Why the PS4 Will Be A Performance Beast

The PS4’s technical details have been kept under wraps for a while, but a recent flood of new information from Sony means that we can give a pretty accurate indication of how it is going to perform. We take a look at the nitty-gritty of the PS4’s technical design and specifications and explain why it might translate into blazingly fast performance which compares favourably to even high-end PCs.

The recent release of several nuclear-powered PC video cards such as the Nvidia Titan and ATI 7990, which are capable of jaw-droppingly ridiculous graphics output, have led some to question whether the upcoming generation of consoles such as the PS4 are still relevant when it comes to the raw performance arena. Some questions you commonly hear are: ‘Isn’t the PS4 equivalent to a standard mid-range PC?’, and ‘Why should I get excited about the technical specifications of the PS4?’

The answer to the first question is simply ‘no’. The PS4 can’t be judged on rudimentary numbers alone such as the amount of RAM or CPU clock speed – to get a true idea of the performance capability of Sony’s new console, you need to look at the architecture the system employs and how it has been specially engineered to reduce processing overhead and address memory more efficiently.

Let’s take a look at some of the key architectural features of the PS4 that will enable it to achieve off-the-charts performance.

A unified pool of high-speed memory

One of the biggest features touted by Sony’s technical staff in the leadup to the release of the PS4 is its unified, high-speed memory capabilities. The PS4 will come with 8 gigabytes of GDDR5 memory, which, implemented in a unified architecture, means that both the CPU and GPU can address the memory concurrently, without the performance penalty incurred from memory having to travel across a relatively slow PCI express bus.

What this means for game developers is that they can customise to precise detail how they want their games to utilize the system’s memory – for example in a high-res texture intensive environment, they could allocate 7 gigabytes of memory to the GPU and 1 to the CPU, or vica versa in a CPU-heavy environment.

Unified memory serves to almost completely eliminate bottlenecks in the system, which contributes a lot to performance in cutting-edge, demanding games, and most game developers tend to agree, with Sony tech lead Mark Cerny recently stating in an interview with GamaSutra that ‘the largest piece of feedback that we got (from developers) is that they wanted unified memory’.

GDDR5 memory itself (Graphics Double Data Rate, version 5) is extremely fast – much faster than your typical DDR3 ram you’d find in a standard PC setup. It provides more than double the memory bandwidth as its predecessor, GDDR3, and has an achieved data rate of 7Gbps (7 gigabits per second) per pin, which translates into an expected 178 GB/s of memory bandwidth for the PS4 – which is slightly more than triple the bandwidth of a current gaming PC.

8 core AMD CPU with Heterogeneous Uniform Memory Access

The PS4’s central processor will be a custom AMD job with a 64-bit x86 architecture (instead of the RISC architecture employed on the PS3 and other Sony consoles), and has 8 Jaguar cores capable of running 8 hardware threads, with 2MiB (mebibytes) of L2 cache per 4 core group, and a 32kib l1 instruction/data cache.

While the CPU clock speed is not yet confirmed, and is expected to be on the low side (1.6 ghz has been touted as an early speculative figure), the 8-core CPU should nonetheless provide blistering gaming performance.

The 8 cores will provide the system with incredible multitasking ability and the APU (accelerated processing unit) means that the CPU is integrated with the GPU, leading to rapid data sharing and reduced bottlenecks, which means better overall performance.

Diagrams released by AMD show how their new APU technology is going to work, and essentially it all revolves around the ability of the CPU and GPU to share data and communicate with one another.

This diagram shows a traditional architecture without hUMA (Heterogeneous Uniform Memory Access), and the resulting bottleneck that occurs when the CPU has to physically copy data from a GPU computation back to its own memory.

The next diagram shows an architecture with unified memory between the CPU and GPU, in which the CPU can read the result of a GPU computation directly without having to copy the entire data array.

Finally, we look at a diagram of a fully realised HAS (heterogeneous system architecture), which the PS4 will employ, in which the CPU and GPU caches can see the same, up to date view of data.

Overall, the CPU has been suggested to be capable of producing 1.8 teraflops (floating point operations per second) of raw processing power. Compared to something like the i7 920 processor seen in many high-end PCs which produces roughly 0.7 teraflops, and you can see that the PS4’s processor is starting to look quite impressive indeed.

Greater number of simultaneous compute commands

Further improvements to the AMD CPU architecture in the PS4 include the ability to process 64 sources of compute commands. This essentially means that asynchronous processing and multitasking are going to be significantly improved over previous consoles.

For the layman, asynchronous processing means processing that doesn’t block, or wait, for a response – so for example sending an email is an asynchronous task. You send the email and don’t necessarily wait around to receive a response – you can continue performing other tasks while you’ve sent that one off.

To relate this term to gaming, if a game has several asynchronous computations it needs to be performed, with the improved CPU architecture it can place commands in one of the 64 compute queues and the hardware will efficiently determine what processes should be run, how they should be run, and when they should be run.

This will lead to noticeable performance gains in multitasking, a feature which Sony have publicized aggressively, meaning that gamers will be able to conduct many tasks at once with little performance degradation, whether that’s playing games, watching movies, browsing the Net, or simply mucking about on Facebook.


image credit: Digitoll

You may also like...