Sony decided to develop its own AI hardware for the PS5 Pro instead of using AMD technology exactly four years ago, and the big question is why

Upscaling graphics using AI is so important for games. Sony has decided to develop its own hardware instead of using AMD technology. That’s one of the key takeaways from the signature deep dive that Sony’s lead architect for its Playstation consoles, Mark Cerny, just gave.

AMD is of course responsible for both the CPU cores and most of the GPU technology in PS5 Pro. But when it comes to AI acceleration for upscaling, Sony has decided to go its own way.

For clarity, what we mean here when we talk about “AI acceleration” is the matrix math hardware used to power neural networks, also known as machine learning or ML. Currently, both Nvidia and Intel have dedicated matrix math acceleration hardware in their graphics chips for PCs. But not AMD. In fact, the existing AMD-based PS5 Non-Pro does not have any dedicated Matrix Math hardware.

In Nvidia and Intel GPUs, this matrix math hardware is primarily used for various upscaling technologies, including DLSS and XeSS. The lack of matrix math hardware in AMD GPUs is the reason why AMD’s FSR upscaling is based on hand-coded algorithms and not ML. And that, in turn, is a big reason why FSR is in many ways not as good as XeSS and is certainly behind DLSS.

According to Cerny, Sony had to make two important decisions for PS5 Pro and ML acceleration. First, it had to be decided whether to use a dedicated NPU or Neural Processing Unit or to use an improved GPU. Sony has chosen the latter.

Then it had to decide whether it wanted to license that technology or develop its own. Cerny says the PS5 Pro project began back in 2020 and it was decided from the start that Sony didn’t want generic ML hardware, but rather something specific to the PS5 Pro’s workloads.

PS5 Pro ML technology

Sony has decided to go its own way with machine learning hardware. (Image credit: Sony)

“Once you license technology, that’s what you do forever. That’s why in 2020, despite the effort required, we decided to develop our own hardware and software technology,” says Cerny.

The goal was ML performance of 300 TOPS. That’s 300 trillion operations per second, or roughly on par with the pure ML performance of an Nvidia RTX 4060 Ti.

To achieve this, Sony modified AMD’s RDNA 2 GPU architecture. “We have made a number of targeted improvements to the RDNA shader core and surrounding storage systems. We call it custom RDNA,” says Cerny.

The problem for Sony was that adding matrix math capabilities to the RDNA 2 shader core, as used in the plain old PS5, wouldn’t be enough. The matrix math consumes so much bandwidth that without improvements to the storage system, the new ML hardware would, in the worst case scenario, only reach 3% of its full potential. Yikes.

Therefore, Sony has developed a tailored on-chip memory solution and tiling algorithms that allow matrix computation to be performed entirely on the GPU without the need to access system memory.

PS5 Pro ML technology

(Image credit: Sony)

According to Cerny, it was a “four-year journey” to design all of this. The result uses the vector registers in the RDNA 2 shaders as RAM. This provides 15MB of storage with a combined bandwidth of 200TB/s. In comparison, the bandwidth of the PS5 Pro’s GDDR6 memory is “only” 576 GB/s. Meanwhile, the tiling algorithms allow the matrix computation to be divided into parts small enough to fit into those 15 MB.

Anyway, it’s this new ML hardware that powers Sony’s new upscaling software, Playstation Spectral Super Resolution, or PSSR, which is essentially Sony’s answer to Nvidia DLSS and Intel XeSS. Oh, and AMD FSR, although as explained above, FSR is not based on machine learning.

“As I look back over the four years since this project began, I am so glad we made the time-consuming decision to develop our own technology. The results are good and, just as important, we’ve learned so much about what AI can do.” “Improving game graphics can only make our future brighter,” says Cerny.

So the big question that arises is why Sony felt the need to develop its own ML hardware. According to Cerny, Sony didn’t want to be tied to a permanent licensing agreement, which means Sony sees ML upscaling as even more important than GPU architecture. Finally, Sony is willing to license AMD’s more comprehensive GPU architectures.

PS5 Pro ML technology

The PS5 Pro’s ML hardware has about the same matrix processing power as an Nvidia RTX 4060 Ti. (Image credit: Sony)

On the other hand, did AMD even have ML hardware to offer in 2020? It’s worth noting that AMD’s current RDNA 3-based GPUs still lack ML hardware. And some rumors suggest that the upcoming RDNA 4 GPUs may not have ML hardware either. Perhaps the only way for Sony to integrate ML hardware into the PS5 Pro was to do its own thing.

If true, that would also mean we probably won’t see an AMD graphics chip with ML hardware until around 2027, when the upcoming RDAN4 chips are replaced by something based on AMD’s new unified CDNA architecture.

Considering that even Intel’s first generation Arc GPUs had ML hardware and Nvidia is about to launch its fourth generation of GPUs with ML hardware, this is truly a damning verdict against GPU technology AMD and is not letting the company go further in the race for AI upscaling.

We hope the rumors are false and AMD will release some GPUs with ML hardware soon.

Breaking News

Dellupodisabato

Sony decided to develop its own AI hardware for the PS5 Pro instead of using AMD technology exactly four years ago, and the big question is why

Leave a Reply Cancel reply