The Next Chapter in our Graphics Story: Intel® Arc™ High Performance Graphics
Author: Roger Chandler, Intel Vice President and General Manager of Client Graphics Products and Solutions
It’s been a busy week for the graphics team at Intel. In case you missed our updates, I wanted to capture several of the key announcements and details we shared regarding our new performance graphics brand, details on the Xᵉ HPG microarchitecture, our AI-accelerated super sampling capabilities, and new software driver capabilities.
Intel’s new performance graphics brand: Intel® Arc™
This week we announced our new performance graphics brand: Intel® Arc™.
Our advancements in integrated graphics have made gaming and creating more accessible to a larger audience, even on ultra-slim devices, but our product line was missing discrete GPUs that scale up into the enthusiast gaming space. In 2018 we began building a roadmap of high-performance graphics architectures for desktops and laptops. Our journey to performance graphics began a new story for Intel.
We all have stories to tell, and we form new ones every day when we work, when we play, and when we create. Each story has a structure, with plot and character inflections that flow in an arc. This inspired the name for our new consumer high performance graphics brand.
Arc is more than just a name; it’s a commitment to the future spanning multiple hardware generations. Each generation has its own codename, and we wanted them to be more relatable for gamers than just 3 letter codes. Our upcoming product family — what has previously been known as DG2 — is now codenamed Alchemist. We are already hard at work on the next generation, which will be codenamed Battlemage, and that will be followed by Celestial and then Druid.
More details on the Xᵉ HPG microarchitecture
At Architecture Day we revealed many new details about our Xᵉ HPG microarchitecture.
Alchemist will be our first family of SoC’s based on the new Xᵉ HPG microarchitecture and will be available in the market in Q1 of 2022. Xᵉ HPG draws from our Xᵉ HPC, Xᵉ LP, and Xᵉ HP microarchitectures to deliver the compute efficiency, graphics efficiency, and scalability required for high-performance gaming.
Alchemist SoCs are built with a set of Xᵉ cores, a new high-efficiency compute building block designed to deliver the foundation required for high performance graphics. We previously discussed our GPUs in terms of execution units, or EUs, but as we scale with Xᵉ HPG and Xᵉ HPC, our focus is moving to a better representation of the microarchitecture’s configurability. Each Xᵉ core combines caches, logic for memory transfers and a set of arithmetic units.
The core Xᵉ ISA, or instruction set architecture, gives us the flexibility to adapt and scale Xᵉ cores for different workloads and products. We’re not ready to share all the details just yet, but I can reveal that the Xᵉ cores in Alchemist SoCs each have 16 vector engines and 16 matrix engines, which we refer to as XMX, or Xᵉ Matrix eXtensions.
While Xᵉ cores provide compute capacity, render slices combine them with the fixed function rendering units required to produce 3D graphics. Each render slice pairs four Xᵉ cores with four ray tracing units that fully support DirectX Raytracing and Vulkan Ray Tracing standards. Render slices also add samplers, pixel backends, and geometry and rasterization pipelines that are all designed for DirectX 12 Ultimate.
To achieve the performance required by gamers and enthusiasts, we can connect up to eight render slices to a high-bandwidth memory fabric and enable up to 32 Xᵉ cores per SoC. This much wider configuration is key, but the improvement compared our current discrete graphics product did not stop there. To deliver more performance and more efficiency, the engineering teams worked tirelessly to optimize the architecture at all levels and to improve the logic design, circuit design and software. We’ve also partnered with TSMC to fabricate Alchemist SoCs on the N6 process, which offers great characteristics for GPUs. These efforts yielded a 50% improvement over the previous generation both in performance per watt and in clock frequency at the same voltage.
Xᵉ Super Sampling scales performance and quality
We are excited to unveil more details about AI accelerated super-sampling capabilities of our Alchemist-based products, which we are calling XᵉSS.
Gaming has long required making trade-offs between performance and quality. Today, it’s common at ultra-high resolutions like 4K and with cutting-edge techniques like ray tracing. But modern upscaling techniques are tipping the balance, and we’ve been working on one of our own.
XᵉSS touches the sweet spot by using deep learning to synthesize frames that are very close to the quality of native high-resolution rendering. It reconstructs subpixel details based on information from not only neighboring pixels, but also motion-compensated previous frames that add temporal awareness. This process runs on a neural network trained to deliver high performance and fidelity, and it’s accelerated by our XMX hardware.
We want the benefits of XᵉSS to be available to a broad audience, so we developed an additional version based on the DP4a instruction, which is supported by competing GPUs and Intel Xᵉ LP-based integrated and discrete graphics. Our XᵉSS APIs are designed to integrate into today’s game engines, and we’re committed to working with the development community. The SDK for the initial version of XᵉSS based on XMX will be available to ISVs starting this month, while the DP4a version is due later this year. We believe in open-source standards and will open up the tools and SDKs for XᵉSS as they mature.
Software gives you more control
Drivers are an important part of the experience. We’ve made big strides recently with our integrated graphics, improving throughput for CPU-bound titles, accelerating load times by enhancing shader compiling, and implementing major changes affecting over 100 games. For the past three years, we’ve also been working closely with Microsoft to co-engineer DirectX 12 Ultimate. In addition to supporting ray tracing effects via DXR, Intel ARC graphics products will be capable of boosting performance with variable rate shading tier 2 and unlocking greater geometry details with mesh shading.
Many gamers are also creators, so we’re developing robust capture capabilities that leverage our powerful encoding hardware. These include a virtual camera with AI assist and recorded game highlights that save your best moments. We’re even integrating overclocking controls into the driver UI to give enthusiasts the tools they need to push the hardware to the limit.
Join us on the journey
We’ll share more of our story later in the year. Please go to Intel.com/arc for more details, and to see some of the videos we produced to help explain Intel Arc and showcase a sneak peak of the pre-production Alchemist SoCs in action. Developers can also follow us through the Intel GameDev Zone, where we’ll make additional Arc resources available over time.
Thank you for reading and for being a part of our story!
Notices and Disclaimers:
Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. No product or component can be absolutely secure.
All product plans and roadmaps are subject to change without notice.
Statements in this document that refer to future plans or expectations, including with respect to future technology and products and the expected benefits and availability of such technology and products, are forward-looking statements. These statements are based on current expectations and involve many risks and uncertainties that could cause actual results to differ materially from those expressed or implied in such statements. For more information on the factors that could cause actual results to differ materially, see our most recent earnings release and SEC filings at www.intc.com.
Code names are used by Intel to identify products, technologies, or services that are in development and not publicly available. These are not “commercial” names and not intended to function as trademarks.
© 2021 Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.