How Mantle, DX12 and Metal make huge gains achievable via software alone...
I had said in a previous post that I had felt it WAS indeed possible for Microsoft to get the scale of performance gains they were advertising in DX12 IF the prior implementations of DirectX were flawed enough.
People, some who even proclaim to actually have degrees in Computer Science, said it was impossible. I really do wonder what passes for a degree in Computer Science these days. That statement is full of spectacular idiocy.
Anyway, I stumbled across this article thanks to someone I follow on Twitter. The author is one of the "fathers" of DirectX at Microsoft and from what I've read of his in the past, knows what he is talking about.
The important things acknowledged by the article are that DirectX and OpenGL are essentially held back by decisions that made more sense in a time when GPU's were FAR less powerful than they are today. He even muses over how it is a wonder that API's like DirectX and OpenGL ever worked at all and that we owe their very existence to a very small group of geniuses able to approximate calculations required well enough and fast enough to yield that something that looked like real time rendering of graphics.
Basically though, the important factor here is that GPU's scaled up in power in different ways and far faster than CPU's. This uneven growth has resulted in API's like DirectX and OpenGL becoming horribly inefficient. These API's have been since before most people had personal computers. Since well before cell phones were common place. And they have simply been built upon more and more. And, when you consider the point above about making it work being dependent on a small number of geniuses, you have to assume that the most critical elements have not evolved much since their inception.
Honestly, if you actually know what you're talking about. When you realize how old DirectX actually is, and where it started, the notion that a substantial rewrite might be able to yield a 200% performance increase... it actually stops sounding crazy and starts sounding a bit underwhelming. And that is where I am on this. The fact that the gains are so small is undoubtedly the work of very smart people refactoring very archaic code over previous iterative release of DirectX. And as a result, the thought of this overhaul yielding JUST double the performance actually sounds a little lacklustre. I imagine true performance gains will vary based on hardware and proving out those gains will require the existence of games built using both the old and new versions.
Out of the 3, Metal, Mantle and DX12... I think only DX12 has a future. DX12 will replace DX11. The others stand alone as brand new API's. Mantle has already failed to get traction. So unless Apple simply kills off support for OpenGL somehow, I see Metal going the same way.
Finding the chatter on forums about this hilarious. Everybody hates the move away from OpenGL. Proprietary API's are quickly taking over. OpenGL was the last standout for open graphics API's. So, while I'm happy that these API's are evolving, I do share in the sadness of the death of an era.
People, some who even proclaim to actually have degrees in Computer Science, said it was impossible. I really do wonder what passes for a degree in Computer Science these days. That statement is full of spectacular idiocy.
Anyway, I stumbled across this article thanks to someone I follow on Twitter. The author is one of the "fathers" of DirectX at Microsoft and from what I've read of his in the past, knows what he is talking about.
The important things acknowledged by the article are that DirectX and OpenGL are essentially held back by decisions that made more sense in a time when GPU's were FAR less powerful than they are today. He even muses over how it is a wonder that API's like DirectX and OpenGL ever worked at all and that we owe their very existence to a very small group of geniuses able to approximate calculations required well enough and fast enough to yield that something that looked like real time rendering of graphics.
Basically though, the important factor here is that GPU's scaled up in power in different ways and far faster than CPU's. This uneven growth has resulted in API's like DirectX and OpenGL becoming horribly inefficient. These API's have been since before most people had personal computers. Since well before cell phones were common place. And they have simply been built upon more and more. And, when you consider the point above about making it work being dependent on a small number of geniuses, you have to assume that the most critical elements have not evolved much since their inception.
Honestly, if you actually know what you're talking about. When you realize how old DirectX actually is, and where it started, the notion that a substantial rewrite might be able to yield a 200% performance increase... it actually stops sounding crazy and starts sounding a bit underwhelming. And that is where I am on this. The fact that the gains are so small is undoubtedly the work of very smart people refactoring very archaic code over previous iterative release of DirectX. And as a result, the thought of this overhaul yielding JUST double the performance actually sounds a little lacklustre. I imagine true performance gains will vary based on hardware and proving out those gains will require the existence of games built using both the old and new versions.
Out of the 3, Metal, Mantle and DX12... I think only DX12 has a future. DX12 will replace DX11. The others stand alone as brand new API's. Mantle has already failed to get traction. So unless Apple simply kills off support for OpenGL somehow, I see Metal going the same way.
Finding the chatter on forums about this hilarious. Everybody hates the move away from OpenGL. Proprietary API's are quickly taking over. OpenGL was the last standout for open graphics API's. So, while I'm happy that these API's are evolving, I do share in the sadness of the death of an era.
Comments
Post a Comment