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.
2016 nMP
- Thread starter ManuelGomes
- Start date
-
- Tags
- nmp
- Sort by reaction score
- Status
What I've understood, Metal is the way to access multi GPU's. When Metal is ready (now it is not) they could implement an openGL overlay for it, and it will use multi-GPU without the original software knowing where the speed is coming from. Metal is also the key for HSA. AMD is not making CPU for Apple, but an APU. And Metal is everything to use it properly.
Metal is Mantle as much as DX12 is. They share a lot of similar ideas, but differ in implementation. AMD doesn't care if they use little or a lot of Mantle API, as long as they follow the principal. Because that's what GCN is built for. AMD can provide a driver, and it will fly.
Nope. Metal has nothing for multiple GPUs beyond what was already there with OpenGL.
Metal wasn't about AMD. It was about Apple's GPUs on their A series processors. There's a lot of stuff Metal is missing for AMD GPUs that Mantle is not.
[doublepost=1457299893][/doublepost]
Metal for iOS was in development before 2014. It's likely that it was in development when Mantle was first revealed.
I think AMD looks better if they take credit for everyone moving this direction, but it was something everyone was working on because everyone had the same set of problems, and everyone knew what changes had to be done to solve those problems. The reason that so many people announced their own solutions is because no one had formed a standard yet.
AMD brought Mantle up to the public in 2013. It had been under development three years before that. There was already game demos available when it came to public, so it really wasn't born in 2013. I believe, that when AMD was designing CGN, they noticed the need for a new API. So, around 2010-2011 AMD started to introduce its ideas to few game developers and to Microsoft.. and what I presume, Apple too. Apple took the concept.. and even AMD has stated (Feb 2014), ".. we do hope that Mantle becomes an industry standard. We'll be releasing a public SDK later this year, and hope that others adopt it. If they don't adopt it itself, then we hope they adopt APIs similar to it that become an industry standard for PC gaming." - - AMD explained that while it's currently dependent on GCN, Mantle "utilizes a certain level of meaningful hardware abstraction that could eventually allow it to be applicable to other architectures."
"Such applicability is necessary in an ecosystem we hope to grow as an industry standard in the years ahead."
So it was built to be open for other architectures. Such as iOS Metal and its ARM/PowerVR chips.
Last edited:
I don't know what else to tell you besides that if you look at them, they're really aren't similar at all. They don't even have the same features. As has come up here Metal doesn't even have built in support for multi GPU rendering.
Saying they came out around the same time and they have similar names isn't enough to say they are related. If you look at them, it's pretty obvious they don't have any common foundation.
Btw, GFXBench has now Metal support, so direct comparison between openGL and Metal is possible. The speed gains in Manhattan and T-Rex are enormous.
Intel HD4000 openGL / Metal
Manhattan (23.39 fps) / (68.73 fps)
T-Rex (33.7 fps) / (114.62 fps)
Nvidia 650M openGL / Metal
Manhattan (26.4 fps) / (70.59 fps)
T-rex (49.98 fps) / (114 fps)
Late 2013 Mac Pro 3.7 GHz , 16GB, 2 x AMD FirePro D300 2048 MB
openGL / Metal
Manhattan 59.482 fps / 119.93 fps
T-rex 59.89 fps / 119.285 fps
It seems that D300 hit the 60Hz wall in openGL and 120Hz in Metal..
UPDATE: offscreen results for D300
Manhattan 141.26 fps / 221.953 fps
T-rex 377.51 fps / 400.346 fps
[doublepost=1457307606][/doublepost]
Ok, let me say: Metal is unfinished. Version 1. A public alfa. OS X is going to have more complex/complete API. I hoped that they'd add second version for El Captain in .4 or .5.. but now I think, it is one of the new features of OS X 10.12.
Intel HD4000 openGL / Metal
Manhattan (23.39 fps) / (68.73 fps)
T-Rex (33.7 fps) / (114.62 fps)
Nvidia 650M openGL / Metal
Manhattan (26.4 fps) / (70.59 fps)
T-rex (49.98 fps) / (114 fps)
Late 2013 Mac Pro 3.7 GHz , 16GB, 2 x AMD FirePro D300 2048 MB
openGL / Metal
Manhattan 59.482 fps / 119.93 fps
T-rex 59.89 fps / 119.285 fps
It seems that D300 hit the 60Hz wall in openGL and 120Hz in Metal..
UPDATE: offscreen results for D300
Manhattan 141.26 fps / 221.953 fps
T-rex 377.51 fps / 400.346 fps
[doublepost=1457307606][/doublepost]
Ok, let me say: Metal is unfinished. Version 1. A public alfa. OS X is going to have more complex/complete API. I hoped that they'd add second version for El Captain in .4 or .5.. but now I think, it is one of the new features of OS X 10.12.
Last edited:
It could, but it still doesn't really make any sense...
To reply to other things you've said... The driver CPU overhead problem has been known about for a long long time. I think Apple first mentioned it in 2005. It really started to take off as an issue with GPGPU. On mobile it started to be more of a problem because you have limited CPU and limited battery. AMD didn't invent this as an issue, and they didn't invent the fix either. Everyone knew what the fix was going to look like. It's just that moving open standards can take a long time.
I think Apple thought that any open standard to fix the problem would take too long, and that they could fix it themselves quicker if they didn't participate in the open standards. On mobile, that turned out to be true. They beat Vulkan to market by a longshot.
On Mac, is that going to be true? Vulkan is out now, and sure Metal beat it by 6 months, but Metal has a fraction of the features and no support from developers. If you've seen the pace at which Apple adds new features to their graphics APIs... let's just say I wouldn't hold my breath for Metal in 10.12 to fix everything. Although I think Apple will spin it and give it their best try.
The benchmarks you've posted are really impressive, and a big reason AMD bet their hardware on these sorts of changes (which I think is going to pay off well for them after Nvidia crippled their GPUs), but none of this requires Metal.
Metal is not as explicit API as Vulkan or DX12. In other words, Metal could use Vulkan if Apple would want to. It is layer of abstraction for Applications, but not as complicated and high level as DirectX11 was.
Metal uses Mantle DRIVER. Not exactly copies Mantle 1:1. It is similar thing to DX12. Core of it has been added to DX12. With progress, Metal will mirror features from Vulkan, as Vulkan is OpenGL Next/OpenGL 5.0, or it will simply integrate Vulkan. Vulkan IS Mantle. Most of the code is 1:1 copied from Mantle, even commands for Vulkan are similar, and differ by 3 first letters from Mantle.
As for AMD GPU performance. Graphics side was on older APIs(DirectX11/OpenGL) bottlenecked by not enough amount of Draw Calls, and few other problems that they created. Vulkan and DirectX 12 lifted that, and thats why we currently see that R9 390X ties in games using DX12 with GTX 980 Ti, because it has such similar amount of raw compute horsepower. Fury is faster at this stage than Titan X, because it has 0.3 TFLOPs more compute power, and proper Asynchronous Compute capabilities. SIMD vs SIMD performance should be similar from now on.
Metal uses Mantle DRIVER. Not exactly copies Mantle 1:1. It is similar thing to DX12. Core of it has been added to DX12. With progress, Metal will mirror features from Vulkan, as Vulkan is OpenGL Next/OpenGL 5.0, or it will simply integrate Vulkan. Vulkan IS Mantle. Most of the code is 1:1 copied from Mantle, even commands for Vulkan are similar, and differ by 3 first letters from Mantle.
As for AMD GPU performance. Graphics side was on older APIs(DirectX11/OpenGL) bottlenecked by not enough amount of Draw Calls, and few other problems that they created. Vulkan and DirectX 12 lifted that, and thats why we currently see that R9 390X ties in games using DX12 with GTX 980 Ti, because it has such similar amount of raw compute horsepower. Fury is faster at this stage than Titan X, because it has 0.3 TFLOPs more compute power, and proper Asynchronous Compute capabilities. SIMD vs SIMD performance should be similar from now on.
Unreal Tournament 4 will have Metal support within few weeks. Beta is already running.
https://forums.unrealengine.com/showthread.php?94321-Unreal-Engine-4-11-Preview
Update: WoW also is using Metal, and Starcraft 2 and HotS should be ready for April-May.
https://forums.unrealengine.com/showthread.php?94321-Unreal-Engine-4-11-Preview
Update: WoW also is using Metal, and Starcraft 2 and HotS should be ready for April-May.
Last edited:
At least the other way around is nearly there. Vulkan runs on Metal to supply Vulkan on OS X and iOS.
https://moltengl.com/moltenvk/
It's already happening. Thanks for the link. This is the future of Metal that I've been predicting. Altough I was thinking that Vulkan could be open source solution. Now Apple just needs to improve & fix Metal and its GPGPU capabilities (which are pretty much broken atm according to some developer sites).
[doublepost=1457342508][/doublepost]
Unity engine supports only SM4.0 shaders for Metal.
Last edited:
Well, it's obvious. Complete lack of SM 5.0 support for Metal, yet
To me it looks like hardware drivers report SM5.0 but API cannot use it.
Interesting. Is it just me, or is it similar to third party additions to Vulkan that are currently touted by Khronos?
http://www.amazon.com/Intel-Xeon-E5...e=UTF8&qid=1457360050&sr=1-4&keywords=xeon+v4
http://www.shopblt.com/item/intel-xeon-e5-2697-v4-lga2011/intel_bx80660e52697v4.html
First Broadwell-EP CPUs already available. Pretty steep prices, however, that 12 core is very tempting.
http://www.shopblt.com/item/intel-xeon-e5-2697-v4-lga2011/intel_bx80660e52697v4.html
First Broadwell-EP CPUs already available. Pretty steep prices, however, that 12 core is very tempting.
Wow! Maybe we'll get a Mac Pro refresh at the march event
LOL. I remember when it was going to be any day now or else. That was two years ago. It's OK to gloat over something so hard fought as this!?
Seriously though, I just went out and got me an Alienware Area 51 R2...very nice. In some ways it's nice to be back in the Windows ecosystem with a really nice box. Huge....but still quite. Doesn't make a whisper while playing games on insane either.
Wow! Maybe we'll get a Mac Pro refresh at the march event
Perfect time for Apple to get it done right before the new GPUs come out!
One big winning from the switch to Broadwell-EP and DDR4 is the immediate availability of 64GB RDIMM this means the updated nMP could offer upto 256GB ECC Ram from day 0.
About an nnMP launched at WWDC unless AMD hides some surprise it will not happen due Polaris delay or at least it could be only announced at WWDC with availability "some time later this year" which means October or November as soon AMD is delivering Polaris 11 in enough quantity.
I'm also intrigued by the possibility of Apple just ditching Xeon nMP and early adopting Zen, while its sounds bizarre it could be a good thing about performance.
About an nnMP launched at WWDC unless AMD hides some surprise it will not happen due Polaris delay or at least it could be only announced at WWDC with availability "some time later this year" which means October or November as soon AMD is delivering Polaris 11 in enough quantity.
I'm also intrigued by the possibility of Apple just ditching Xeon nMP and early adopting Zen, while its sounds bizarre it could be a good thing about performance.
No, add an "n". Are you trying to fight the reality distortion field?
The 2016 tube will be the "nnMP".
[doublepost=1457400451][/doublepost]
Broadwell-EP is available now... (https://forums.macrumors.com/threads/2016-nmp.1952250/page-28#post-22645411)
Or are you making a joke that Apple will continue in its penchant for offering last-generation components at premium prices?
That will certainly silence the people who have been pointing out that a Z840 already supports 2TiB of DDR4 RAM compared to 64 GiB for the MP6,1.
Mac professionals should be hoping that Apple realizes that the tube is a nice upgrade for a MiniMac, but is a disaster for power users and goes back to adding a dual socket form factor system to the lineup.
And they should be hoping that Apple sends a pink slip to "Phil my ass" for grievous damage to an iconic brand.
Last edited:
Only if you can really consider that thing a MacPro. Other than the name, I don't consider it a suitable replacement yet.
AMD was talking about their live webcast from GDC when they teased about "Capsaicin".
http://www.amd.com/en-us/press-releases/Pages/amd-radeon-to-2016mar07.aspx
Next Monday the 14th. " The webcast will feature an inside look at the latest technological advancements affecting the enthusiast and developer communities as they approach the burgeoning virtual reality market."
http://www.amd.com/en-us/press-releases/Pages/amd-radeon-to-2016mar07.aspx
Next Monday the 14th. " The webcast will feature an inside look at the latest technological advancements affecting the enthusiast and developer communities as they approach the burgeoning virtual reality market."
- Status