Monday, April 6, 2020

Power vs. Speed: The Xbox Series X And PS5 are different animals


An exploded view of the Xbox Series X

Microsoft/Digital Foundry

Some were of the opinion there would be little difference between what we now know as the Xbox Series X and PlayStation 5 when news about the core technology in the two consoles began to surface. The opinion was based on both consoles having an AMD Zen 2 CPU, an AMD RDNA 2 GPU, 16GB of GDDR6 memory and lightning fast SSDs.

Those who promoted this idea should have known better. The core technology doesn’t determine the outcome, it establishes a work space for platform holders to create complex gaming systems. The systems they create provide developers with a playground of opportunities and constraints. How games perform on different systems depends on how the developers take advantage of the opportunities available within the limits imposed by the constraints. Identifying the core tech doesn’t address questions about how systems are designed let alone questions about what developers can do within the confines of different systems.

Last week Microsoft shared a wealth of technical detail about the Series X with Digital Foundry and Sony presented a faux public presentation of its cancelled GDC talk about the PS5 featuring lead system architect Mark Cerny. At this point we know more about the Series X than the PS5 but we know enough to see the engineers at Microsoft and Sony have created machines that are very powerful, very fast and very different. Microsoft built for power while Sony emphasized speed. This is going to be a very interesting console generation.

The heart of the Xbox Series X

Microsoft/Digital Foundry

Power: The CPU

The Xbox One and PS4 consoles were limited by relatively weak CPUs. That won’t be the case with the Series X and PS5. Both of the next-gen machines feature customized AMD Zen 2 CPUs with 8 cores and 16 threads. Microsoft is reserving one core (two threads) for the operating system and front-end shell. The remaining 7 cores run at 3.8 GHz when SMT (simultaneous multi-threading) is disabled. That’s very fast. For comparison purposes, the AMD Jaguar chips ran at 2.3 GHz in the One X and 2.1 GHz in the PS4 Pro.

Microsoft is also giving developers the choice to enable SMT on 5 cores (10 threads) at 3.6 GHz. Microsoft claims multi-threading (if it’s done well) provides a performance boost of 30% or more. Nevertheless, it’s expected many developers will choose the higher clock speeds until they learn to use multi-threading effectively.

Cerny did not disclose how many cores in the PS5 are reserved for the operating system. However, he made it clear that clock speeds are not locked as they are in the Series X. Instead, clock speed is variable and capped at 3.5 GHz. What this means is that the CPU will slow down when the demands on it are high. If the CPU is operating at less than full capacity, it’s unused power can be shifted to the GPU.

The CPU setup in the Series X will deliver more compute power than the PS5. Its fastest clock speed (3.8 GHz) is clearly faster than than the PS5’s 3.5 GHz cap. The slowest clock in the Series X (3.6 GHz) is still marginally faster than the PS5’s cap but it’s capable of a 30% performance boost over the 3.8 GHz clock. Also, the PS5 clock will slow down when demand is high.

What does this mean for games? The GPU creates what you see on the screen, the CPU does everything else including handling input from the player, running the game simulation and handling physics and movement. The power difference between the Series X and PS5 could be significant. Or not. It will depend on how developers make use of the different opportunities and constraints provided by Microsoft and Sony’s very different approaches to CPU power.

"Hellblade 2: Senua's Saga" rendered in engine on the Xbox Series X

Ninja Theory

Power: The GPU

Microsoft locks CPU clocks in the Series X while Sony lets clocks vary with demand in the PS5. The companies follow the same approach with their custom GPUs. The GPU in the Series X is a customized AMD RDNA 2 chip with 52 compute units running at a locked 1.825 GHz. The PS5’s RDNA 2 chip has 36 compute units capped at 2.23 GHz. The compute units in the PS5 are faster but there are fewer of them. The setup in the Series X produces 12 TFLOPs (teraflops or trillions of floating-point operations per second) while the PS5 is rated at 10.28 TFLOPs.

The Series X has a more powerful graphics engine when measured in terms of TFLOPs. Cerny argued that TFLOPs are not the only way to measure the capabilities of a GPU, however. TFLOPs measure compute power which is the core function the chip is designed to carry out. However, the GPU carries out other functions such as moving data in and out of various memory locations. Cerny’s argument was that these other functions also benefit from the GPU’s clock speed and measurement in terms of TFLOPs does not take these additional benefits into account. He provided a hypothetical example without test data to back it up.

Will Cerny’s additional benefits be sufficient to reduce or eliminate the Series X’s advantage in compute power? The situation is complicated by the variable clock in the PS5 GPU. 10.8 TFLOPs is the peak performance average. What happens to performance on a moment-by-moment basis when the GPU is under consistent sustained load? Cerny believes reductions in speed will have marginal effects but supporting test data were not referenced in his talk.

The GPUs in the Series X and PS5 both support hardware accelerated ray tracing. Microsoft has also stated its GPU allows for variable rate shading (which links screen resolution with visibility) and mesh shading (which allows more refined detail in shapes and movement). In addition, Microsoft worked with AMD to build machine learning capability into its GPU. Cerny talked about “primitive shaders” in the PS5’s GPU which sounds like something akin to mesh shading. He did not mention variable rate shading or machine learning.

Will the series X meet Microsoft’s goal of providing 4K/60fps for all games? Will the PS5 fall back on techniques like upscaling or dynamic resolution scaling if its GPU clock slows down too much under load? As was the case with the CPU, we won’t know the answers to these questions until we see how developers handle the opportunities and constraints inherent in Microsoft and Sony’s approaches to GPU power.

The PS5's SSD has a blazing 5.5 GBs throughput

Sony

Speed: The SSD

The CPUs and GPUs in the Series X and PS5 are much more powerful than their current counterparts but the most radical difference between the next-gen consoles and their predecessors is the SSD. Consoles have never been built to take advantage of any SSD let alone the inordinately fast SSDs custom designed for the new consoles. These SSDs open up possibilities never seen in consoles before such as virtually unnoticeable load times.

When it comes to speed, the PS5 is the clear winner. Sony’s machine has an 825GB NVMe drive delivering 5.5 GBs. The Series X has a 1TB NVMe drive delivering 2.4 GBs. The SSD is smaller in the PS5 but it’s much faster.

Functional speed is determined by the slowest component in the system. A lightening fast SSD isn’t going to be much use if a slow up or downstream process introduces a bottleneck. To eliminate choke points, Sony created a custom decompressor, a controller that directs the data flow from the SSD, and separate processors to take care of I/O and memory mapping. In addition, developers can use a six-level system to prioritize data streaming from the SSD in terms of how fast the game needs it. Microsoft also created a custom decompressor and rewrote DirectX to handle I/O at their SSD’s increased speed. The Series X also incorporates refined memory management techniques that eliminate the need to load textures and assets that end up not being used.

These SSDs are going to have very noticeable effects on games. The virtual elimination of loading times has been widely publicized but that’s not the end of the story. The increased throughput between the SSD and the CPU/GPU gives developers access to more data more quickly which translates into more lifelike images based on richer textures and denser underlying geometry.

Microsoft is also touting Quick Resume which lets players return to the exact point where they left off when switching between one game and another. Digital Foundry saw a demonstration of Quick Resume working with four games and taking about 6.5 seconds to switch from one game to another. Sony hasn’t mentioned anything like a Quick Resume feature, but it’s hoped the PS5 will have something similar. If it does, the time to switch between games may be even faster.

The SSDs have important consequences for games but it’s too soon to tell whether the PS5’s speed advantage will give it a performance advantage over the Series X. How speed plays depends on how developers handle opportunities and constraints.

The next-gen consoles will open up new worlds of gaming

Sony

Conclusion

The Series X and PS5 are very different animals. Microsoft fixed the CPU and GPU clocks and focused on power; Sony used variable CPU and GPU clocks and prioritized speed. How all this plays out remains to be seen. I expect “Which is the better console?” will be a pointless question. Based on what we’ve seen thus far, they’ll both be great.

For those who care more about gaming than brand names, the next gen consoles offer an exciting opportunity to compare the effects of power and speed. Cross-platform games from developers who are equally competent with both systems will tell the tale. I expect some games will shine on the Series X while others will be better on the PS5. If this happens, working out why it happened can provide new insights into the contributions power and speed make to different aspects of gaming. This is going to be fun.



from Hacker News https://ift.tt/34fVZ1c

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.