On November 11, 2020, Intel officially released its first data center GPU, which is based on the Xe-LP microarchitecture and is designed for high-density, low-latency Android cloud gaming and streaming services. At the same time, it is announced that Intel’s oneAPI Gold toolkit will be officially delivered in December this year; Intel’s software stack introduces new features as part of its hardware-software co-design approach.
At first glance, the first half of the sentence seems to be a piece of news with a common sense error. How could Intel release a GPU? And once the GPU is completed, Intel will have a strong dominance in the entire XPU link, just like Qin unified the six countries, scolded Fang Qiu, and sent troops to major processor types. As the layout becomes more and more complete, it also begins to play an overall coordination effect.
Everything Intel has planned has to start with the release of the GPU.
What does Xe GPU mean?
The Xe GPU released this time is fundamentally different from the previous integrated GPU. This GPU is the first independent Display product for data centers. Featuring Intel’s most energy-efficient graphics architecture, the Intel Xe-LP microarchitecture, it features a low-power, discrete system-on-chip (SoC) design with a 128-bit pipeline and 8GB of dedicated onboard low-power DDR4 memory.
Intel server GPU
From the perspective of Xe architecture, Xe is a basic GPU architecture, a very flexible and highly scalable unified architecture. Previously, Intel was divided into four micro-architectures (Xe LP, Xe HP, Xe HPC, Xe HPG), which can be applied to almost all computing and graphics fields, including high-performance computing, deep learning and training, cloud services, multimedia Editing, workstations, gaming, thin and light notebooks, portable devices, and more.
For Intel, GPU has the advantages of efficient parallelism, high-density computing, and ultra-long graphics pipeline in high-performance technical computing. Transformation strategy.
Not only that, but combined with Intel’s own advantages, a synergistic effect is formed. This general release of Xe-LP GPUs can be combined with Intel Xeon Scalable processors, allowing service providers to independently expand graphics capacity to support more streams and subscriptions per system without changing the number of servers users, and at the same time achieve a lower total cost of ownership (TCO). By packing four Intel server GPUs in a H3C 3/4-length, full-height x16 PCIe 3.0 expansion card, it can support more than 100 concurrent users of Android cloud games in a typical dual-card system. This number can be scaled up to 160 concurrent users, the actual number depending on the specific game and server configuration2. Developers can take advantage of the common API in the current Media SDK, which will also be migrated to the oneAPI Video Processing Library (oneVPL) next year.
Currently, Intel is working with a number of software and service partners to bring Intel server GPUs to market, including Gamestream, Tencent and Ubitus.
In terms of specific applications, Fang Liang, deputy general manager of Tencent Cloud Gaming, said: “Intel is a very important partner for our Android cloud gaming solutions. Intel Xeon Scalable processors and Intel server GPUs have created a high-density, low-cost The solution of latency, low power consumption, and low TCO allows us to generate more than 100 game instances on each dual-card server, such as “Honor of Kings” and “Legendary Showdown”.”
In addition, Intel server GPUs based on the Xe-LP microarchitecture will ship this year, according to Intel. Together with the recently launched Intel Iris Xe MAX stand-alone Display, the GPU will further enhance the visual computing experience for users around the world as Intel’s Xe architecture products and software plans continue to evolve.
Looking at all the processor types Intel has, GPUs seem to be a belated bit in the data center space. When the outside world is constantly thinking about “a certain PU” to adapt to future computing, Intel has proposed the XPU strategy, overlooking the entire computing field from the perspective of God. The development of science and technology always needs someone to hit the wall first. The proposal of the XPU strategy seems too ambitious, but it is a very promising lead and attempt.
Intel’s “X” Mindset
Coincidentally, the Xe architecture, XPU, and intelligent X effect all hide an “X”, and the meaning behind it is “flexible” and “unified”. This concept seems to be deeply rooted in Intel’s bones, so much so that it is named after all. Lean on the “X”.
Looking at the overall background of the development of science and technology, X also represents chaos, the chaos in an era of data explosion, the types of data are different, the types of sensors are also very different, and the ways to obtain data are also very different. For Intel, it has mastered four different architectures: CPU, GPU, ASIC, and FPGA. Among them, the CPU is suitable for scalar operations, the GPU can be used for matrix operations or accelerators, and customized programmable hardware such as ASICs can be used as AI-specific accelerators. Spatial operations, i.e. computing with the fewest paths in the matrix, can be customized with FPGAs.
The “super-heterogeneous computing” proposed by Intel before is a typical integrated and unified thinking, that is, using multi-functional, multi-architecture chips to process and accelerate different computing loads, using packaging integration technology to package computing units in a SoC, and have a unified Computing Architecture for Heterogeneous Computing Software. Through heterogeneous arrangement and combination, customized requirements can be realized.
Taking the XPU as the center, continuing to go upstream is Intel’s “six technical pillars”, and going downstream is Intel’s “intelligent X effect”.
The former is process and packaging, architecture, memory and storage, interconnection, security, and software. Song Jiqiang of Intel China Research Institute said that computing needs are ubiquitous and increasingly diversified. CMOS scaling, 3D process technology, new architectures, new functions, etc. will continue to drive Moore’s Law forward, but any single factor can no longer be satisfied. Diversified future computing needs, and exponential innovation based on six technology pillars will be the driving force for Intel to enter the next 10 years and even the next 50 years.
The latter is both an industrial effect and an economic effect. When he first proposed the “Intelligent X Effect” at the media conference, Yang Xu, Vice President of Intel Corporation and President of China, said that the intelligence of all things brings exponential data explosion, and the development and application of intelligent technology will release the value of data. , so that industrial upgrading and economic upgrading move towards a new turning point. Specifically, it is the intelligent and cross-industry integration of value-added services, which penetrates into all walks of life.
From technology to ecological layout, the integrated and unified strategy represents Intel’s vision for the future, but the challenges behind it are very huge. Every technology branch will have challenges from the world’s top companies, which makes the outside world feel that Intel is tired of coping, so some The technical branch is slightly behind. But from a holistic point of view, all the technical layouts are the fuel for Intel to roll forward.
The charm of oneAPI
The XPU strategy described above covers all of Intel’s hardware layouts, and if that’s all, it’s messy for developers. The existence of oneAPI perfectly solves the problem of hardware interconnection of different architectures on the software side, and also conforms to Intel’s idea of unified integration. oneAPI will define and lead a programming era in which artificial intelligence is increasingly converged, heterogeneous and multi-architecture.
oneAPI is a unified, simplified programming model designed to simplify the development process across multiple architectures (e.g. CPU, GPU, FPGA, accelerators). Provides a common, open programming experience that gives developers the freedom to choose an architecture without compromising performance and greatly reduces the complexity of using different codebases, programming languages, programming tools, and workflows. Compared to today’s closed programming environments based on a single vendor, oneAPI provides developers with a highly competitive and advanced alternative to help them build a seamless bridges to create more rich applications for the multi-architecture world of the future.
The Intel oneAPI Gold toolkit will be available for free on-premises and on Intel DevCloud in December, along with a commercial version that includes worldwide support from Intel technical consulting engineers, Intel announced. Intel is also migrating the Intel Parallel Studio XE and Intel System Studio toolkits to the oneAPI product.
Additionally, the Intel DevCloud platform, which allows developers to test code and workloads on a variety of Intel architectures, has now added new Intel? Torch? Xe GPU hardware. Intel Iris Xe MAX graphics cards are also now publicly accessible; at the same time, Intel Xe-HP has been made available to select developers.
oneAPI has been supported by the industry. Recently, Microsoft Azure and Google’s TensorFlow have announced support for oneAPI; many leading research institutions, companies and universities also support oneAPI.
If XPU is a reflection of Intel’s own extensive and comprehensive strength, oneAPI is an ecological gang established by Intel based on technology. Intel’s corporate strategy is based on unification and integration. Just like Qin unified the six countries, after having its own hardware foundation, it began to unify the soft power of words, currency, and weights and measures. Judging from the acquisitions of competitors, it seems that they also want to follow Intel’s strategy. At present, no one can draw a final conclusion on the form of computing in the future. The best strategy is to focus on a certain general direction, take all computing forms, and become a giant wheel rolling forward until it hits a wall and smashes it.
The Links: MIG200J6CMB1W TPS61230ARNSR