Technical Comparison of the SoCs of the PS5, Xbox Series X and S Consoles

The following comparison is not to say which console is better, since that depends on the tastes of each one and if the system’s game catalog is more or less liked. A value that is completely subjective to the personal experience of each one of us. What we have done is analyze each of the three consoles from our hardware knowledge and we have drawn conclusions about the design of each one of them.

Size differences in PS5 and Xbox Series SoCs

The first thing that attracts your attention is that the SoC of PlayStation 5 is smaller than that of Xbox Series X, despite being two consoles that are in the same price range, since while that of Xbox Series X measures 360 mm ^2. , that of PlayStation 5 is a little below 300 mm ^{2> / sup>. Although the smallest is the Xbox Series S, with a size of 190 mm 2 .}

The factors that most influence the size of each SoC are, on the one hand, the configuration in the number of Compute Units of the GPU integrated in each of the SoCs, and on the other, the memory interface. In the case of Xbox Series S we have a 128-bit GDDR6 interface, on PS5 256-bit GDDR6 and on Xbox Series X 320-bit GDDR6. Since in all processors the interface with the memory is always placed on the external perimeter of the same, this affects the size of each of the SoCs.

Obviously by increasing the perimeter the available area is also increased, in any case it must be clarified that the specifications come after choosing the size of a processor, since this will mark the amount of SoCs that will come out per wafer and therefore what will be production volume of a console.

Another important factor to take into account is the issue of costs, consoles are sold with lower margins than traditional hardware and are financed in the long term by the royalties that their manufacturers collect from the games and services associated with it.

Differences between CPUs

Officially, the three SoCs use AMD Zen 2 cores, specifically they are the same as those of the AMD Ryzen 4000, Renoir, for PC. Each of the three SoCs uses 2 CCXs, each comprising 4 cores, the L1 and L2 caches for each core, and 4 MB of L3 cache.

In the case of PS5 we find the following omissions:

• The registers for the AVX units have been shortened, this prevents SIMD calculations with 256 bits of operands, the maximum being 128 bits per instruction.
• The floating point summation unit, FADD, has been removed. Such an instruction can be done in the FMADD unit that performs an addition and multiplication operation in an instruction cycle, but with a higher latency in clock cycles, 5 instead of 3.

As for the clock speed of each console is where we also find differences. In the case of SONY, the CPU works at 3.5 GHz, but it cannot go beyond that speed due to the limitation imposed by SONY itself regarding the use of SmartShift in the console. In the case of Xbox Series, these may or may not have the core SMT enabled. Without the active SMT Xbox Series S can reach 3.6 GHz and Xbox Series X 3.8 GHz, speed that is reduced to 3.4 GHz and 3.6 GHz when it is active.

Is it the PS5 RDNA 2 GPU?

There is a controversy in which many people claim that the PS5 GPU is not RDNA 2, despite AMD and SONY claiming that it is. However, it is necessary to clarify where the confusion comes from and is that when choosing the different parts in the AMD portfolio, SONY has dispensed with certain pieces of the RDNA 2 that Microsoft has added in their consoles. In any case, if we are purists, none of the consoles is RDNA 2 in full, as all three lack the Infinity Cache that the AMD RX 6000 PCs have.

What part has SONY decided to do without? Well, the rasterized unit and its Render Backend are the same as in RDNA and therefore they are not the RBs of RDNA 2+. Which means that the non-SONY console can use hardware Variable Rate Shading in games. This function groups the repeated Pixel Shader instruction lines into a single instruction, which is executed only once and not several times. Once it has been solved, it applies the data to the different pixels on the screen with the same value.

The lack of Variable Rate Shading is a performance advantage for Microsoft consoles, especially considering that the most powerful part of Pipeline 3D is the application of Pixel Shaders. Already in PS4 Pro SONY asked AMD to make changes to the RB units, which generate the ID Buffer for the Checkerboard Rendering. So SONY would have kept the PS4 Pro units and would not have adopted the RDNA 2 RB +s for backwards compatibility.

General specifications of the three GPUs

The first thing that stands out is the specifications in terms of the number of Compute Units. The least ambitious of the three, Xbox Series S, has a configuration of 20 Compute Units and runs at a stable clock speed of 1565 MHz to reach 4 TFLOPS. Going up in clock speed, the next is Xbox Series X that works at 1825 MHz, but with a configuration of 52 Compute Units that allows it to reach 12 TFLOPS of power. PS5, on the other hand, reaches higher clock speeds and can reach 2.23 GHz thanks to the use of SmartShift technology, but it only does so at certain times and despite this it does not reach beyond 10.28 TFLOPS.

Summarizing the data, Microsoft has not given a Boost speed to its consoles so that they perform the same all the time, while SONY has given us the Boost speed, but we do not know the base clock speed at which its console runs the highest part of the time.

For its SONY console, it adopted SmartShift technology, which varies the CPU and GPU clock speeds depending on the workload of each. So when the CPU is under low workload then the speed of the GPU can go up and vice versa. Although the SmarShift is a bit more complex, but it helps the SONY console to compensate for the lower number of Compute Units on the GPU.

Shader Arrays and GPU Cache Hierarchy

The first thing we will look at is the Shader Array, where they share space with the Render Backend or RB, the raster unit and the primitive unit, also known as the tessellation unit.

Each Shader Array is connected to an intermediate cache, called GL1 cache by AMD, this cache is 128 KB and is only read for its clients, since when writing export data these go directly to the GL2 cache of the GPU. In RDNA architectures GL2L2 cache partitions depend on the number of memory channels that are available, each partition is 256 KB. Consoles using GDDR6 memory where each memory channel is 16 bits have the following GL2 partitions;

• Xbox Series S: 128-bit GDDR6 bus, 8 GL2 cache partitions.
• PlayStation 5: 256-bit GDDR6 bus, 16 GL2 cache partitions.
• Xbox Series X: 320-bit GDDR6 bus, 20 GL2 cache partitions.

Now, in both RDNA and RDNA 2 the ratio of GL2 partitions per GL1 partition is 4. This means that Xbox Series S has 2 Shader Arrays in total, while PlayStation 4 Shader Arrays. The exception to the rule? Xbox Series X should have 5 Shader Arrays, but it has 4, since the ratio in your case is 5: 1. Given the configuration with a greater number of Compute Units, the Xbox Series X SoC houses more units in each of its Shader Arrays than those of the PS5 and Xbox Series S

GPU Compute Units on PS5 and Xbox SoCs

If we extract each Compute Unit of the PS5 SoCs and the Xbox Series individually, we will not observe differences. They all use RDNA 2 CUs, regardless of the console we are talking about. The differences between RDNA 2 and RDNA 2? Really null if we do not take into account the unit for acceleration for Ray Tracing.

What does affect performance is the organization of the Compute Units in the Shader Array and this will depend on each game. When a polygon is rasterized through the unit for it included in each Shader Array, the number of pixels in that polygon is distributed among the different Compute Units in it. So the degree of use will depend at first glance on the size of the fragments.

Nowadays, games use very small polygons, so it is normal that in the first shipment not all Compute Units are used, to this we must add that there are tasks related to the computing pipeline that require the use of Compute Units leftovers. And at this point where the strengths and weaknesses are seen, especially between PS5 and Xbox Series X.

If we talk about Ray Tracing, it purely benefits from the calculation speed that Compute Units have, so Xbox Series X has an advantage in this. It is as if SONY had designed its console to take PS4 Pro games to the highest possible level and had not taken into account elements such as Ray Tracing until the last minute in the design.

Since the intersection unit is within the Compute Unit, the PS5 is also at a disadvantage in that regard, having only 36 of these units compared to the 52 of the Xbox Series X.

Audio implementation on the PS5 and Xbox Series SoC

The next point in our comparison between the PS5 SoCs and the new Xbox is the audio. Where both SONY and Microsoft have covertly integrated the previous generation audio systems, as they are necessary for backwards compatibility in games. At the same time they have added new audio systems, which are more advanced than those of the previous generation.

We do not have much data from the two systems, since both companies have said very little about them. In both cases we find that the computing power of the audio system is greater than the previous generation CPUs. Beyond that the details are scarce. but there is a detail in which both consoles differ and it says a lot about the strategy of each one in terms of audio. The console does not support Dolby Atmos by hardware, this means that the games cannot use it and instead are forced to use a proprietary standard from SONY.

On the other hand, the Xbox Series X and S support Dolby Atmos by hardware, but to use it then the user has to pay a user license that is sold separately. This license has a cost of $ 15 and both manufacturers have decided to do without it on their consoles. In the case of Microsoft, making the payment optional for users, but in SONY’s case, dispensing with it in games, although it is implemented in multimedia content, so in movies and series it is the CPU that decodes the Dolby Atmos.

Codecs for the SSD

The SSD is the new storage unit, for the way both consoles work is different when it comes to accessing data and communicating with the console’s storage at the SoC level:

• Both consoles have two different channels to the SSD, one for the CPU and one for the GPU. This is because the RAM in the three consoles is divided into two spaces, depending on whether the CPU or GPU make the request for data to the SSD, the information will go to one part of the RAM or another.
• Each channel between the SSD and the RAM has a different codec, and each of the consoles uses its own codecs.
• For textures PS5 makes use of Oddle Kraken, while Xbox make use of BCPack.
• The bandwidth with the PS5 SSD is higher, but in return the compression ratio is lower per transmitted byte. On the other hand, the maximum transfer speed of Xbox is lower.
• The Xbox Codec is designed to be used in DirectStorage, allowing Sampler Feedback.

Hardware codecs for the SSD are essential to fully utilize these drives, as without these drives it would take a lot of CPU power to decode data fast enough. We would even be talking about whole cores just for this task.