OS X Metal for OS X?

[TallManNY]

thanks for the explanation. guess I'm out of luck then.

Me too, I guess. Though I have a nice GPU in my 2011 iMac.
Oh well, I generally consider an upgrade after five years. So maybe I buy the next refresh of the iMac. Though dropping three grand for retina with top GPU is a pretty big pill to swallow. Hopefully the price has come down a bit in a year or so.

 

[Irishman]

its a long thread so apologies if its been covered somewhere in here, i saw something about it being compatible with all macs since 2012. does this mean nothing before 2012 will be compatible, or that all models from 2012 plus some from earlier will be compatible. since my 2011 27" is more powerful than a lot of machines produced afterwards it'd be dissappointing if it wasnt compatible.

We don't honestly know (non-Apple and non-developer Muggle folk) what the reasons are for the cutoff point. The first video on the Dev Site suggests that Macs since 2012 are compatible with Metal. At 29:16 in that video, it's suggested - in the section on app thinning - that only GPUs with 1GB of VRAM or higher are supported. Those with 512MB or less will get legacy OpenGL 4.1 implemented. The graphic he uses shows iPads , so I don't know if that means the limitations and application of app thinning applies only to iOS , as several sites are reporting.

If it applies to OS X El Cap, unfortunately it would include my late 2012 iMac, if I'm understanding the slide properly.

 

[koyoot]

What I really hope for is that MEtal will bring Asynchronous Shaders to the Mac, and... dual GPU support for Mac Pro.

That will be huge.

 

[antonis]

What I really hope for is that MEtal will bring Asynchronous Shaders to the Mac, and... dual GPU support for Mac Pro.

That will be huge.

Regarding the dual GPU, isn't that a driver issue rather than the API's ? Is really Metal able to contribute to that ?

 

[Ferazel]

Regarding the dual GPU, isn't that a driver issue rather than the API's ? Is really Metal able to contribute to that ?

The second GPU in the nMP is pretty underutilized because the APIs currently don't take advantage of it. OpenGL didn't really have a good way to access the nMP's secondary GPU. After perusing the documentation it appears on desktop system you actually specify the device that you want to execute the command buffer with, which could potentially include the integrated GPU as an option or the second nMP's GPU. The good news, is that devs should now have a way to split the execution between these Metal devices. So on a MBP with a discrete GPU, you could keep the discrete GPU rendering shaders, but have the integrated GPU performing compute operations.

Another thing, is Apple could add an automatic load balancing mechanic to MetalKit. Devs could give up some control and maybe a small overhead to let the API manage the threads that need to be executed and balance the load between the GPUs properly. This is where stuff that Metal can make it easier for devs to take advantage of multiple GPU setups.

 

[antonis]

Another thing, is Apple could add an automatic load balancing mechanic to MetalKit.

Thanks for the clarifications. This is getting interesting. But, as I understand it, you are referring in a generic context of taking advantage of the unused GPU, even if it is a different one from the main, in order to throw to it some auxiliary tasks. I guess that would be better than nothing at all (as it is the situation currently).

Regarding the quoted text, I understand correctly that you are referring to a simulation of crossfire ? That is, load-balancing the rendering task of each frame between the 2 GPUs ? But wouldn't that be better done from the driver ? (considering that the driver works in even a lower level than Metal, and that is essential for such tasks regarding speed).

 

[antonis]

Double post.

 

[madeirabhoy]

Me too, I guess. Though I have a nice GPU in my 2011 iMac.
Oh well, I generally consider an upgrade after five years. So maybe I buy the next refresh of the iMac. Though dropping three grand for retina with top GPU is a pretty big pill to swallow. Hopefully the price has come down a bit in a year or so.

i suppose though, its not as annoying as some obselescence, as it wont stop me playing any game, just mean that ill be playing them on bootcamp rather than os x native, but it wont mean there's any game i wont be able to run that i might have.

 

[Ferazel]

Metal on iOS is only a modest 10% gain in real-world situations (http://www.anandtech.com/show/9223/gfxbench-3-metal-ios). This is also inline with some Mantle implementations (http://www.anandtech.com/show/7728/battlefield-4-mantle-preview). Granted, I would imagine/hope that iOS's OpenGL drivers are significantly better than OS X's. However, if we hold the same low-overhead API performance increase guidelines to OS X, Metal isn't the equalizer that we want it to be. Since the most optimized Mac games from dedicated houses run at best 20% slower than Window's DirectX, we're still behind the Windows curve. Especially, since some offenders can go 40-50% of their performance on Windows. Then stack the fact that Windows is also getting their own low-level API in DirectX 12, the numbers are probably just going to remain where they are currently, but the goal posts will shift 10% or so.

 

[koyoot]

Why did you took those benchmarks, as a mark how will OSX games behave in comparison to Windows?

Are there any benches from Android? Are there any Benches from Windows Phone? No. You completely forgot that Games on OSX are "muted" by 20% in comparison to Windows by drivers, OpenGL, system, game ports etc. Get rid of that, and add another 10%. Now - things look completely different?

 

[Ferazel]

Are there any benches from Android? Are there any Benches from Windows Phone? No. You completely forgot that Games on OSX are "muted" by 20% in comparison to Windows by drivers, OpenGL, system, game ports etc. Get rid of that, and add another 10%. Now - things look completely different?

I did say that the benchmarks were for iOS and Mantle respectively. I'm saying these are the improvements that other OSes have gained by switching to a low-level API from a higher-level one. I think it's reasonable to expect similar results from Metal. I don't think anyone knows the actual OS X platform overhead of OpenGL besides the OS X OpenGL developers (of which there are few). I think that 10% is actually a good baseline improvement that we should be expecting. I just wanted to set some realistic expectations of how much we should be expecting from the Metal change in terms of performance, not 10x improvements that Apple is advertising.

Unless you're gaming on a low voltage Mac (MacBook or Air) or playing with low quality settings, I don't think the game is going to be CPU limited anyways. Sure you'll gain some performance from reducing the CPU batch and feeding the GPU more (10% for platforms that have had the switch so far). The fact is that most games are not frame rate limited based on this metric.

From what I understand, most games use a double-buffer rendering pipeline. So while the GPU is processing the current frame, the CPU works on the next frame. Let's say that it takes the CPU 10ms to run game logic and 6ms to create the rendering buffers. Then there is a frame time to transfer new vertex, texture, and shader information to the GPU which say takes 2ms. Then the GPU will process the vertex information to render the current frame's rasterized image which can be a basic 2-stage process of vertex and fragment shaders. Let's assume that this gpu rendering takes 18ms. So in total the CPU and GPU wait time is about 2ms, the GPU takes 18ms to process so in total the frame's time is still 20ms. This is a 50 frames per second rendering situation regardless of how much you reduce that 6ms batch time. The performance improvements of Mantle and Metal on iOS (10%) I assume are at the driver level and being able to being fed instructions from the CPU in a more efficient manner (the 2ms time from above). The good news about reduction in batch time is that NEW games can spend more time on game logic (physics, AI, etc.) or sending more geometry batches without risk of reducing the total frame time. Or the CPU can just relax while the GPU is processing, thus reducing battery consumption.

Anyways, my goal was to set some reasonable expectations for performance increases not claim that I have developer knowledge of the overheads of OSX OpenGL vs Metal. Although to be honest, my reason for being here is to see if I can acquire some.

 

[Dirtyharry50]

I did say that the benchmarks were for iOS and Mantle respectively. I'm saying these are the improvements that other OSes have gained by switching to a low-level API from a higher-level one. I think it's reasonable to expect similar results from Metal. I don't think anyone knows the actual OS X platform overhead of OpenGL besides the OS X OpenGL developers (of which there are few). I think that 10% is actually a good baseline improvement that we should be expecting. I just wanted to set some realistic expectations of how much we should be expecting from the Metal change in terms of performance, not 10x improvements that Apple is advertising.

Unless you're gaming on a low voltage Mac (MacBook or Air) or playing with low quality settings, I don't think the game is going to be CPU limited anyways. Sure you'll gain some performance from reducing the CPU batch and feeding the GPU more (10% for platforms that have had the switch so far). The fact is that most games are not frame rate limited based on this metric.

From what I understand, most games use a double-buffer rendering pipeline. So while the GPU is processing the current frame, the CPU works on the next frame. Let's say that it takes the CPU 10ms to run game logic and 6ms to create the rendering buffers. Then there is a frame time to transfer new vertex, texture, and shader information to the GPU which say takes 2ms. Then the GPU will process the vertex information to render the current frame's rasterized image which can be a basic 2-stage process of vertex and fragment shaders. Let's assume that this gpu rendering takes 18ms. So in total the CPU and GPU wait time is about 2ms, the GPU takes 18ms to process so in total the frame's time is still 20ms. This is a 50 frames per second rendering situation regardless of how much you reduce that 6ms batch time. The performance improvements of Mantle and Metal on iOS (10%) I assume are at the driver level and being able to being fed instructions from the CPU in a more efficient manner (the 2ms time from above). The good news about reduction in batch time is that NEW games can spend more time on game logic (physics, AI, etc.) or sending more geometry batches without risk of reducing the total frame time. Or the CPU can just relax while the GPU is processing, thus reducing battery consumption.

Anyways, my goal was to set some reasonable expectations for performance increases not claim that I have developer knowledge of the overheads of OSX OpenGL vs Metal. Although to be honest, my reason for being here is to see if I can acquire some.

Apple is not advertising 10x improvements.

"Federighi stated that “running a game in Metal instead of the standard Open GL graphics standard gives an app 50 percent improvement in rendering performance, 40 percent improvement in efficiency (which gives a better battery life), and 10 times improvement in drawing performance.”"

That's up to 10 times improvement in drawing performance. That does not of course translate to a 10x improvement in overall performance of any game.

Otherwise, I read quite a number of assumptions above. Ultimately, as I was talking about earlier, people cannot accurately be making predictions about frame rates nor overall performance at this time. There's no data to support these ideas and it is very important to remember there are many variables in play, such as the skill and effort applied in writing low level routines for optimal performance that were not possible before when much of this functionality was inaccessible.

This does not go so far as writing directly to the hardware in assembly language as was the practice many years ago to greatly improve performance but the basic idea is the same. Developers can more directly access the hardware via the Metal API. I had a friend who used to write assembly programs to run signs and he used to jokingly refer to that as being where the software meets the metal. I am sure this is why they named Metal what they did. Maybe everyone already had that figured out but if not, that's why I'd bet.

In skilled hands, I think Metal is going to enable developers for OS X to deliver some impressive results. Beyond that I could not begin to venture any guesses that would try to nail that down to FPS or what the CPU is doing relative to the GPU in SPECIFIC terms because it is going to vary by game.

 

[Dirtyharry50]

Excellent news. This article that just got dumped in my mailbox is further exciting me:

WWDC 2015: Why Metal for Mac OS X is a Big Deal for Mac Gaming

...we spoke with Aspyr Media’s Director of Technology, Jez Sherlock, to see what he thought about Metal, how it will impact the future of Mac gaming, and how Aspyr is incorporating the technology into their own titles.

Q: What are the potential drawbacks for gamers?

JS: If there is a drawback it relates to OpenGL. In some cases there will be a need to support legacy solutions. This is time consuming for us and eats into time we could put to better use (such as focusing on Metal), which in turn is a drawback to the end user. The goods news for everyone is that Aspyr has been thinking about this problem for some time.
​

This one here worries me. I own a lot of Mac games I've yet to get to of varying ages. I expect Aspyr and Feral will do their best to take care of us on this. I also expect Blizzard will given their games are all ongoing online experiences. Likewise, I am not worried about the various MMOs, etc out there for Mac.

What I am worried about is all the other stuff from various other companies, most especially the indie developers of the world.

I guess I should try to think positively and embrace the reduction of my backlog whether I want to or not and start playing games that are more recent anyway.

I also wonder about the impact of this on Wine and Wineskin.

Maybe I am misunderstanding that comment. That's always a possibility. ;-)

 

[antonis]

I guess wine could benefit a big time if updated / rewritten to convert DirectX to Metal instead of OpenGL (which, as it seems so far, doesn't seem to be updated much in the upcoming OS anyway).

Personally, I'm not very optimistic about blizzard as their products have the wider gap between windows and os x versions regarding performance, but I wish they'd prove me wrong and adopt metal as soon as possible.

For indie games, now that's interesting. I think we'll get some nice surprises from this scene as - sometimes - they prove to be more versatile in new technology adoptions, than big companies.

 

[Dirtyharry50]

I guess wine could benefit a big time if updated / rewritten to convert DirectX to Metal instead of OpenGL (which, as it seems so far, doesn't seem to be updated much in the upcoming OS anyway).

Personally, I'm not very optimistic about blizzard as their products have the wider gap between windows and os x versions regarding performance, but I wish they'd prove me wrong and adopt metal as soon as possible.

For indie games, now that's interesting. I think we'll get some nice surprises from this scene as - sometimes - they prove to be more versatile in new technology adoptions, than big companies.

Blizzard has already indicated that they are updating WoW with Metal in a future patch. I think they will do it for all their games on Mac considering their history of supporting the Mac platform. I am confident they are making money because otherwise they would not bother.

I don't think the future looks bright for Wine personally but we'll see.

 

[mysticalos]

The GPU support of metal is already known, at least as of DP1, whether or not that will be expanded upon is another story. But taken from driver by netkas, it looks like this

"Nvidia – GeForce gtx 4xx and newer
Intel – HD4000 and newer (ivy bridge and newer)
AMD – HD7000 and newer"

Blizzard has also confirmed this with their own post stating what gpus they will support metal on
http://us.battle.net/wow/en/forum/topic/18000263457?page=4#61

"We will support Metal for these video cards:
Intel HD 4000 and later
NVidia GeForce 600 and later
AMD Radeon R9 M290 and later
AMD Radeon HD7000 and later
AMD Fire Pro D300 and later"

Which matches netkas findings in drivers. There technically weren't any nvidia 400 or 500 on mac so 400 or newer is basically 600 and 700.
Whether or not 900/maxwell will be supported by a future update is a mystery as of right now though. Currently only up to 700 is supported which could leave some mac pros using 970/980 cards out of the fun unless nvidia continues their amazing web driver support in 10.11 with working metal. Or apple does it's usual switch to nvidia after an AMD cycle and updates OS X for newer 900 series hardware.

 

[Eric5h5]

Maybe I am misunderstanding that comment. That's always a possibility. ;-)

Yes...you're acting like Metal will somehow stop your old games from working, but that's not what happens. The comment is about supporting both OpenGL and Metal for new games, which takes more time. They'd rather dump OpenGL entirely, but it's probably too soon for that considering that only new(ish) Macs will be able to use Metal. Unless they're porting a game which will only run on the latest Macs anyway, in which case they can safely drop OpenGL for that port.

--Eric

 

[Dirtyharry50]

Yes...you're acting like Metal will somehow stop your old games from working, but that's not what happens. The comment is about supporting both OpenGL and Metal for new games, which takes more time. They'd rather dump OpenGL entirely, but it's probably too soon for that considering that only new(ish) Macs will be able to use Metal. Unless they're porting a game which will only run on the latest Macs anyway, in which case they can safely drop OpenGL for that port.

--Eric

Well, I am not acting like anything. I am just wondering what is meant by this comment in particular which I find vague:

"If there is a drawback it relates to OpenGL. In some cases there will be a need to support legacy solutions."

I am not sure precisely what he meant by that. He could easily and more clearly have said legacy systems or legacy hardware if that was what he meant thus indicating a need to create both OpenGL and Metal versions of some particular game. He didn't elaborate on that though so I am not sure how you'd be able to do so.

Are you a games developer who is intimately familiar with the Metal API and the implementation of Metal on OS X such that you can speak with authority and state that under no circumstances will the introduction of Metal on OS X have any impact whatsoever on any existing title that pre-El Capitan ran fine with OpenGL?

I'm not saying you're wrong because I don't know or I wouldn't have posted what I did. I am questioning though how it is that you know this, what is the basis of your statement? Where does your info come from that I could read it and understand better, in more detail, what is going on with this?

 

[marksatt]

Well, I am not acting like anything. I am just wondering what is meant by this comment in particular which I find vague:

"If there is a drawback it relates to OpenGL. In some cases there will be a need to support legacy solutions."

For some games it will be necessary to support Yosemite & possibly even Mavericks, which means continuing to support OpenGL for those titles. That requires development effort that then can't be applied to maximising Metal support. In reality there's always a tradeoff to be made between backward-compatibility to reach a larger audience and requiring newer features that might reduce it.

Existing OpenGL games will continue to work just fine but I wouldn't expect many to be reported to Metal until it has become the established & dominant API in use & there's a business case for doing the work.

 

[Eric5h5]

Are you a games developer

Yes, and while not "intimately familiar" with Metal, I do know how these things work in general. What you're suggesting would be like if Apple, back during the PPC->x86 transition, just dumped PPC right away and expected that somehow everybody would switch to x86 immediately. But as we all know, the transition took years and PPC was phased out gradually. Likewise, OpenGL may go away eventually, but it will take a while. The full quote is this:

If there is a drawback it relates to OpenGL. In some cases there will be a need to support legacy solutions. This is time consuming for us and eats into time we could put to better use (such as focusing on Metal), which in turn is a drawback to the end user.

It's very clear to me. Simplifying this a bit results in: "The drawback is that it takes time to support OpenGL. We could do ports faster if we dropped that and only used Metal." But, as I said, it's too soon for that. Let's see where things are a few years from now when presumably the majority of the userbase is on 10.11 or later. I wouldn't be surprised to see most games using Metal by then.

--Eric

 

[Dirtyharry50]

For some games it will be necessary to support Yosemite & possibly even Mavericks, which means continuing to support OpenGL for those titles. That requires development effort that then can't be applied to maximising Metal support. In reality there's always a tradeoff to be made between backward-compatibility to reach a larger audience and requiring newer features that might reduce it.

Existing OpenGL games will continue to work just fine but I wouldn't expect many to be reported to Metal until it has become the established & dominant API in use & there's a business case for doing the work.

Thanks marksatt. That is good news so far as all the stuff I already own goes. I'm glad to hear it.

 

[Dirtyharry50]

Yes, and while not "intimately familiar" with Metal, I do know how these things work in general. What you're suggesting would be like if Apple, back during the PPC->x86 transition, just dumped PPC right away and expected that somehow everybody would switch to x86 immediately. But as we all know, the transition took years and PPC was phased out gradually. Likewise, OpenGL may go away eventually, but it will take a while. The full quote is this: /snip

--Eric

Okay, thanks. This stuff is clearer to me now so far as my question and worries were concerned. I do still wonder though if there is rapid adoption of Metal in graphics applications and games how many versions of OS X going forward will continue to support OpenGL especially where Apple cranks them out annually. I better get playing the old stuff now not because there is impending doom but because I am so behind I need a lot of time even if they are slow about dropping it.

I can see it now. I'll be playing old OS X games in Yosemite running in a Parallels VM because I didn't get around to too much stuff before OpenGL went away.

 

[whooleytoo]

Yes, and while not "intimately familiar" with Metal, I do know how these things work in general. What you're suggesting would be like if Apple, back during the PPC->x86 transition, just dumped PPC right away and expected that somehow everybody would switch to x86 immediately. But as we all know, the transition took years and PPC was phased out gradually. Likewise, OpenGL may go away eventually, but it will take a while.

Isn't that a slightly flawed comparison - given that many/most Mac gamers can upgrade to El Capitan for free and get a Metal-capable Mac, with no broken backwards compatibility (unless Apple actually exclude OpenGL from El Capitan). Upgrading a PPC to x86 was a bit more tricky and expensive.

Obviously older Macs won't support El Capitan and/or Metal, but many of those those Macs would be below the required/recommended gaming specs for modern games anyhow.

I'd have thought OpenGL would start being dropped for new ports pretty soon after El Capitan's release. Given how willing and quick Apple users generally are to upgrade - and the fact it's free - by the time the first post-El Cap port is ready the penetration of the new OS would probably be quite high.

 

[koyoot]

Maybe Im making a bit of a mistake, but...

You guys should see this: https://forums.macrumors.com/thread...erence-thanks-to-low-overhead-of-cpu.1895894/

OSX is lagging currently in about 20% in performance to Windows 8. Metal will completely get rid of the gap.

That makes Apple claim of 50% performance improvement possible.

 

[Stacc]

Maybe Im making a bit of a mistake, but...

You guys should see this: https://forums.macrumors.com/thread...erence-thanks-to-low-overhead-of-cpu.1895894/

OSX is lagging currently in about 20% in performance to Windows 8. Metal will completely get rid of the gap.

That makes Apple claim of 50% performance improvement possible.

You are jumping the gun way too much here. Games that support metal will likely run better on OS X, but just because there are performance increases on directx 12 does not mean they will translate to OS X. Remember, people have been optimizing games for windows for decades, not so much OS X and openGL/metal. Its a completely different software stack, including a different set of drivers than those available for windows.