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.
The Vega RX Thread (Rumors and Info)
- Thread starter fendersrule
- Start date
- Sort by reaction score
Its not entirely correct. Microsoft has been working on next generation API that would be suitable for Windows and Xbox One. It was called: DirectX 11.3. Then Andersson from DICE came to AMD and pushed them to develop low-level API, so it can be used on desktops.
That is what happened. Then AMD decided to Open Source it. Microsoft decided to use the core, low-level features, out of its feature set: low-overhead, Explicit mutlithreading, explicit multi-GPU capabilities. The rest is DX11.3, on top of Mantle features.
Similar thing happened with Metal.
Vulkan is a different story(slightly). Vulkan is the OpenGL Next, and in fact it is from the ground up - Mantle API.
Source? There is almost no similarity between Mantle and Metal, other than they are both a clean start.
https://en.wikipedia.org/wiki/Mantle_(API)
AMD originally developed Mantle in cooperation with DICE, starting in 2013.
https://en.wikipedia.org/wiki/Metal_(API)
Metal has been available since June 2, 2014[2] on iOS devices using the Apple A7 or later.
There is no way it took Apple less than a year to design and implement Metal.
Are there? You, yourself, have posted the exact similarity, I was talking about. Haven't you?
How does multi GPU work in Metal? Is it somewhat similar to DX12? Is it possible that two GPUs reporting to Metal API are able to work on one part of the display? I wonder where this came from...
Who says that they had to incorporate all of it?
It is the same situation as was with DX11.3.
Microsoft has been developing DX11.3 for quit some time. Then they decided that they want low-level control to give to developers, and rather than rewriting the API - they added core features, out of Mantle Feature set to DX11.3 and called it: DX12. Those low-level features are: low-overhead, explicit multithreading, explicit multi-GPU capabilities, and few more. All of those features are shared in all of APIs that share Mantle core: Metal, Vulkan, and DX12.
Funniest part. DX11.3 is DX12, but without those low-level features.
Apple did not had to write Metal in just one year. They could've added Mantle to what they already had. Judging by the state of Metal, and what we have even now - not much. Metal is still the weakest API of them all, but has the biggest potential.
Best API does not mean that it is open source and widely available.
highest potential in terms of explicitness, and easy of use. Highest potential of features, and ease of use.
Again, please explain exactly how Metal is similar to Mantle. How can you do multi-GPU stuff in Metal when you don't have any kind of GPU synchronization primitive (e.g. fences)? Mantle is about as low-level as an interface can get. Metal is much more like OpenGL ES (you bind resources directly, you don't have to manage your own memory allocations, etc). Or are you suggesting things like command buffers, command queues, precompiled objects is enough to claim that Metal is based on Mantle?
Because you say so?
https://community.amd.com/community/gaming/blog/2015/05/12/on-apis-and-the-future-of-mantle
AMD appears to be disagreeing with you.
LOL no!
MAntle is dead. AMD is saying that people should move on to the two other API that are still being worked on.
Are you even understanding what you're reading or do you just go by what the title seems to imply?
Serously, man, just stop it already.
It specifically says that if you are developer interested in Mantle feature set 1.0 you better focus on DX12 and OpenGL Next.
Its only your problem that you resist reality. But let me help you. This is what AMD said Mantle and relation to DX12 and OpenGL Next/Vulkan, in this blog post.
"Proud moments also call for reflection, and today we are especially thoughtful about Mantle’s future. In the approaching era of DirectX® 12 and the Next-Generation OpenGL Initiative, AMD is helping to develop two incredibly powerful APIs that leverage many capabilities of the award-winning Graphics Core Next (GCN) Architecture."
You have to bring proofs to substance your claims that DX12 is not based on Mantle. Otherwise it is not me who is the problem here. Its only you arguing constantly with me, while having zero clue what you are talking about.
Man oh man... Amd is helping, guess what koo, so does nvidia! Are you now saying that nvidia as adopted mantle? Now don't be silly...
And the proofs have been presented to you in many other threads previously, you just fail to understand what you read.
AMD is helping to develop two APIs, because they have gave out Mantle to Microsoft and Khronos, and both companies are using it in their APIs.
Show proofs that show that low-level features are not coming from Mantle. Show proofs that Microsoft did not used Mantle as base for DX12. Show proofs that DX11.3 is not DX12 without low-level Mantle features: low-overhead, explicit multithreading, explicit multi-GPU.
AMD has said that plainly: "If you are interested in Mantle 1.0 focus on DX12 and OpenGL Next". I do not know why you argue against it. It is plain as hell.
You'll notice that they didn't mention Metal there at all, so I'm not sure why you keep insisting that Metal is based on Mantle as well.
Today you have been proven that DX12 is based on Mantle core features. I hope this graphic is settling that Metal also has those features used in its core.
"Strong Mantle adoption primes AMD to win API inflection points"
The inflection points are on that straight arrow. And one of them is Metal.
Today you have been proven that DX12 is based on Mantle core features. I hope this graphic is settling that Metal also has those features used in its core.
"Strong Mantle adoption primes AMD to win API inflection points"
The inflection points are on that straight arrow. And one of them is Metal.
Okay, we can just agree to disagree. Apple probably had Metal in development for several years before they announced it, and this reads more that AMD is trying to take credit for something that they shouldn't be. I'd also suggest that "low overhead" and "multi-threaded" are not revolutionary concepts that AMD invented, in fact it's a pretty logical next step from DX11/GL4. DX11 was already pretty low overhead (i.e. NVIDIA's DX11 driver beats AMD's Mantle driver in directed overhead tests like StarSwarm) and had multi-threaded command lists (even if AMD never implemented them, NVIDIA definitely did). Precompiled objects are also a pretty logical next step, in fact OpenGL already had them for shaders (and extending them to textures makes a lot of sense). The implementation of all those points are just so fundamentally different between Metal and the other next-gen APIs that I still think it's a real stretch to claim that Metal and Mantle are similar.
I understand that this is your opinion, about situation, not a fact. Similar logic you can use for DX12 and Vulkan, but I rather base my thinking on facts, not opinions.
The facts are simple. Metal, Vulkan and DX12 are based on Mantle feature set. You have enough proofs for this already.
They are derivatives of the same ideas but consoles have used 'close to the metal' optimised APIs since the mid 90s, and especially since the first Xbox.
For years PC gamers asked why can't we have console level optimisation? Then finally game developers pushed AMD and Microsoft to provide solutions.
In other news.
New version of OpenGL is out. Not on macOS of course.
New GeForce drivers have unlocked the performance of Titan Pascal cards. They now perform like Quadro cards. Why didn't Nvidia do that before?
Because they could get away with the tier scam.
About Low-level APIs you do not have to tell me that
.Nvidia released new Titan Pascal drivers because of Vega. They had to.
Are you sure that Nvidia didn't release Pascal drivers so that they could sell some Pascal cards?
Or, NVIDIA decided to enable some of their Quadro optimizations on TITAN to put the final nail in the coffin re: Vega.
So what will happen when AMD will release drivers that will improve performance of Vega out of reach for Titan Xp?
There are rumors that Nvidia will stop releasing webdrivers for macOS when high sierra is released. Could have to do with Metal 2 compatibility or maybe Nvidia doesn't think it will get Apple's business anytime soon.
So what will happen when AMD will release drivers that will improve performance of Vega out of reach for Titan Xp?
Get back to me when that happens, sure. AMD is claiming Vega will trade blows with a 1080, not a 1080 Ti or TITAN Xp.
[doublepost=1501712876][/doublepost]
NVIDIA stopped doing web driver releases for beta OSes a while back, probably because there were so many of them. We'll wait and see once 10.13 is actually released, I would expect that there will still be a web driver for that OS version. Things might've changed so much that the web driver requires more work etc.
It is odd if Nvidia and Apple have still not improved relations, in light of the Mac Pro press conf. I think many people consider the ability to choose a Nvidia card a big selling point of a 'modular mac pro'. Another MP that is all AMD all the time would be a big letdown.
1000 words...So what will happen when AMD will release drivers that will improve performance of Vega out of reach for Titan Xp?