Despite the fact that PS5 is having an unbeatable success and is undoubtedly the best-selling next-generation console of the moment and the most popular, we have decided to use our knowledge of hardware to build what will be PS6 from a realistic perspective . That is why we have decided to put all our ideas on the table and in a totally ordered way. Do you want to see what we expect from the PlayStation of the future?
This article is largely an exercise in informed speculation about what we might expect from a future PlayStation 6 or PS6 from SONY. A console that we know is under development through a job offer that was published a few months ago by AMD where engineers were sought to create the chip for a next-generation video game console. On the other hand, this will serve as an exercise to show you some future technologies, which at the moment are of minority use or simply found on a road map.
A multi-chip setup for PS6
We must start from the fact that the era in which the cost per square millimeter was maintained has already passed into history from several manufacturing nodes. Which has caused that even with increasingly powerful systems since the launch of PS4 and with a greater number of transistors, the area that the chip of each of the new models has has been decreasing with each new iteration. The objective? keep the cost per processor within margins. Thus, the original PS4 chip measured 348mm 2 , while the newer PS5 model, the 1200, measures in total close to 280mm 2 .
AMD’s solution in its desktop PC systems has been to completely break the chip’s functions into several distinct ones. This allows them to make not only smaller and cheaper chips, but to keep the parts of them that don’t require such an advanced node to function separately. We’ve seen it on desktop Ryzens where we have the processor cores and their cache separated from the memory controller. Also recently in GPUs where the last level cache and memory controller is divided into several small chips.
PS6 will be a multi-GPU console
Well, SONY in one of its patents, in which it talks about a Pn system that is much more complex than the current PS5, suggests a multichip configuration and presents us with several possible configurations. This is titled SCALABLE GAME CONSOLE CPU/GPU DESIGN FOR HOME CONSOLE AND CLOUD GAMING and in it we can read the following statement:
In a multi-GPU simulation environment, image buffering can be implemented with multiple GPUs, each rendering their respective video frames or their respective portions of each frame.
That is, the same system that we saw in its day with SLI by NVIDIA and Crossfire by AMD. Which means that PS6 games should be programmed to take advantage of this functionality. On the other hand, it would reduce the manufacturing costs of the console by having much smaller chips, which considerably increases the number of good chips per wafer. What’s more, one of the things that we have observed in the new AMD RX 7000 is that there is a division between the Frontend of the chip and the rest of the GCD in terms of clock speed, which gives clues that it will be one of the parts that we can disintegrate.
What would the final design look like?
In the end, in the design we would have a multichip system, which would have the following chiplets:
- CPU
- IO Die (Integrated memory controller, audio and video codecs, display controller, peripheral management (SSD, USB ports, Bluetooth) and FrontEnd and GPU L3 cache)
- 2 x Shader Chiplet Die (GPU Compute Units, L1 and L2 Cache)
So in total we would have 4 different chips, some of them using less advanced manufacturing nodes such as the Memory Cache Die. The objective of separating the FrontEnd from the GPU is due to the fact that this allows to have a configuration of several GPUs, but that it is seen as one by the applications.
PS5 Slim will come before PS6
When SONY launched PS4 Pro, it did so under a different directive from the current one and under the objective of using its console to motivate the sale of 4K televisions, which were sold under its Bravía brand. It was a hardware that was a small evolution on the 2013 design, sharing a good part of the architecture, but making use of a more powerful graphics. Its business impact? Well, not too good, since less than 20% of the PS4s sold were the Pro model and today it is a model that SONY no longer manufactures.
However, the rumors about a possible PS5 Pro do not stop from the moment we know that both AMD and SONY are already working on the main chip of a future PlayStation . The conclusion of the majority of media? Comment on the existence of a potential PS5 Pro for 2024. Our bet? It is a PS5 Slim that we will see in 2023 and that has some improvements compared to the current PS5 with the aim of being able to reach 4K resolution at the highest possible fidelity that the console can and at 60 FPS. And believe us that these improvements are slight compared to having to make a Pro Model.
What is our argument based on? The current PlayStation leadership hates having several models with different hardware on the market. We are not going to see a PS5 Slim sharing space with a PS5 Pro, but a single model, which, as you know, we will see in 2023. Just three years after the launch of the original model. A timing that gives us the idea of when we should expect the launch of PS6
How can PS5 Slim reach 4K and 60 FPS?
One of the peculiarities between PS5 games is that users usually have to choose between two different graphics modes, usually called performance and quality. For example, in God of War Ragnarok we have a 4K quality mode, but at 30 FPS, while the performance mode fluctuates between 1440p and 2160p. Another example is Horizon Forbidden West, whose performance mode works at 1800p, but if we want to enjoy the game at 4K we have to do it at 30 FPS. That is to say, the console in its most cutting-edge games has problems maintaining even an Ultra HD resolution at a stable 60 FPS and showing all the visual capacity.
Now, well, times are different from PS4 Pro and we are in the age of auto-upsampling algorithms. In the case of AMD we have FSR 2.0 and in a few months we will probably have version 3.0. The operation of these algorithms is simple, since they are based on the GPU generating the frame with a lower resolution and then generating the same image at a higher resolution and with a larger number of pixels.
In other words, we hope that a possible PS5 Slim will allow us to play console games in graphical fidelity mode with the same speed as performance mode. Even if it’s using tricks like FSR 2.0. Which will force the PS6 to aim at a much higher horizon.
8K, the standard resolution of PlayStation 6
With PS5 Slim achieving 4K performance excellence, SONY will need to aim for the next big resolution build for its PS6. Currently and in the middle of 2022, 8K is something that is only available to a privileged few. And to be able to do it with ease and without any type of performance problem can require a system as powerful as a very high-end PC such as an RTX 4090 in order to play at said resolution and at 60 frames per second.
The problem is that we are talking about 8K being a resolution with 4 times the number of pixels than 4K, this means that the raw computing power will not only need to be quadrupled, but also the bandwidth with which it is used. communicate the graph. So we must not only use a higher power in the graphics card, but also get higher bandwidths to be able to operate properly.
What limitations can be found in the face of 8K?
Currently, and oddly enough, we can build a video game system capable of generating 8K graphics, but with incredible costs and a bottleneck. For example, the NVIDIA RX 4090 has memory at 1 TB/s of bandwidth and is the only one that can hold up. But for this it requires a 384-bit bus and 21 Gbps GDDR6X memory. In the case of the RX 7900 XTX we have a similar performance with a GDDR6 at 20 Gbps.
| Characteristic | ps4 | PS5 | Differential | RX 7900 XTX | Differential |
|---|---|---|---|---|---|
| GPU TFLOPS | 1.84 TFLOPS | 10.36 TFLOPS | x5.63 | 61 TFLOPS | x5.88 |
| Fill rate | 25.6 Gpixels/s | 144 Gpixels/s | x5.63 | 441.6 Gpixels | x3 |
| Texturing rate | 57.6 Gtexels | 324 Gtexels | x5.63 | 883.2 Gtexels/s | x 2.75 |
Interestingly, if we take into account only the computing power of the RX 7900 XTX we will see that this is the same as that between PS5 and PS4, so it could be considered an intergenerational leap. However, if we talk about fill rates and texturing the impact is not that big. The reason? Bandwidths have not grown so much and believe us that in the case of the NVIDIA RTX 4090, it is exaggeratedly limited in its memory speed.
The problem is that we need fewer memory chips in order to reduce system costs, this is achieved with a much faster standard than the current one. Which we already know exists today and is GDDR7 memory . With which it has reached 36 Gbps speed and therefore 1152 GB/s in bandwidth . At the same time, new manufacturing nodes for RAM allow for much higher densities. Therefore, we should think about configurations of 4 GB per memory chip , that is, if the current 256-bit bus is maintained, we would speak of 32 GB of capacity for the system.
Balance between computing power and bandwidth
One of the concepts when designing a GPU is arithmetic intensity, which is achieved by dividing the total bandwidth with the video memory of a graphics card by its total computing power. Said calculation will help us to have a figure in TFLOPS of power on the table, always theoretically, in order to have figures with which to speculate. The ideal in this regard would be for each byte transmitted, the system would have a capacity of 1 FLOP. However, this is not the case and with each new generation this number decreases due to the fact that the video memory lags far behind the processor.
In other words, with each new generation the ratio between computing power and bandwidth decreases, however, this must be qualified.
On PS4 Pro we saw an increase in computing power, but barely in bandwidth. In the case of PS5, the rate has been maintained due to the fact that despite the increase in the amount of TFLOPS, the leap to GDDR6 has been made. With the following system we have to start from the fact that we will have a new type of memory that will make the arimetic intensity not so high, but will follow a downward trend.
Bearing in mind that with each generation of PlayStation consoles the jump in computing power is less than the previous one and taking as a reference the worst scenario in terms of arithmetic intensity, then we are talking about a power of 49 TFLOPS . Which will surprise many of you due to the fact that it is less than that of the RX 7900 XTX and we are talking about a console for 2026.
Building the base system
Once we have a series of premises that allow us to get an idea, we have to build the system. However, we are going to focus on the graphic configuration and the processor or CPU, which are the most important of all. So we will leave the rest of the elements aside for the moment. As for memory? Remember that we are talking about a system with a bandwidth of 1152 GB/s and 32 GB of shared memory for all the elements.
What would the PS6 GPU look like?
PlayStation consoles since PS4 have maintained a configuration based on multiples of 9 in which the number of Compute Units or GPU cores is concerned. On PS4 there were 18 and on both PS4 Pro and PS5 there are 36. The number for an eventual PS6? Well, you should keep a multiple as long as you can use the console in PS5 mode. In this mode the system could disconnect one of the two GPUs to save power and lower the clock speed.
The configuration we expect? 72 Compute Units at 2.65 GHz sustained speed. If that seems little to you, keep in mind that the RX 7900 XTX stays at 2.3 GHz . Said clock speed could be higher and it is possible that when the console is finished AMD will achieve speeds of 3 GHz on a sustained basis without problems. In any case, it is the worst of our scenarios on the table.
AI, Ray Tracing and other elements
In order to achieve optimal performance for 8K and 60 FPS, it will not be enough to use raw power, but we will need the help of things like image reconstruction with more advanced versions of the FSR. The addition of AI accelerators in the RX 7000 opens the possibility for an increase in this capability in future AMD graphics architectures. Since we are talking about 2026 for the launch of the console , we would be talking about two generation leaps in terms of PS6 graphic architecture and therefore an eventual RDNA 5 .
Therefore, we expect that the jump in computing power capacity will be much greater in real figures than the 49 TFLOPS that we have mentioned before and that would be limited to only a part of the power of the PS6 GPU, but that cannot be quantify in the same way. More than anything we are talking about complementary elements that serve to increase graphic performance through small processors for specific tasks.
No, we have not forgotten about the processor
In an interview about the development of PS5 Mark Cerny stated that some other developer had asked him for a 16-core configuration for the PS5 CPU . However, they could not do so due to lack of space. Games today are fully multithreaded and divide their work among several processes that are computed in parallel and it is more than likely that such loads on games will increase again in the future. That is why we think that PS6 will have this configuration, using Zen 5 architecture for this, which by 2026 will already be on the market.
SONY will have to choose how much of the design they want to go to the CPU and how much to the GPU. If the first one falls short then it will be a drag on a too powerful GPU. So they have to find the balance between one part and the other. However, it is possible that we will see the trend of hybrid configurations applied . That is to say, use less complex cores, but that occupy less area as is the case with the last two generations of Intel processors with the E-Cores.
