Nvidia RT Cores vs. AMD Ray Accelerators – Ampere vs RDNA 2 Ray Tracing

The first generation of the RTX Graphics cards was released in 2018. With this release, Nvidia presented a cutting-edge feature that was believed to transform the scene of gaming. The working of the first generation RTX 2000 series graphics cards use the latest Turing architecture. Also, it was the first to offer support for the real-time Ray Tracing in the games. In the professional 3D animation as well as synthetic fields, Ray Tracing already existed. However, Nvidia facilitated real-time manifestation of games with the implementation of Ray Tracing technology. Instead of using conventional rasterization, Nvidia considered the use of Ray Tracing technology.

Rasterization is the conventional approach using which games are rendered. On the other hand, Ray Tracing uses complex computations to precisely illustrate how light would interrelate and perform in the game environment if it would be in practical life. This post discusses more on the Rasterization and Ray Tracing.

Also Read: Best Ray Tracing GPUs – AMD & Nvidia

In 2018, AMD was kept in the dark about graphics cards from Nvidia’s RTX series and their Ray Tracing feature. The Red Team was unprepared for Nvidia’s ground-breaking introduction. Consequently, it placed its top offerings at a noteworthy disadvantage compared to Team Green. At the price of $399, the AMD RX 5700 XT was a wonderful graphics card. Its performance exceeds that of the RTX 2070 Super, whose price was around $499 retail.

The largest concern for AMD was that the competition offered a technology that they did not own. This, along with the various feature set, stable drivers, DLSS support, and superior performance, turned the offerings from Nvidia offerings into a noteworthy benefit about the Turing vs. RDNA generation of GPUs.

AMD RX 6000 series with Ray Tracing:

In 2020, AMD presented competition to the top offerings from Nvidia. In addition to presenting support for the Real-Time Ray Tracing in games, AMD has also released 3 graphics cards which are so much cutthroat to the topmost graphics cards from Nvidia. The AMD RX 6800, the RX 6800 XT, and the RX 6900 XT compete with the Nvidia RTX 3070, RTX 3080, and RTX 3090 correspondingly. Ultimately, AMD again proves to be a competitive gain on the highest end of the product stack that assures recognition for its consumers.

Related Reading: Best RDNA2 RX 6000 Video Card Series

But, things are not completely constructive for AMD either. Even though AMD has presented support for Real-Time Ray Tracing in the games, it is observed that their Ray Tracing performance got an unenthusiastic response from general consumers as well as reviewers. It is explicable because this is AMD’s foremost attempt at Ray Tracing. Therefore, it would be slightly unjust to anticipate the Ray Tracing performance right from the initial attempt. But it arouses questions regarding how the AMD’s Ray Tracing implementation functions compared to the Nvidia’s implementation, which we observed with the Turing architecture and presently the Ampere architecture.

Nvidia’s suite of RTX Technologies:

Image: Nvidia

The key reason why the attempt by AMD appears to be dull compared to that of Nvidia is that AMD was trying to compete with Nvidia. It only had around 2 years for developing and perfecting their execution of Ray Tracing. On the other hand, Nvidia is involved in the development of this technology for a long time. Also, there are no competitors for this at the extreme top of the product stack. In addition to delivering Ray Tracing support before AMD, Nvidia also owns a better support system for the technology.

During the design of the RTX 2000 series graphics cards, the key focus of Nvidia is Ray Tracing. The same is apparent all through the design of the Turing architecture. In addition to multiplying the number of CUDA Cores, they too added explicit Ray Tracing cores recognized as “RT Cores.” Such cores deal with a huge number of computations needed for Ray Tracing. Moreover, Nvidia developed “Deep Learning Super Sampling (DLSS)” technology that implements AI and deep learning to accomplish tasks like upscaling and rebuilding. This technology also recompenses for the performance failure of Ray Tracing.

Nvidia also presented explicit “Tensor Cores” in GeForce series cards designed to assist in AI and deep learning tasks like DLSS. Nvidia paired with the In-game studios to optimize the forthcoming Ray Tracing games for the allocated Nvidia hardware to make sure the performance can be amplified.

Nvidia’s RT Cores:

Source: Nvidia

Before diving deep into this section, it is essential to understand what it means by Nvidia’s RT cores. The latest RT Core resides at the heart of Turing’s ray tracing acceleration that is based on hardware. This core is found in each SM. Generally, RT Cores carry out visibility testing in favor of threads operating in the SM.

Ray Tracing Cores or RT are customized hardware cores from Nvidia. They are exceptionally designed to deal with the computational tasks lined with real-time Ray Tracing in games. Allocating special cores for Ray Tracing unlades a huge workload from the CUDA Cores, which are allocated for standard rendering in games. Hence, the overload of core utilization does not much influence the performance. RT Cores compromise versatility and execute hardware with a special architecture for exclusive algorithms or calculations to attain faster speeds.

The typical Ray Tracing acceleration algorithms that are famous are Ray Packet Tracing and BVH. Moreover, the more common Ray Tracing acceleration algorithms commonly known are BVH and Ray Packet Tracing and the illustrative diagram of Turing architecture introduces BVH (Bounding Volume Hierarchy) Transversal. Those commands that relate to Ray Traced rendering in games are identified and accelerated by the RT Core.

In the Turing Architecture White Paper, Nvidia mentions that the RT Cores work collectively with the latest denoising filtering. This filtering feature is an extremely efficient BVH acceleration configuration developed by NVIDIA Research. It also works as RTX compatible APIs to accomplish ray tracing in real-time on a solely Turing GPU.

RT Cores separately go over the BVH. With the acceleration of traversal and ray/triangle intersection trials, they unlade the SM. As a result, it allows the SM to deal with another pixel, vertex, and compute shading work. The driver manages functions like BVH building and refitting. The application handles other functions like shading and ray generation via novel types of shaders. As a result, it releases the SM units to perform other computational and graphical tasks.

AMD’s Ray Accelerators:

Image: AMD, RDNA2 Architecture

For improving the Ray Tracing pursuance of AMD’s RDNA 2 GPUs, AMD has embedded a Ray Accelerator component within its core Compute Unit Design. It is expected that such Ray Accelerators boost the efficiency of the typical Compute Units within the computational workloads associated with Ray Tracing.

With the introduction of the RX 6000 series, AMD entered the Ray Tracing competition. Along with that, AMD also implemented several major components to the architectural design of RDNA 2 through this feature. The working mechanism of the Ray Accelerators is yet unclear. But AMD has drawn some recommendations to how such elements are expected to function. As per AMD, such Ray Accelerators have a disclosed intention of passing through the Bounded Volume Hierarchy (BVH) structure. Moreover, it intends to determine intersections between boxes and rays efficiently.

Source: AMD

The design is entirely compatible with DirectX Ray Tracing, which is recognized as the industry standard for PC Gaming. Besides, AMD uses a Computing-based denoiser to regulate the specular effects of ray-traced scenes in lieu of depending on purpose-built hardware. Perhaps, this will pose additional pressure on the mixed-precision abilities of the latest Compute Units.

Ray Accelerators are competent in executing 4 bounded volume box intersections or a triangle intersection in each clock. This processing speed is faster than per clock, which is much faster than executing a Ray Traced scene in the absence of explicit hardware. The approach by AMD presents a huge benefit. It implies that the RT Accelerators from RDNA 2 can interrelate with the Infinity Cache of the card. It can simultaneously store a huge volume of Bounded Volume Structures in the cache. Therefore, some load can be reduced from memory read cells and data management.

Key Difference:

A massive difference is instantly perceptible when carrying out a comparison between the Ray Accelerators and the RT cores. This difference is when both carry out their functions quite identically, the RT Cores are distinct hardware cores that possess a singular function. On the other hand, the Ray Accelerators belong to the standard Compute Unit configuration within the RDNA 2 architecture.

The Nvidia’s RT Cores exist on their second generation with Ampere and ample architectural and technical advancement. As a result, Nvidia’s RT Core implementation proves to be a quite efficient Ray Tracing mechanism compared to AMD’s implementation using the Ray Accelerators.

There is one Ray Accelerator embedded in each Compute Unit. The number of Ray Accelerators is 80 in AMD RX 6900 XT, 70 in 6800 XT, and 60 in RX 6800. Keep in mind that these numbers cannot be directly compared to the Core numbers of Nvidia’s RT. This is because those are the dedicated cores put together considering a single function.

Source: AMD

The number of 2nd Gen RT cores is 82 in RTX 3090, 60 in RTX 3080, and 46 in RTX 3070. In all these cards, Nvidia has equipped distinct Tensor Cores. Ultimately, it assists in AI and machine learning applications such as DLSS.

Future Optimization:

At this moment, it is difficult to predict the future of Ray Tracing for AMD and Nvidia. However, you can carry out some knowledgeable guesses by inspecting the present scenario. At the time of writing, Nvidia excels in Ray Tracing performance compared to the offerings from AMD.

AMD has noted an outstanding commencement for RT Cores, but they are still lagging Nvidia by 2 years in development, research, optimization, and support. In 2020, Nvidia bounded most of the Ray Tracing titles for using Nvidia’s dedicated hardware more effectively than that by AMD. Another fact is that Nvidia’s RT Cores are more efficient than the Ray Accelerators by AMD. Both these facts prove that AMD is not much suitable in terms of the existing Ray Tracing condition.

This does not imply that AMD has stopped there. It is already proclaimed that AMD is working on an AMD alternative to DLSS. This is a huge assistance in enhancing Ray Tracing’s performance. Furthermore, AMD is also dealing with game studios to enhance forthcoming games for their hardware. This improvement is visible in titles such as GodFall and Dirt 5. In these game titles, AMD’s RX 6000 series cards work quite powerfully. Henceforth, we can anticipate that Ray Tracing support by AMD would be continuously improved with the forthcoming titles and with the development of imminent technologies like the DLSS Alternative.

Anybody willing to obtain significant Ray Tracing performance needs to value the efficiency of Nvidia’s RTX Suite. Our definitive recommendation is the latest RTX 3000 graphics card series from Nvidia rather than AMD’s RX 6000 series. This recommendation is useful for those who consider Ray Tracing a vital aspect in the buying decision. With the forthcoming AMD’s offerings, this may and must change. Also, with time, we can expect advancements in in-game optimization and drivers.

Final Words:

AMD works on offering efficient Ray Tracing performance with their RX 6000 series of graphics cards. These graphics cards use the RDNA 2 architecture. They do not surpass the performance by the Nvidia’s RTX 3000 series in direct Ray Tracing criterion. But they surely offer cutthroat rasterization performance and notable value. The same will entice the gamers who do not value Ray Tracing. But AMD is working on enhancing Ray Tracing performance using some significant steps in a series.

The approach which AMD and Nvidia adopt for Ray Tracing is quite identical. However, both companies implement unique hardware methods. Preliminary testing has demonstrated that dedicated RT Cores by Nvidia surpass AMD’s Ray Accelerators, which are put together into the Compute Units on their own. To the end-user, this may not be a huge concern. However, it is a significant aspect to cosine for the future because game developers now need to improve their RT features using one of these approaches.