hey it's Bob Duffy with Intel here with another insightful episode of techcraft and today we're going to go through AI Denoising for 3D rendering applications so I'm excited we're going to do it on the... What no not the... A770, are you sure Okay well we're going to do AI Denoising for 3D rendering applications on this guy, the MSI Prestige 16 with an Intel Core Ultra processor and built-in Intel art GPU, so it should be fun and maybe we'll throw in the a770 if we got time. Yes we are going to run AI Denoising for 3D rendering on an Intel Core Ultra laptop in blender using Intel Open Image Denoise. Now before we get into the demo I've got to give you guys some background on Denoising if you feel you already know all this stuff feel free to go ahead and scrub forward till you see blender in the UI. But for for everybody else let's go to school on Denoising. Okay let's discuss Denoising what it is how it speeds up the time of ray tracing and how the Intel Open Image Denoise uses advanced methods to decrease rendering speeds with quality output. Most modern 3D render engines use a form of ray tracing called path tracing, to create beautifully realistic renders. Path tracing works by mimicking light, shooting virtual light rays from each pixel location in an image to objects in a virtual scene. If an object is translucent then the ray will pass and bend through the object to objects behind it. Information on light, shadows, color, and surface properties of each object along the path defines the color and the luminance of the pixel. Each light ray per pixel is called a sample. In this image we have 32 virtual arrays for every single pixel. Fewer samples or rays per pixel, will result in a grainy and noisy image but with faster rendering times. To get pristine renders it may take thousands of samples per pixel or even take hours per frame to generate. Denoising is a computer algorithm that cleans up the noise when using a lower sample per pixel count. This can create beautiful clean renders in a fraction of the time. Historically many denoisers did a fair job but in areas of low light or textures denoisers often created unwelcome patterns in a final render. The Intel Open Image Denoise Library uses artificial intelligence to more intelligently Denoise a render. And the results can be stunning. The Intel Open Image D noise library has been part of blender for many years now, but now supports GPU compute for rendering. allowing users with either Intel Arc discrete GPUs, as well as Intel Core Ultra processors with built-in Intel Arc GPUs to use open image Denoise. Learn more about Intel Open Image Denoise at openimagedenoise.org Okay here we are back in Blender and we got to do the basic stuff we always do in Blender so for GPU rendering first go to edit and you want to go to preferences um and we're going to be doing GPU uh rendering here with Open Image Denoise. So select preferences then go to system, and that's where you can select One API and you can select your Intel Arc GPU, and then you're going to want to select Embry uh for faster rendering as well which is also uh for for GPU. So now we're going to take full advantage of our Intel Arc GPU on our Core Ultra here. So this is a laptop that I'm going to use my virtual laptop to demonstrate this. Other couple of things you do is make sure over in render properties you are using the Cycles Render engine, otherwise we're not ray tracing. And then your device is set to GPU, even though that we did it before we got to do it here too. Okay now under the viewport side you've got a denoise option here um and you want to have that turned on and make sure that you've got Open Image D noise uh selected. And then the new thing is use GPU which now allows you to use the GPU for that denoise compute. And you can do the same thing under the render setting as well. So we have render settings and viewport settings for denoising. But I think what I'm going to do is, let me turn off the GPU and the denoising so I'm going to just turn everything off. We're going to see what the experience is without denoising. So going to move around the scene and you can see it's just noisy. It's just constant noise samples are going out, Up, sorry and it's still noisy. Here we can let those samples go up hit a high number and it's still it's still really really really noisy. And this is why we do denoising here, in order to clean that up and speed up our renders. So let's go ahead and turn it on but let's leave it to CPU and You can see right away it's working all that noise is cleaned up but at first it's it's you know a little funky looking and the sample count is really slow. So we're doing CPU Denoising for all of those samples and ray bounces that are happening. And that's at a cost um you know to do that um but um before I turn on a GPU just notice the sample count as it goes up. It's it's going up quite slowly here right right but let me go ahead and turn GPU on and you'll see. Wow look at that look at that go up. So you can see that we are Computing that denoising much faster we're getting more of those Rays more of those uh samples per pixel up faster so we're getting a crisper cleaner render um along the way and we're doing all of this on an Intel Core Ultra processor with built-in Intel Arc GPU. Pretty cool Okay so what I thought I'd do is show you path tracing um visualized. So we did Denoising here, but here's a game board scene I've shown before here, and let me go ahead and and turn on the render view in the viewport here. So you can see that this is path tracing. I like this because we have this glass game piece here. So the light's refracting through it. So those rays are actually doing those virtual bounces and you can see it is actually bending that light, but what I can do is go down to the index of refraction. I'm going to change this down to 1, which is going to make it just like clear and transparent like it's invisible. But if I change that number a little bit you can actually see the rays bending in near real time. So this is a really good example of actually seeing path tracing work on a device like this an Intel Core Ultra with built-in Intel Arc GPU. Okay before we get to the last bit and give the a770 a go. Let's give some context on scaling this denoising compute for Intel Arc GPUs. With Intel Core Ultra you experience the built-in Intel Arc GPU providing Graphics processing with 8 Xe cores for lower power mobile devices like the MSI Prestige 16 AI PC laptop and then here we're going to see the Intel Arc a770 GPU with 32 Xe cores which supports more demanding gaming and creator workloads on desktop systems. Okay as I said we're going to try this also on the in Arc a770 so same exact scene here, so um I've got this set where let me just make sure yeah denoising is off. GPU is off. So we're just doing exactly the same thing and I'm going to go ahead and turn on the render and you can see right away geez look at those samples going up I mean that time it went licky split. Right so obviously a more powerful GPU. So um this is this is scaling. So but what I want to do is it's the same exact thing is let me turn on denoise for CPU Denoising. You can see it slows down quite a bit so um again this is it just without Denoising and still we have a lot of noise uh and then I turn on denoising now the CPU is de noising as every sample it gets generated. Now I'm going to go ahead and turn on GPU Denoising we can see that is much faster. You know so it's it's definitely the same type of situation that if you don't have any denoising going on um you've got a lot of noise in the scene um but obviously because we're scaling up with you know uh more Xe cores this is happening faster, but as I turn on denoise CPU is in the in the process of getting in front of all of those samples and then if I turn on GPU that is much faster um and this this is a small scene there's only you know 48,000 triangles in this scene here but what I love about Blender is we can do two things at once so I've got Blender also open over here uh this is a very different scene we've got eight million triangles in this scene so this is the Barber Shop scene from uh Blender.org uh and I'm going to go ahead and turn, we'll do the same thing so we've got denoising off we're going to go ahead and we're going to do GPU rendering here and you see it it's going through it um it's going pretty fast but there's a lot more geometry for all of those virtual light rays to bounce around in order to calculate the luminosity and color of every single Pixel here. So it it's working in fact we can you know also just jump right in here um let me frame this you can see this is a glass object here so path tracing is working away uh but a very grainy image obviously so what we can do is turn on denoise now we're doing CPU denoising which is typically what we've been able to do um and the sample account is going up it's it's actually pretty good and we're getting a nice render and this is the great thing about viewport um rendering with Cycles is you get to see the actual lighting lighting Shadows all that kind of stuff all the shaders are being calculated in near real time but let's go ahead and turn on GPU and there we go much a much faster experience than what we had there before so this is the value of what's happening with uh software from Intel like Open Image Denoise. These things are continuing to improve, our Partnerships out in the ecosystem like Blender and getting Intel Open Image Denoise out there with advancements that support things like GPU denoising. So there you go a beautiful render here on an Intel Arc a770 as well as the Intel Core Ultra with built-in Intel Arc GPU. Hopefully you like this video. More things to come on this channel. So please like and subscribe. Until next time I'm Bob Duffy and we'll catch you [Music] later
