Intel have recently launched their 12th Generation Alder Lake processors, and have taken a step away from their usual release process by re-designing this generation from the ground up. Not only does the Alder Lake series feature 10nm process, but it combines two different types of core — E-Cores and P-Cores — to deliver a single solution built on a similar principle to ARM's Big/Little core design, which supports hyper-threading and a host of new integrated technologies.
Smaller process, faster processing
Let’s begin with the new 10nm process that these 12th Generation chips feature. 2014 may seem a long time ago, especially after the trials and tribulations of the last two years, but in general terms seven years is not a significant amount of time. In the world of computing, however, in particular that of computer chips, with every two years that ticks by Moore’s Law dictates that advances in technology should be doubling transistor counts exponentially.
To stay true to Moore’s Law, in a perfect world we should essentially be three-times more advanced than we are now, but due to technical challenges, the global pandemic, and IC shortages, it has taken much longer for manufacturers to develop their technology. For seven long years the 14nm FinFet architecture has reigned supreme, with its debut coming in the form of Skylake in 2014, but with astonishing 10nm silicone allowing for an extra 63 million transistors in the same area, these new 12th Generation processors have finally knocked the 14nm FinFet architecture off its perch.
E-Cores and P-Cores
The new 12th Generation Alder Lake processors from Intel combine two different types of core in one chip, allowing for the processing of different types of tasks in the most efficient way. E-Cores, or Efficient Cores, are based on Intel’s low-power architecture and used for background tasks, where P-Cores, or Performance Cores, do the heavy lifting.
Efficient CoresThe Efficient Cores in the Alder Lake processors are single-threaded cores based on Intel’s low-power architecture found in Atom CPUs, which have been a staple in
embedded computing for around the last 13 years or so. Labelled ‘Gracemont’, this new architecture is a continuation of Intel’s Airmont, Goldmont, and Tremont architectures found in their Bay Trail, Apollo Lake, and Elkhart Lake processors respectively.
These E-Cores do share some of the features found in the recently launched Elkhart Lake CPUs, such as a dual 3-wide decoder unit, but the instruction cache has been doubled to 64KB, with an L1 cache of 96KB, and an L2 cache of 4MB per 4 cores. L2 cache serves to speed up the accessing of data by providing the processor all the necessary stored information without interruptions, especially data which has already been accessed so it doesn’t have to be loaded again, so these E-Cores provide a significant boost in performance where it counts. Combining these improvements, a single E-Core can deliver 40% more performance in a single-threaded application over previous generations while consuming the same amount of power. Impressive considering four of these E-Cores fit into the space of a single 6th Generation Skylake core, for example.
Performance CoresThe P-Cores, or Performance Cores, function much like traditional Core processors, in that they take care of priority tasks and those with high processing requirements such as AI-related computing. Named ‘Golden Cove’, these cores are a continuation of the Willow Cove processors found in the 11th Generation Tiger Lake U-series, and again feature sizeable upgrades and improvements to the design. With more AI-based applications being developed in both the commercial and industrial sectors, focus has been placed on better support for this by including AI acceleration technology for advanced matrix multiplication (AMX) as standard via a co-processor.
Thread Director brings balance
Having two types of processing cores responsible for processing certain types of task can’t be achieved through just hardware. This is why Intel have partnered with Microsoft to develop Thread Director, a new software-based system built directly into the hardware which routes processing to both E-Cores and P-Cores depending on type of task and other performance factors.
Optimised for Windows 11, Thread Director schedules tasks and monitors runtime instructions, providing feedback to the operating system to enable it to make optimal scheduling decisions for any workload or workflow. For instance, where the P-Core is predominantly where processing occurs for foreground apps, or bigger tasks like processing graphics instructions or playing games, Thread Director can make the decision to switch the roles of P-Cores and E-Cores to balance processing, but always within the boundaries of providing the most effective processing while using power as efficiently as possible.
More innovative technology and features
Intel’s 12th Generation Alder Lake processors bring some major feature updates geared towards applications that require high speed compute, communication, and I/O technologies.
Intel Deep Learning Boost
Intel’s Deep Learning Boost is a new built-in function aimed to accelerate training and inference workloads such as image classification, speech recognition, language translation, and object detection. Intel Deep Learning Boost introduces Vector Neural Network Instruction (VNNI) to extend their AVX-512 architecture and accelerate AI/Deep Learning inference, by combining the three instruction cycles adopted by the AVX-512 into a single instruction. This combining of instructions delivers a significant uptick in performance through better utilisation of computing resources and cache management, while also avoiding bandwidth bottlenecks.
DDR5 Memory Support
The new version of SDRAM, DDR5, has as many new features and improvements as the new 12th Generation Alder Lake CPU that supports it. We discuss in detail here, but some key points are as follows: DDR5 has higher bandwidth with lower voltage, higher capacity, longer burst length, and also a reengineered power control system which essentially replaces the requirement to draw power from the board by powering itself via power rails directly on the DIMM.
Up to 16 lanes of PCIe 5.0 and 4 lanes of PCIe 4.0
Alder Lake provides support for up to 16 lanes of PCIe 5.0 and 4 lanes of PCIe 4.0. This is significant for applications in the AI/Machine Learning space as it means high speed SSDs can be served by the PCIe 4.0 lanes while the 5.0 lanes are being used by high-performance graphics cards, or even 400GbE network cards.
Thunderbolt 4 - Universal Cable Connectivity
Thunderbolt 4 builds on its predecessor (aptly named Thunderbolt 3), delivering high-speed data transfer, high-definition video, and power, all via a single port. Furthermore, with a bidirectional data transfer speed of up to 40Gbps, Thunderbolt 4 can serve up to five Thunderbolt devices from a single port. With a standard USB-C type connector, Thunderbolt 4 is backwards compatible with all previous versions of Thunderbolt, along with USB, DisplayPort, and PCIe.
Intel’s Time Co-ordinated Computing (TCC)
Intel’s Time Co-ordinated Computing is all about real-time processing. Time-sensitive applications such as robotics and automotive often require tasks to be actioned within an allotted time span, with the failure to do so possibly resulting in a failure of the entire process in which that one task is just a component. Intel’s TCC prioritises real-time workloads' access to cache, memory, and networking, and minimises disruption from non-real-time workloads. This helps time-sensitive applications carry out their tasks uninterrupted and unhindered by otherwise non-critical processes.
In conclusion
We’re excited to see the potential of Intel’s new 12th Generation Alder Lake processors as our manufacturing partners begin rolling out supporting hardware. While we’re not expecting to see anything in mass production until some time in Q1 2022 and beyond, Alder Lake processors are sure to change the game for AI, machine learning, and IIoT applications as more products are developed and released. We look forward to providing innovative Intel 12th Gen powered solutions for your next generation of AI and embedded applications very soon!