Future PC and data center processors will be based on mobile design principles

For many years, mobile processor manufacturers have been committed to optimizing designs to achieve the best performance within a limited power budget, memory space, and bandwidth. In the past, these considerations were obviously not taken seriously in markets such as data centers or personal computers (PCs). Today, changes are quietly taking place in the traditional data center and PC markets - changing the rules of processor design and forcing developers to rethink their chip architectures to achieve higher performance-power ratios.

Mobile processor design principles applied to PCs and data centers

Today, more and more cloud gaming, data mining, artificial intelligence/data analytics, and high-performance computing are being implemented in the cloud. Although the requirements of these applications vary, they all share the same need for increasing computing power.

Data centers cannot meet this demand by continuously expanding their physical footprint. In order to keep operating expenses (OpEx) within acceptable limits and achieve Net Zero goals, enterprises need to increase computing density within limited space to achieve higher computing performance. Processing elements such as graphics processing units (GPUs), central processing units (CPUs), and artificial intelligence (AI) accelerators must achieve the highest performance within the minimum power/heat dissipation and area budgets. To this end, designing processors following mobile design principles is an ideal starting point.

Analyzing the trends in the PC market, similar conclusions can be drawn. In the traditional PC model, most functions are integrated into separate modules. However, as most organizations want to expand the hybrid office model, people are gradually replacing desktops with laptops. Integrating more and more functions (including graphics processing, neural network acceleration, security, I/O, etc.) into a single system-on-chip (SoC) with a unified memory architecture can both improve performance and keep power consumption to a minimum. Therefore, the next generation of PC processors looks more and more like smartphone processors.

Big tech companies are using custom chip design to differentiate themselves

With the end of Moore's Law, it is no longer possible for the industry to see performance improvements every two years. In this context, companies have joined the design "race" to create the best user experience with the best chips.

The world's large technology companies are well aware of this and are well prepared. They are focusing on designing custom chips for use in consumer products, PCs, or data centers. These companies are turning from off-the-shelf chips to custom chips, hoping to gain better control over the design and gain an advantage. As a result, we see Amazon investing in Graviton CPU design and Google launching the TPU-centered Tensor CPU. Apple's M1 processor will bring a chip optimized with mobile design principles to Mac computers, providing higher integration and better performance-power ratio.

OEM Alternative

For original equipment manufacturers (OEMs) that do not have in-house hardware and software design teams and have not yet embarked on the development of custom chips, the challenge they face is how to make their designs stand out and compete with highly optimized architectures. Most of the chips these OEMs use are off-the-shelf, which may put them at a disadvantage. Many chips designed for PCs and data centers are "brute force" solutions that can provide the required performance, but are usually too power-hungry and memory/bandwidth-intensive to be competitive. In addition, these chips are also limited in terms of the specific software and operating systems that can be used.

Some SoC suppliers in the mobile market have begun to enter the data center and PC markets, hoping to carve up some of the market share of existing players, but their number is very small, and it is difficult to help OEMs achieve differentiation in both innovation and cost control. Therefore, some industry alternatives are emerging. We have seen many manufacturers considering CPU solutions based on the RISC-V architecture. However, a single CPU design cannot completely solve the fierce competition dilemma currently faced by OEMs. OEMs need to look at the entire data center architecture to enhance the innovation of the overall solution and thus increase their own competitiveness.

Scalable heterogeneous architecture is key

Through heterogeneous computing, CPU, GPU and other computing units can be flexibly used to achieve maximum hardware utilization to optimize computing performance while meeting efficiency and power consumption optimization. Heterogeneous computing architecture provides a flexible array working solution for the ever-increasing computing requirements of data centers. Currently, many semiconductor manufacturers are researching related products and applications to enable OEM market competitiveness. Imagination, a traditional GPU IP company, launched its CPU product line last year and strengthened the research and development of heterogeneous computing, intending to provide customers with more complete heterogeneous computing solutions through the optimization and improvement of product portfolios, so as to better serve customers and adapt to the needs of future high-performance computing.

Mobile GPUs lay the foundation

Mobile GPUs are an ideal entry point for creating efficient heterogeneous designs. Compared with trying to force high-end GPUs into the mobile power budget, it will make more sense to upgrade mobile GPUs and apply them to data centers and PCs. Because mobile GPUs are born to be "small and beautiful". Mobile GPU manufacturers have developed many patented technologies to maximize GPU high performance and low power consumption. In the context of data centers and PCs pursuing high-performance GPUs in a crude and single-minded manner, these technical advantages can make mobile GPU manufacturers more competitive and provide OEMs with more management added value.

When it comes to patented mobile GPU technology, we have to mention Imagination, a long-established company that focuses on GPU design. Compared with its competitors, Imagination has focused its main efforts on GPU research for many years, especially in the more complex field of GPU rendering. Imagination is a pioneer developer of many technologies, such as GPU hardware virtualization, block-based deferred rendering (TBDR), and real-time hardware ray tracing (Ray Tracing). Block-based deferred rendering (TBDR) technology divides geometric data into small areas (tiles) and processes them uniformly. Since each tile is rasterized and processed individually, the rendering size is very small, so all data can be saved in fast-running on-chip memory. This technology lays the foundation for M1's graphics processing.

For application scenarios such as Android cloud gaming, data centers need to flexibly handle different gaming consumption scenarios for multiple users. Processing multiple small concurrent workloads on multiple small GPUs is more efficient than using traditional desktop GPUs. The cloud gaming industry chain is strengthening the development and application of GPU hardware virtualization technology to reduce costs. Mobile GPUs can support more users per GPU by expanding the distributed multi-core mobile GPU architecture upwards, while providing higher energy efficiency for many users in the cloud.

Take Innosilicon as an example. As the industry leader of the first domestic high-end GPU chip, the company has expanded the mobile GPU architecture upward to high-performance server-level hardware based on Imagination's mobile GPU IP, aiming to break the existing pattern of the desktop graphics card market. In this high-end market that has long been dominated by a duopoly, no one expected the emergence of new competitors, but Innosilicon is using changing market forces and highly scalable and efficient technology to provide alternatives.

Adding efficient on-chip AI processing (as shown in the M1) is another opportunity for OEMs. Since on-chip AI processing is not yet standard for PCs, OEMs can use this capability to support emerging applications such as super-resolution noise reduction, audio commands, security, etc. This AI function usually requires huge computing power, and the use of neural network accelerator (NNA) IP designed based on mobile design principles can integrate efficient and highly reliable AI inference functions on the SoC. In the field of AI edge accelerators on the terminal side, compared with other competitors, Imagination's NNA edge accelerator hardware not only inherits the high-performance and low-power DNA of its GPU design, but also has superior performance than competitors in computing fields of different orders of magnitude.

Designing specialized chips – not just for big tech companies

SoC manufacturers need to create energy-efficient, high-bandwidth, and high-performance designs through scalable IP cores designed based on mobile design principles. With this processor designed for heterogeneous architecture, they can create dedicated, efficient new solutions. This can help OEMs provide highly competitive and differentiated products and firmly grasp the future development direction of the enterprise.