Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Open GL Vs Metal
- Thread starter dfritchie
- Start date
- Sort by reaction score
Mac Pro is in my sig. Tests were done on a 1440p monitor. (Running 10.11.5)
Added two more results; running Sierra Public Beta 1 using Apple's stock Nvidia driver (355.10.05.05b04) and Nvidia web driver (367.05.10.05b07).
Added two more results; running Sierra Public Beta 1 using Apple's stock Nvidia driver (355.10.05.05b04) and Nvidia web driver (367.05.10.05b07).
Attachments
Last edited:
Well it seems with one card, the 7950 all perform about the same with Open GL being about 60 FPS and Metal about 120 FPS. If you have two though the Open GL rocks! I find it interesting that my old 1,1 is on par with the 5,1's in this program, wish I could run the higher res to see just how close they are.
My understanding is like perform in "headless", so, won't affected by something like v-sync (cap at 60FPS). The offscreen result should be more accurate.
Last edited:
Yes, headless. An offscreen benchmark runs in memory at a resolution that the screen doesn't support.
Not even one of those people bragging about how the Nvidia web drivers are optimised for 9 series posted a benchmark. They are so vocal when they need to be and then run for the hills whenever asked for some numbers or if they have noticed the number of OpenCL bugs that plague pros.
I could do the OpenGL Benchmark with GTX 770 in the Mac Pro, GTX 630 as eGPU, and GTX 750 Ti as eGPU connected to a Mac Mini 2012.
But the Metal Benchmark crashed with the only Maxwell GPU I have, the GTX 750 Ti. So no Metal benchmark for Maxwell...
But the Metal Benchmark crashed with the only Maxwell GPU I have, the GTX 750 Ti. So no Metal benchmark for Maxwell...
Attachments
Last edited:
It seems the onscreen test is no longer locked at the 119FPS in 10.12. The T-Rex (onscreen) score 3x more than on 10.11 (please see my previous post).
I've been on vacation for the last few weeks, but will run these tests when I get home just to keep you happy. I'm also not sure that a benchmark that runs at 3-400 frames per second is that representative of a modern game, these tests have typically been used on phones and tablets rather than a high end desktop GPU.
It's the only thing I know of that test Metal on a desktop Mac, that's free anyway. That's why I used it.
I'll run on the hack tonight if you promise to make your response to it your last post on the subject.
Here's one on a Titan X. Monitor runs at 30Hz.
Does this benchmark comparison mean anything? It seems to prove that the devs managed to program these graphics scenes more efficiently on metal than on opengl. Now that depends more on their ability to program these APIs than on the APIs themselves, am I wrong ? The doubling of framerates with metal on the onscreen test is odd (double the screen's frequency). Does it mean we have to divide the metal results by 2 to get a fair comparison with the opengl test ? Well the offscreen test still seems to scale with GPU speed so there's that.
It's also the first graphics benchmark I try that doesn't tax the GPU enough to ramp up the fan even slightly. I did however experience hearing the coil whine for the first time, with the fans being so silent and no sound.
Does this benchmark comparison mean anything? It seems to prove that the devs managed to program these graphics scenes more efficiently on metal than on opengl. Now that depends more on their ability to program these APIs than on the APIs themselves, am I wrong ? The doubling of framerates with metal on the onscreen test is odd (double the screen's frequency). Does it mean we have to divide the metal results by 2 to get a fair comparison with the opengl test ? Well the offscreen test still seems to scale with GPU speed so there's that.
It's also the first graphics benchmark I try that doesn't tax the GPU enough to ramp up the fan even slightly. I did however experience hearing the coil whine for the first time, with the fans being so silent and no sound.
Last edited:
I'm hearing Metal has matured nicely in macOS. Should be awesome with Polaris
Yeah, but I want P10, I think that's quite a good match with cMP. However, at this moment, only the weaker P11 has full support in Sierra, what a shame.
Here are my results for my Core i7-4790K with a TITAN X, using 10.11.5 with the 346.03.10f02 web driver. I've attached a comparison of the tests that are likely to be GPU limited with the GTX 770 posted earlier, for ease of reference. I don't know how you can look at results like these and conclude that the Maxwell cards are running in compatibility mode or are missing optimizations or whatever else people are claiming.
Attachments
Poor in comparison to the 770. If you can't see how different that delta is compared to how well the Titan X is supposed to run in comparison to the 770 then...what am I saying...of course you can't see the issue here. I wasted one year trying to tell you. Haha....facepalm.
Okay, so just to be clear, you're complaining that Metal performance isn't as good as DirectX under Windows? I've spent a year trying to explain to you what it means to be CPU limited, and how the Apple OpenGL framework makes it impossible to match Windows OpenGL or DirectX performance, so I'm hoping we're past that.
Let's look at how the original 3.0 version of Manhattan performs. D3D gets 735.5 FPS, Metal gets 625.2 FPS, or about 15% slower. Is your whole argument that Metal should match D3D on a test that renders more than 10 frames every screen refresh? Given that NVIDIA has more people working on D3D performance than they have working on all aspects of the Mac drivers, it's not surprising that they've been able to squeeze out more performance here, and/or that their Mac team has been focusing their effort elsewhere.
Let's look at how the 3.1 version of Manhattan performs. Metal gets 423.8 FPS and beats Windows OpenGL with 345.3 (though the max score looks like it's much closer). What are you complaining about here, exactly?
So, what exactly would make you happy with respect to GFXBench results? Are you upset that GFXBench isn't matching what you expect based on the paper specs of the two GPUs? What makes you conclude that the Mac driver is running in compatibility mode, or doesn't have color compression enabled? I get that you think the NVIDIA web drivers are **** and that they should be doing a better job of supporting your usage cases, and you've raised some legitimate concerns about issues with Adobe apps (many/most of which have already been addressed).
Enjoy your Polaris GPU, I'll enjoy using my 1080 Ti when it gets released. I was hoping to be able to include comparisons to the high-end AMD GPUs like the Fury X, but I wasn't able to find any data. So yeah, let's keep complaining about the NVIDIA web drivers because they're at least enabling all their GPUs to work.
Attachments
I'm hearing Metal has matured nicely in macOS. Should be awesome with Polaris
Yes indeed, I just installed Sierra. Below comparison of the GTX 770 between OS X 10.11 and Sierra:
