AMD's Radeon VII Supports DirectML - An Alternative to DLSS?

[BubbySoup]

Secodnly, DirectML is not an open standard, just like DirectX is not an open standard. It is am industry standard controlled by Microsoft.

It's open in as much that it is not locked to any particular GPU. It's closed in the fact that MS ultimately have control on the direction it takes.

 

[opethdisciple]

ehh, Nvidia supports DirectML as well, in fact they worked closely with Microsft to help define the specification and ensure the API is generalizable enough to benefit form dedicated acceleration hardware like Nvidia's TensorCores


Secodnly, DirectML is not an open standard, just like DirectX is not an open standard. It is am industry standard controlled by Microsoft.

Fair enough.

I think what I meant was that AMD seems to be more friendly when it comes to open source than Nvidia.

 

[BS Dave]

ehh, Nvidia supports DirectML as well, in fact they worked closely with Microsft to help define the specification and ensure the API is generalizable enough to benefit form dedicated acceleration hardware like Nvidia's TensorCores


Secodnly, DirectML is not an open standard, just like DirectX is not an open standard. It is am industry standard controlled by Microsoft.

Nvidia can do it but they’d rather you used their own DLSS which sounds very similar, and oh yeah you’ll need a new card for that...

 

[shankly1985]

Nvidia have a nice slide describing their support for DirectML
https://developer.nvidia.com/using-ai-slide



By leveraging Turing' and Volta's Tensor cores, DirectML is 8x faster than using direct compute on FP32 cores.

It was using FP16 compute

 

[D.P.]

Fair enough.

I think what I meant was that AMD seems to be more friendly when it comes to open source than Nvidia.

NVidia are also very friendly when it comes to opensource. For example, they beat AMD to the market with Vulkan compatible drivers, support more features in Vulkan and have far more resources available for Vulkan deelopers.


Nvidia always supports wide spread Open source standard where they make sense, but are equally happy supporting closed source standard like DX, or using a proprietary standard

 

[D.P.]

It was using FP16 compute

That execute on the Tensor cores, which is why there is such a large performance increase.
EDIT: Slides 57-58

 

[shankly1985]

That execute on the Tensor cores, which is why there is such a large performance increase.

Titan V has tensor cores?

Edit
Ignore seems it does

 

[opethdisciple]

NVidia are also very friendly when it comes to opensource. For example, they beat AMD to the market with Vulkan compatible drivers, support more features in Vulkan and have far more resources available for Vulkan deelopers.

Nvidia always supports wide spread Open source standard where they make sense, but are equally happy supporting closed source standard like DX, or using a proprietary standard

Maybe. It's just from what I've noticed AMD always seem to go down the open source route with their tech whilst Nvidia will choose proprietary standards such as Gysnc, Physx, GameWorks etc...

 

[shankly1985]

Based on the video also it seems performance isn't that much effected.
Also this

 

[4K8KW10]

We couldn’t write a graphics blog without calling out how DNNs can help improve the visual quality and performance of games. Take a close look at what happens when NVIDIA uses ML to up-sample this photo of a car by 4x. At first the images will look quite similar, but when you zoom in close, you’ll notice that the car on the right has some jagged edges, or aliasing, and the one using ML on the left is crisper. Models can learn to determine the best color for each pixel to benefit small images that are upscaled, or images that are zoomed in on. You may have had the experience when playing a game where objects look great from afar, but when you move close to a wall or hide behind a crate, things start to look a bit blocky or fuzzy – with ML we may see the end of those types of experiences.

This sounds similar to what Unlimited Detail will achieve:

 

[Wrinkly]

This sounds similar to what Unlimited Detail will achieve:

The toad still makes me smile.

 

[Nasher]

I think they is more chance devs using Microsoft own in house tech than Nvidia DLSS.

DirectX 12 vs Vulkan at the moment speaks for itself. Even though everyone knows Vulkan is the clear winning yet still devs keep on using DirectX 12

Both AMD and Nvidia can use it!!

That is key here!

Some developers actually went back to dx11 due to stability issues with dx12 (E.g. Hitman 2 uses dx11 even though 1 used 12).

 

[opethdisciple]

How does DirectML differ to DLSS? Why would Nvidia support both?

 

[shankly1985]

Some developers actually went back to dx11 due to stability issues with dx12 (E.g. Hitman 2 uses dx11 even though 1 used 12).

That isn't the fault of DX12 though. Its the devs and engine to get DX12 working. That is why i said above they is some really good DX12 games and then they is some really bad games.

 

[shankly1985]

How does DirectML differ to DLSS? Why would Nvidia support both?

Hard to say when we dont have games supporting either lol

Based on the images showing the two screenshot DirectML looks better.

Performance difference or realtime game they isn't much to compare.

 

[4K8KW10]

The toad still makes me smile.

Everything in Crysis 3 is impressive - the toad, the crocodile in the water, the real area lighting, the leaves of the vegetation... It has something that looks like active ray tracing....

 

[D.P.]

How does DirectML differ to DLSS? Why would Nvidia support both?

DlSS is one particular implementation of an algorithm. It actually operates his g the DirectML stack, at least on Windows.

Directly is just a middleware API, just like DirectX. Directly doesn't provide any algorithms, it just defined the instructions.

Directly let's you develop all kinds of machine learning models, DLSS is just one.


Where it gets interesting is Directly has been developed to be agnostic to architecture, so this way it fully supports Nvidia's tensor cores. This way Turing GPUs could be 8x faster than equivalent AMD hours for tasks like DLSS. Moreover, by using the Tensor cores the CUDA cores are not know Ted allowing them to soothe game razerization faster

 

[shankly1985]

DlSS is one particular implementation of an algorithm. It actually operates his g the DirectML stack, at least on Windows.

Directly is just a middleware API, just like DirectX. Directly doesn't provide any algorithms, it just defined the instructions.

Directly let's you develop all kinds of machine learning models, DLSS is just one.


Where it gets interesting is Directly has been developed to be agnostic to architecture, so this way it fully supports Nvidia's tensor cores. This way Turing GPUs could be 8x faster than equivalent AMD hours for tasks like DLSS. Moreover, by using the Tensor cores the CUDA cores are not know Ted allowing them to soothe game razerization faster

I like how you take that 8x from a benchmark ran on a nvidia GPU and automatically say its 8x faster than a Vega GPU.

It's a wild claim since we haven't seen a single test from DirectML ran on AMD.

For all we know it could be 5x faster, it could be much more, it could be even less. Amd might be working on something for Vega 7 when this releases spring windows 10 update.

 

[Poneros]

Unfortunately this sort of tech is just empty promises atm:

 

[doody]

great news for all gamers / freesync / now this well done amd