While we are still waiting for an official Arc launch, Intel in an ‘Ask us Anything‘ showcased the Arc architecture along with the A770 and the A750. We got some pretty juicy details to share. You can watch the complete video here.

Starting off, we present to you the complete Arc lineup. While we have seen this picture before, Intel reassures us that Arc is en route and not cancelled. The initial launch lineup will only feature 4 GPUs namely:

Team blue stated a while back that the 8GB variant for the Arc A770 will be exclusive to the AIB partners. Only the A770 16GB will be treated as the ‘Limited Edition‘.

Intel then showcases their RT performance which is on-par with NVIDIA. We unearthed the XeSS technology yesterday. In short, the image quality is softer as compared to DLSS. In extreme testing conditions (bushes, trees) pushing most AA (Anti-Aliasing) technologies to their limits, we see XeSS fall prey to soft edges and miscalculated shadows.

The RT performance has been shown before by team blue. We’re seeing them move closer and closer to NVIDIA’s territory, however, the actual hands-on experience will tell the full story. 

Coming to the main topic of today’s discussion, we find the Arc A770 and the Arc A750. This is the most detailed specs sheet we have seen so far. Both the A770 (16GB) and the A750 have the same TDP (225W). Intel stated that the Arc A770 can overclock to 2.7GHz with air cooling. Additionally, we see support for PCIe Gen 4 x 16. These GPUs ship with a standard 3-year warranty and support the Windows 10/11,Ubuntu OS.

Interestingly, the Arc series ‘officially’ supports only DX12 Ultimate. This is where Intel’s architecture starts to fall apart because the difference between DX11 and DX12 in terms of performance is night and day. All in all, the specs seem solid for a 1st Gen GPU. This might have actually threatened both NVIDIA and AMD, if these GPUs were launched on time. However, Intel did do its homework and we have to give them that.

This is the ‘nerdy’ part for most. However, lets first understand a few terminologies.

Basic rasterization works like this. You have a few 3D objects inside a scene which by using some algorithms are squished to form a 2D image (with a 3D looking effect) on your screen. Your screen essentially goes in the x-y coordinate plane with no concept of a 3rd coordinate (z).

After this, each pixel is shaded with colors using another algorithm baked into the program. This is how you see images on your screen.

Ray Tracing requires accurate reflection of light after hitting a surface, but how do you know which surface reflects how much light? To combat this, a ray is casted (Ray-casting) from you camera (screen) as an imaginary line which strikes the triangles (building blocks of 3D Models) present in a scene. But there are millions of such triangles. Similarly, you will need quite a lot of ray casts for all the lighting effects. Here BVH comes into play. 

BVH stands for Bounding Volume Hierarchy. Each object is surrounded by an imaginary box consisting of smaller boxes. The working is as follows, the ray-cast hits the large box (A), which then opens up the smaller boxes (B), which after striking the perfect box from B, go into further smaller boxes (C) for accurate ray tracing. 

BVH is a data structure stored in the GPU’s cache memory for easy access when needed later. This is different from NVIDIA’s approach and can open up different opportunities for game developers. It will be interesting to see this technology in a few years.

To sum it up, Intel really outdid itself because such persistence even after so many set backs puts to show that Intel is here to stay. Including RT in the 1st iteration was never necessary, but they did so and with surprisingly good results. Similarly, XeSS also looks great albeit with a few caveats. These GPUs are expected to launch ‘anytime’ now given how Intel is fueling the Arc hype train.

(Featured Image credit goes to hothardware)