssfire.

Thanks for the clarifications. When I owned the nMP, I had mixed results regarding crossfire in games. Sometimes it worked, sometimes it didn't. However, I wasn't sure if crossfire was able/smart enough to lower the faster gpu in order to keep things in order. Glad that it does that.

 

[wubsylol]

If you look at the results on the first post of this page, Metal is still a huge benefit on Maxwell cards.

Sorry, I think you've misunderstood what I said. Metal using the nVidia web driver is currently not performing as well as Metal using the (older) drivers included with OS X. As Apple have not manufactured any system with a Maxwell-based card, the stock OS X drivers do not support Maxwell cards and users (only people with an older Mac Pro) need the web driver.

I am saying it would be a disappointment if a similar trend as I have experienced (web driver performing 15-20% slower than stock drivers) were happening with Maxwell-based cards, and thus were unable to operate at their full potential and users having no alternate.

If you are running a Maxwell card, this is likely not as big an issue for you anyway as there are still improvements with Metal and your card is powerful enough to handle (mostly) whatever you want to throw at it anyway (and you have no alternate so there's not a lot you can do anyway). This is more of an issue for people running older (not that you can really consider 2013 to be "old") iMacs and older Macbook Pros with Kepler (and maybe Tesla) cards, as OpenGL performance using the web driver is far, far, far better than using the stock drivers.

Having to choose between OpenGL performance for the bulk of your apps or Metal performance for a smaller number is not ideal. Hopefully nVidia can work this out.

 

[FireArse]

Mac Pro 2006 running 10.11.4 with Mac Edition ATI 7950 and 8-cores of X5355 at 2.66GHz:

Test Name (value): Metal / OpenGL
Manhattan 3.1 (fps): 177.701 / --
Manhattan (fps): 256.527 / 126.31
T-Rex (fps): 469.038 / 347.34
ALU2 (fps): 614.192 / 536.7
Driver Overhead2 (fps): 48.2632 / 19.631
Texturing (MTexel/s): 58198 / 58987

For anyone with a 2006 Mac Pro, the addition of the GPU and 10.11.4 enables video AirPlay to the newest AppleTV too

Metal providing some real kick for this 2006 machine. Bring on more Metal games.

 

[Irishman]

I'm still waiting for the first metal games to come out!!

 

[throAU]

Metal is still new and i doubt things are optimised for it yet.

The whole point is the driver overhead being reduced, so "the driver overhead was the only advantage" is kinda the point.
[doublepost=1459346360][/doublepost]

As ever, we hear of developers 'planning' to adopt Metal, but, what is it, 9 months on and nothing has been announced as far as I know.
Another dead duck? Hope not.

Games are typically in development for several years. The mac is not a gaming platform, and only got metal a few months ago on El Cap, which not everyone runs.

You'll see more support for Metal when more iOS applications use it and they start getting ported.

Most cross platform computer based developers (i.e., devs doing windows/mac/mac) will continue to code for OpenGL until there's a compelling advantage because the one rendering tech will work across windows/mac/linux.

 

[jeanlain]

The gains brought by Metal in this benchmark are too good to be true. I'd be surprised if we get anything remotely near those kind of improvements in a real game, even at low resolution.

 

[Marshall73]

They're comparing the same presets in live vs alpha without realising that the graphics have been bumped up massively in alpha at the same presets (view distance etc). Which makes the comparison void. Metal is already faster.

The latest Alpha build positively flies on my 27" retina iMac, getting higher FPS by 50-75% than the current live build and with higher settings across the board. Shadows no longer cause a massive FPS hit which is nice, now if they could only apply this to Diablo 3

 

[wubsylol]

DirectX still flies ahead of Metal in GFXBench and the WoW: Legion alpha :\

 

[oldmacs]

I've found metal great on my Macbook. Sadly not so much on iOS