Unveiling the PS5: Cutting-Edge GPU Innovations and the Future of Gaming Technology Through Sony and AMD Amethyst Collaboration

When Sony introduced the PlayStation 5 Pro, it did so in a distinctly different manner compared to the original PlayStation 4 Pro and PlayStation 5. Whereas lead system architect Mark Cerny served as the ‘master of ceremonies’ for the Pro reveal, the debut was a concise 9-minute event – and a significant portion of that runtime focused on the successes of the original PlayStation 5 console. It was a stark contrast to the extensive presentation given for the launch of Sony’s first ‘Pro’ console and was notably lacking in details compared to the now-iconic ‘Road to PS5’ presentation delivered by Cerny in March 2020. We were left wanting for more crucial information – and now, that information has finally arrived.

Just three months after the unveiling, Sony has provided a new in-depth presentation regarding the hardware design and indeed the vision behind the PlayStation 5 Pro – and it’s impressive material. The philosophy guiding the console represents a transformation in how consoles are likely to be developed. While the core graphics rendering technology of the PlayStation 5 Pro has been improved from the original model, Cerny envisions a future where machine learning plays a pivotal role in console development, and while rasterisation approaches its limits, he anticipates significant advancements in ray tracing capabilities moving forward.

This vision bears similarities to Nvidia’s trajectory over the past six years, and our observations indicate that AMD will continue to be Sony’s partner of choice in realizing this new vision. In pursuit of that goal, a new ‘Amethyst’ collaboration between Sony and AMD has been announced.

Digital Foundry’s Oliver Mackenzie had the opportunity to observe the recent Cerny presentation several weeks ago and to interview Mark Cerny himself, alongside Mike Fitzgerald – Insomniac’s director of core technology. However, the presentation itself is something that all Digital Foundry team members were eager to discuss, so here is our response to the conversation – and, of course, an overview of everything we have learned.

GPU overview: RDNA 2.x, 16.7TF

The most noticeable advancement from the original PS5 to the PS5 Pro is evident in terms of its graphics processing capabilities. Rather than enhancing CPU performance, improving storage speeds, or focusing on higher resolution displays, as one might expect from a new console, the primary emphasis with PS5 Pro is bolstering performance and visual fidelity through a more powerful GPU. Interestingly, the PS5 Pro GPU is still fundamentally based on AMD’s RDNA 2 (RX 6000-series) architecture, but includes some features backported from RDNA 3 (RX 7000-series) – and even some from RDNA 4 (RX 8000-series, expected in 2025) as well.

With a current clock speed of 2.17GHz, the PS5 Pro achieves a performance level of 16.7TF – half of the previously stated 33.5TF which was incorrectly reported prior to launch. This is much closer to the approx. 10TF number associated with the original PS5, compared to the inflated (or “flop-flated”, as Mark Cerny describes it) RDNA 3 teraflop figures. Other specifications align with early PS5 Pro leaks, including 30 WGPs, 60 computing units, and a maximum boost of 2.35GHz – although this frequency is unlikely to be attained under normal operating conditions, likely due to power constraints. The original PS5 has a maximum clock speed of 2.23GHz, but the presentation suggests that the operational speed is the same 2.17GHz as the PS5 Pro, so the current TFLOPs is about 10TF, rarely exceeding the 10.23TF indicated in the latest specifications. As this generation has demonstrated, however, TFLOPs is increasingly proving to be an inconsequential metric – a point acknowledged by Mark Cerny in the recent presentation.

As for why PlayStation 5 Pro is not entirely adopting the latest AMD graphics architectures, the explanation is quite straightforward: the shader code for the PS5 GPU would not function on newer AMD hardware, and there is no way those shaders could be recompiled for existing games. Shader compilation on-the-fly? It’s not feasible for PS5 and – trust us – you wouldn’t want it either. Thus, the PS5 Pro must utilize the same RDNA 2.x foundation, with extensions added for new features – such as enhanced ray tracing.

Regarding additional features, we also know from separate disclosures to developers that the PS5 Pro boasts the full RDNA 2 feature set, including mesh shaders and hardware VRS, along with an additional 2GB of (slower) DDR5 memory onboard. Of that, over one gigabyte of extra memory is available to developers. The GDDR6 memory used here is also approximately 28 percent faster than on the original PS5 – 576GB/s versus 448GB/s – despite being connected on a similar 256-bit bus.

Ray tracing enhancements

A significant aspect of the PS5 Pro’s allure is its remarkable improvement in RT performance, enabling developers to utilize these capabilities more freely in PS5 Pro-enhanced titles without compromising visual quality or frame rates to the same extent as on the original PS5. This is achieved through the RDNA 2.x architecture of the original PS5, with new extensions derived from later RDNA generations.

The PS5 repurposes the texture mapping units to perform field and triangle intersection tests for ray tracing, with the PS5 Pro accelerating this segment of the process by two to three times. The Pro also supports BVH8 instead of BVH4, which further enhances performance. Additionally, the PS5 Pro incorporates stack management in hardware, which again aids the traversal stage in ray tracing and was previously only seen in Intel and Nvidia hardware, not AMD.

Finally, more divergent RT sees a greater performance boost than more coherent RT on PS5 Pro. This divergent/coherent spectrum fundamentally describes
“`the intricacy of RT computations, with shadows and reflections on even surfaces exhibiting greater coherence, whereas reflections on irregular or uneven surfaces tend to be less consistent. Nvidia and Intel again excel by implementing suitable methods to handle these less consistent RT scenarios, including hardware sorting mechanisms and shader execution reorganization. Although we do not have direct examples of these solutions here, the advancements in traversal and the transition to BVH8 suggest that the PS5 Pro is far more capable than its predecessor in managing these resource-intensive RT computations. This development enables developers to more easily utilize a much broader spectrum of material roughness for reflections, rather than being limited to just mirror-like or map-like reflections.

Mark Cerny theorizes that there is a threshold to how far we can advance rasterization. One way to push beyond these limits is through RT, which offers significant scalability potential. Observing these improvements from Sony indicates that RDNA 4 is making strides in this direction, which is quite exciting.

However, we still do not know for certain how RDNA 4’s RT performance will manifest. It appears to represent a considerable advancement, yet it does not yet incorporate the RT core seen in Nvidia’s RTX 20-series architecture and later models, which allows for handling multiple workloads at once. AMD has illustrated in presentations how it is utilizing larger amounts of cache on its PC GPUs to incorporate more into the BVH, so it will be intriguing to see how the desktop RDNA 4 chips perform with this enhancement.

Machine Learning and PSSR

In addition to the PS5 Pro’s foundational RDNA 2.x being enhanced with what Sony refers to as “future RDNA” for ray tracing, it also incorporates “custom RDNA” for machine learning, specifically PSSR upscaling. Instead of relying on a pre-existing upscaler, Mark Cerny explained that Sony focused on speed and reducing the time per frame that their upscaler…