TECHnalysis Research Blog

August 19, 2021
Intel Lays Out Multi Chip Architecture Plans

By Bob O'Donnell

Intel is on a roll. Fresh off an impressive discussion of its long-term chip manufacturing process roadmap at its Intel Accelerated event, and the debut of its new Intel Arc brand for discrete graphics chips (GPUs) aimed at gamers, the company provided an equally impressive and comprehensive view of the designs that will be powering these and other future chips at its latest Architecture Day 2021.

Specifically, the company talked about a huge range of items including:

• Two new CPU core designs—one termed an efficiency core and the other a performance core
• A finished CPU chip for PCs (likely to be called Intel Core 12th Generation, but codenamed Alder Lake) that leverages those designs and integrates an intriguing element Intel calls a Thread Director
• A data center-focused CPU (codenamed Sapphire Rapids) that uses the new performance core
• Two new GPU architectures based on the company’s latest Xe GPU core
• Two complete GPU SOC designs (one focused on gaming graphics codenamed Alchemist, the other on AI and high-performance computing codenamed Ponte Vecchio)
• A next-generation IPU (infrastructure processing unit) leveraging FPGAs and a custom ASIC that’s designed to simplify the creation of cloud computing-focused data centers
• Numerous software advancements, including an AI-powered graphics tool called XeSS that can dramatically improve the speed and quality of 4K gaming visuals, and the latest developments in its OneAPI framework for simplifying the process of writing software that can use all these different chips

To be honest, it was a bit overwhelming, but there are several key takeaways from the event. First, it’s clear that Intel is serious about not only retaking its crown in the world of high-performance CPUs, but it also wants to make its mark in GPUs, IPUs, ASICs and multiple different types of chips. In addition, the company is making a strategically critical pivot towards integrating smaller, individual elements together to achieve that success. To put it succinctly, it’s planning to use multiple tile-based designs across multiple types of chips.

In the CPU space, this type of hybrid architecture represents a big change for the company, but it is something competitors like AMD and Apple have shown can provide a big performance advantage. Truth be told, much of this stems from Arm’s Big.little mobile chip design methodology that it first debuted back in 2011 with the Cortex-A7 CPU. In these early Arm designs, the focus was on power savings because of its use in smartphones and other battery-powered devices. As AMD, and then Apple with its M1 chip, recently illustrated, however, similar hybrid computing concepts can be used to combine multiple cores together for big performance gains as well. These chiplet-based designs, as they’re sometimes called, use various types of fast interconnect technologies to make multiple computing “tiles” essentially function as one big unit.

In Intel’s case, the company has created a new efficiency core and a new performance core that it is combining in its next generation Alder Lake CPUs for both desktop and mobile PCs. A new twist that Intel is also bringing to Alder Lake—which is expected to make its official debut in the fall—is a new hardware element the company is calling Thread Director. Thread Director is going to provide some unique advantages to Alder Lake PCs running Windows 11 (and should show some benefits on Windows 10-based Alder Lake machines as well).

The reason for this goes back to how these hybrid, multi-core CPU designs function. As appealing as the idea of combining multiple computing units into a single logical unit may sound, without the appropriate software to take advantage of that design, it won’t be of much use. For Windows-based PCs, it turns out that much of the hard work of determining what elements of a software application workload should be sent to what type of core is handled by the Windows Thread Manager. Unbeknownst to many, however, is that, up until now, those decisions were often made on intelligent guesses of what a given process on the CPU was doing. With the new Thread Director, however, Intel has created a hardware-based mechanism for providing telemetry information about the types of instructions and specific needs that a given thread is using on the CPU. As it dynamically collects that data, Thread Director sends the details back to the Windows Thread Manager, allowing it to more intelligently allocate where different threads should go. The bottom line is more efficient use of the capabilities of a hybrid, multi-core CPU, which should translate into much better performance overall.

In addition to the technical benefits of Thread Director, the work clearly shows a new level of cooperation between Intel and Microsoft to achieve this tight level of integration. By focusing the biggest efforts on Windows 11, it also shows how the two companies are once again working to drive the PC industry forward—a very positive development in my opinion.

The new GPU-focused efforts are also important steps forward for Intel. Challenging both Nvidia and AMD in the discrete GPU market is no small task for the company—especially given their previous missteps in the market. However, given the increasing importance of GPUs, not only for gaming, but also AI acceleration, data analytics, and other applications, it’s easy to understand why Intel is choosing to make these efforts.

Finally, while it is clearly worthy of significantly more detailed coverage on its own, a few comments are in order about the data center-focused Ponte Vecchio GPU SOC detailed at Architecture Day. What’s particularly interesting about this massive new chip isn’t only the impressive early benchmarks—Intel demonstrated it beating Nvidia’s current best performance for inference and training on the ResNet 50 convolutional neural network—but also the fact that it incorporates so many of Intel’s new initiatives into a single component. Not only does it include pure architectural advancements in discrete GPUs, but it also integrates multiple process and packaging technologies in a way that only Intel can. From Foveros chip-stacking and EMIB high-performance interconnects to the combination of the new Intel 7 process technology along with cutting edge tiles from other foundries, Ponte Vecchio looks to be a technological tour-de-force.

Obviously, Intel needs to execute on its aggressive vision across chip design, packaging, and manufacturing to achieve success with Ponte Vecchio, but it certainly stands as an impressive statement on where the company looks to be headed.

Here’s a link to the original column: https://www.techspot.com/news/90859-intel-lays-out-multi-chip-architecture-plans-12th.html

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on Twitter @bobodtech.