principal dev lead Shawn Hargreaves.
The truth is, DX12Ultimate, announced before the console was fully revealed - has a complete Machine Learning API overhead. Why would Microsoft then go on to release a machine, that does not fully utilize DX12Ultimate. Microsoft Assures everyone that
"if you need Machine Learning for your game, DirectML will meet your Machine Learning needs."
Machine Learning, in computer science - when applied to Graphics/Computer Workloads - sound's like some kind of far fetched solution that won't be possible for another hundred years. But ML saturation across all platform's and software is just around the corner according to Computer Science. And on such subjects, Computer Science never falters.
And within the DirectML doc's, they insinuate ML can be applied to shaders - not just a specific group of shaders but ALL shaders - shader's are standard rendering fare - which mean's the DX12U implementation of DirectML will support and boost all rendering techniques eventually. According to Computer Science, those gains are no less than exactly, precisely - 3 months away.
Direct Machine Learning (DirectML) is a low-level API for machine learning (ML).
docs.microsoft.com
"If you're counting milliseconds, and squeezing frame times, then DirectML will meet your machine learning needs.
Why does DirectML perform so well?
There's a good reason why you shouldn't just write your own convolution operator (for example) as HLSL in a
compute shader. The advantage of using DirectML is that—apart from saving you the effort of homebrewing your own solution—it has the capability of giving you much better performance than you could achieve with a hand-written, general-purpose compute shader for something like
convolution, or
lstm.
DirectML achieves this in part due to the Direct3D 12 metacommands feature. Metacommands expose a black box of functionality up to DirectML, which allows hardware vendors to provide DirectML access to vendor hardware-specific and architecture-specific optimizations. Multiple operators—for example, convolution followed by activation—can be
fused together into a single metacommand. Because of these factors, DirectML has the capability to exceed the performance of even a very well-written hand-tuned compute shader written to run on a breadth of hardware.
Metacommands are part of the Direct3D 12 API, although they're loosely coupled to it. A metacommand is identified by a fixed
GUID, while almost everything else about it (from its behavior and semantics to its signature and name) are not strictly part of the Direct3D 12 API. Rather, a metacommand is specified between its author and the driver that implements it. In this case, the author is DirectML. Metacommands are Direct3D 12 primitives (just like Draws and Dispatches), so they can be recorded into a command list and scheduled for execution together.
DirectML accelerates your machine learning workloads using an entire suite of machine learning metacommands. Consequently, you don't need to write vendor-specific code paths to achieve hardware acceleration for your inferencing. If you happen to run on an AI-accelerated chip, then DirectML uses that hardware to greatly accelerate operations such as convolution. You can take the same code that you wrote, without modifying it, run it on a chip that's not AI-accelerated (perhaps the integrated GPU in your laptop), and still get great GPU hardware acceleration. And if no GPU is available, then DirectML falls back to the CPU."
"Why DirectML
Many new real-time inferencing scenarios have been introduced to the developer community over the last few years through cutting edge machine learning research. Some examples of these are
super resolution, denoising, style transfer, game testing, and tools for animation and art. These models are computationally expensive but in many cases are required to run in real-time. DirectML enables these to run with high-performance by providing a wide set of optimized operators without the overhead of traditional inferencing engines.
To further enhance performance on the operators that customers need most, we work directly with hardware vendors, like Intel, AMD, and NVIDIA, to directly to provide architecture-specific optimizations, called metacommands. Newer hardware provides advances in ML performance through the use of FP16 precision and designated ML space on chips. DirectML’s metacommands provide vendors a way of exposing those advantages through their drivers to a common interface. Developers save the effort of hand tuning for individual hardware but get the benefits of these innovations.
DirectML is already providing some of these performance advantages by being the underlying foundation of
WinML, our high-level inferencing engine that powers applications outside of gaming, like Adobe, Photos, Office, and Intelligent Ink. The API flexes its muscles by enabling applications to run on millions of Windows devices today."