PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true.

This is part of the current PlayStation 5 digital foundry coverage. You can find the rest here:

  • These are the full specifications of the PS5: the technology behind the PlayStation 5
  • PlayStation 5: what the boost function means for the PS5
  • PlayStation 5: expandable memory and unprecedented 3D audio fidelity on PS5

The graphic heart of the PS5

At first glance, the PlayStation 5 offers a lot of performance, but it seems that developers need to optimize these new features. The question is, what happens when the processor hits its performance limit and the clock speeds of the components are reduced? In his presentation, Mark Cerny freely admits that the CPU and GPU don’t always run at 3.5 GHz and 2.23 GHz, respectively.

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

“When this worst-case game happens, it runs at a lower clock speed. But not too low. To reduce performance by 10%, only a frequency reduction of a few percent is needed, so I do a rather barely noticeable downclocking. would wait, ”he explains. “Overall, switching to a variable frequency approach will bring significant benefits to PlayStation gamers. «

At the feature level, Cerny reveals features that indicate parity with other upcoming AMD and AMD derivative products based on RDNA-2 technology. A new block, known as the Geometry Engine, gives developers unprecedented control over triangles and other building blocks, as well as easy optimization for geometry removal. The functionality extends to creating “primitive shaders,” which sound very similar to Nvidia Turing’s mesh shaders and upcoming RDNA-2 GPUs.

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

While Cerny doesn’t mention technologies like machine learning support or variable rate shading, the PS5 actually provides accelerated ray tracing through its intersection engine, which Cerny says is « based on the same strategy as the next AMD PC GPUs ». There was speculation about an external block, but that’s not the case – as with next-gen navigation and the Xbox Series X, the RT hardware is built into the shaders and therefore fully integrated. Similar to the RDNA implementation, the PS5 can access the same RT implementations we’ve seen in the PC realm – reflections, ambient occlusion, shadows, and overall lighting go well together.

“How far can we go? I’m starting to become quite optimistic, ”explains Cerny. “I’ve seen a PS5 title before that successfully uses ray tracing-based reflections in complex low-cost animated scenes. «

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

How SSD Helps Realize the Next Generation’s Dream

The type of SoC and its functions are important – and much of it is new and important information based on feedback from our readers. However, the presentation shows that Mark Cerny has very different priorities – and there is the dream of the next generation that needs to be considered. Two very specific components play a role here: the SSD and a remarkable 3D audio hardware known as the Tempest engine.

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

Sony places great importance on solid-state storage to deliver a truly transformative next-generation experience. Every few years, Mark Cerny travels the world, meeting dozens of developers and publishers, and integrating an SSD was the next generation’s biggest wish. Sony’s actual implementation is different: performance is twice as fast as on the PlayStation 4. 2 GB of data can be loaded in a quarter of a second, which means that theoretically the 16 GB of the PS5 can be filled in only two seconds. « As game developers we no longer try to distract the player from the time the journey takes – like on those Spider-Man subway rides – we are so blindingly fast that we may even have to slow that transition down. » . Said Cerny.

To make such a leap is performance, a lot of special hardware was needed to connect the SSD to the processor virtually seamlessly. A dedicated flash connects the SSD modules to a 12-channel interface, which provides the necessary 5.5 Gb / s, with a maximum storage capacity of 825 Gb. This might seem an odd size for the SSD, considering the sizes of the SSD. Consumer SSD of 512GB, 1TB or more, but Sony’s proprietary 825GB is the optimal size for the 12-channel interface and has other benefits as well. In short, Sony guarantees the freedom to adapt the design: “We can look at the available NAND flash components and build something with the best value for the money. Anyone who uses an M.2 drive probably doesn’t have that freedom and it would be difficult to market if they didn’t have one of these standard sizes, ”says Mark Cerny.

The controller itself is connected to the main processor via a four-lane PCI-Express 4.0 and brings with it a number of hardware units mentioned to eliminate SSD bottlenecks. The system offers six levels of prioritization so developers can prioritize the different data streams that the game needs.

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

The controller supports hardware decompression for the ZLIB standard, but also supports RAD Game Tools’ new octopus format, which offers 10% more compression efficiency. What does all this mean? 5.5 GB / s of bandwidth means that eight or nine gigabytes per second can be injected into the system. « Only marginally, converted to pure performance, does this mean that your own decompression hits nine to ten times Zen-2 cores, which is what it would take to decompress the Kraken stream on a normal processor, » explains Cerny.

A dedicated DMA controller (equivalent to one or two Zen-2 cores in performance) gets data to where it’s needed, while two dedicated custom processors handle I / O and memory mapping. There are also separate consistency engines that allow for a bit of tidying up if needed.

“Consistency shows up in a lot of places, the biggest problem being the expired data in the GPU cache,” explains Cerny. “Emptying the entire GPU cache every time the SSD reads is not an attractive option, as it could drastically reduce GPU performance. So we’ve found a better way to do it when consistency engines overwrite the GPU. The filled in address zones and special purifiers in several dozen GPU caches then have the task of freeing exactly these address zones.

PlayStation 5: the graphics core of the PS5 and how SSD is helping the next generation’s dream come true

Developers get all of this without doing anything. The special hardware even supports decompression. “You just say what data should be read from your own uncompressed file and where that data should go. The actual loading process then takes place invisibly and very quickly, ”explains Cerny.

Source : Twitter Feeds