Currently, the digital transformation of the industry has entered a deep-water zone, with the explosive penetration of GenAI technology driving the demand for computing power across various sectors from &#34;linear growth&#34; to &#34;exponential leap&#34;—from real-time interaction in gaming AI and inference operations of large models to handling sudden traffic surges in e-commerce flash sales and long-term storage of massive business data. Enterprises have unprecedented requirements for computing power in terms of &#34;magnitude,&#34; &#34;performance,&#34; and &#34;energy efficiency.&#34;
However, behind the surge in demand, the industry generally faces dual challenges: on one hand, the supply of computing power is difficult to precisely match the scenarios, leading either to resource idle waste due to the pursuit of extreme performance or to a dilemma of insufficient computing power due to a focus on cost control; on the other hand, the balance of energy efficiency under green data center policies, as well as the secure storage and low-latency access of PB-level data, have also become &#34;roadblocks&#34; for enterprises advancing digital intelligence. In the face of these common challenges, how should cloud vendors and hardware manufacturers collaborate to innovate and break through?
<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null"/> https://static001.geekbang.org/infoq/2a/2a3599aa0898e0543ae6c72c3387a9b2.webp<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null"/>
This issue of &#34;Cloud Evolution Theory&#34; invites three technical experts from Tencent Cloud and Intel—Tencent Cloud Senior Product Manager Xia Shiyue, Tencent Cloud Senior Product Manager Chen Hao, and Intel Chief Software Engineer Xie Yi—to conduct an in-depth discussion on core topics such as industry trend changes, cloud server product layout, cloud-native database innovation, and software-hardware collaborative practices, providing references for the digital implementation of enterprises.
The following is the edited and organized essence of the content.
<h2>Differentiation of Computing Power Demand: The Inevitable Logic of &#34;Dual-Core Architecture&#34;</h2>
As enterprises move to the cloud from &#34;scaled migration&#34; to &#34;scenario-based deep cultivation,&#34; the differentiated characteristics of computing power demand have become increasingly evident, and the traditional &#34;one-size-fits-all&#34; model of general computing power supply is becoming unsustainable. This has also become the core entry point for hardware manufacturers and cloud service providers to work together.
From the perspective of business scenarios, the current demand for computing power has formed two distinct camps:
One type is &#34;AI-driven high-performance demand,&#34; typified by gaming AI robots—previously requiring only high clock speeds to support game operation, now needing to simultaneously meet the real-time interaction of &#34;can listen and speak,&#34; posing dual high demands on AI computing power and single-core performance; another example is vector databases, which require large memory bandwidth and fast data processing capabilities to support high-frequency retrieval for large model inference. The core demand of such scenarios is &#34;performance first,&#34; with a willingness to pay reasonable costs for an extreme experience.
The other type is &#34;steady-state high cost-performance demand,&#34; covering traditional businesses such as video encoding and decoding, web services, and lightweight microservices. They do not require top-tier computing power but pursue long-term stable operation and controllable costs, especially under the trend of green data centers, where &#34;performance per watt&#34; becomes a key decision-making indicator. Additionally, the demand for elasticity in handling sudden traffic and the need to reduce costs for storing PB-level cold data further complicate the supply of computing power
Tencent Cloud Senior Product Manager Xia Shiyue mentioned in a live broadcast: &#34;Single hardware upgrades or software optimizations can no longer cope with this differentiation; only precise hardware adaptation and deep software collaboration can truly unleash computing power potential.&#34;
Intel has keenly captured this trend and innovatively launched a dual product line of &#34;Performance Cores (Granite Rapids, abbreviated as P-core)&#34; and &#34;Efficiency Cores (Sierra Forest, abbreviated as E-core)&#34; on the Intel Xeon 6 processor platform. The Performance Cores focus on &#34;high single-core performance, large cache, and Intel Advanced Matrix Extensions (AMX) instruction set,&#34; targeting compute-intensive scenarios; the Efficiency Cores emphasize &#34;multi-core, high performance per watt, and low power consumption,&#34; suitable for high-density steady-state scenarios. This &#34;on-demand core selection&#34; hardware design not only provides cloud service providers with a computing power foundation that precisely matches scenarios but also fundamentally resolves the industry contradictions of &#34;performance surplus&#34; and &#34;cost waste,&#34; laying the groundwork for subsequent cloud product innovations.
<h2>Dual-Core Computing Power Implementation: Technical Breakthroughs of Tencent Cloud CVM and TDSQL-C</h2>
The dual-core architecture of the Intel Xeon 6 processor ultimately needs to be transformed into perceivable value for enterprises through cloud service products. Tencent Cloud, centered on the ninth-generation CVM cloud server (computing power foundation) and TDSQL-C cloud-native database (data engine), deeply integrates hardware advantages with software innovations, creating a full-link solution covering &#34;computing power supply - data processing,&#34; becoming a typical practice for dual-core technology implementation.
<h2>Ninth-Generation CVM Cloud Server: Dual-Track Computing Power Adaptation for All Scenario Needs</h2>
As the core carrier of cloud services, Tencent Cloud's ninth-generation CVM relies on the Intel Xeon 6 processor to construct a dual-track system that runs &#34;extreme performance&#34; and &#34;inclusive computing power&#34; in parallel, achieving precise coverage of different scenarios.
On the extreme performance side, CVM has launched the S9e/S9Pro models, equipped with Intel Xeon 6 Performance Core (P-core) processors — this processor can have up to 128 performance cores in a single socket, with all-core turbo frequency reaching 3.6GHz, memory frequency increased to 6400MT/s (up to 8800MT/s with MRDIMM memory), L3 cache up to 504MB, and integrates the AMX matrix computation instruction set, with AI computing power improved by 8-16 times compared to AVX512.
To fully unleash hardware performance, Tencent Cloud has also equipped a self-developed 2×200G intelligent network card (single machine throughput capability reaching 67.5 million PPS), a new Star Sea server (overall stability improved by 40%, heat dissipation efficiency increased by 43%, memory ECC verification coverage rate 100%), ultimately achieving a 18% performance improvement in database scenarios, a 20% performance improvement in web applications, and a 15% performance improvement in online games, becoming one of the preferred solutions for high-performance scenarios such as game AI and image rendering
On the inclusive computing power front, CVM globally debuts with Intel's Xeon 6 efficiency core (E-core) processor — with a maximum of 288 physical cores (without hyper-threading design) in a single socket, offering more cores at the same power consumption compared to traditional processors, and achieving a 3-fold performance improvement per watt over the second generation Intel Xeon processors, while directly increasing rack density by 6 times. With the media processing capability unchanged, the E-core model helps users reduce overall costs by more than 30%.
Xia Shiyue introduced that a domestic technology company deployed web services using &#34;S9 cloud servers equipped with E-core + TencentOS Server 3.3 operating system,&#34; which not only smoothly handled the peak computing power during the school season at the beginning of September but also achieved a cost reduction of over 20%, becoming a cost-effective choice for traditional stable businesses.
In addition, to match the elastic demand for computing power, Tencent Cloud has also launched a combination billing plan of &#34;pay-as-you-go + annual/monthly subscription + bidding instances + savings plans + reserved instances&#34;: for burst traffic scenarios, pay-as-you-go or bidding instances can be selected, while long-term stable businesses can choose annual/monthly subscriptions or reserved instances, and periodic fluctuation businesses can balance costs and elasticity through savings plans, allowing enterprises to &#34;pay for what they use.&#34; This model further amplifies the cost advantages of the Intel® Xeon® 6 processors.
<h2>TDSQL-C Database: Hardware-Software Collaboration Drives Performance Revolution</h2>
If CVM is the &#34;computing power supply base,&#34; TDSQL-C is the &#34;data processing engine.&#34; As Tencent Cloud's self-developed cloud-native database, TDSQL-C leverages the hardware capabilities of Intel's Xeon 6 processors to address the pain points of traditional databases, such as &#34;low performance ceiling, slow fault recovery, and weak scalability,&#34; through architectural innovation, acceleration engine adaptation, and Serverless elastic design.
At the architectural level, TDSQL-C adopts a &#34;log as database&#34; design with a separation of computing and storage, completely decoupling them: The computing layer is equipped with Intel Xeon 6 performance core processors, utilizing their high single-core performance and multi-core advantages, combined with full-link remote direct memory access (RDMA) high-speed networks (reducing communication latency by 80%), responsible for core business tasks such as SQL execution and transaction processing; the storage layer builds a secondary cache based on persistent memory (PMEM), caching warm data closer to the computing end, achieving over 3 times performance improvement in I/O intensive scenarios, while supporting PB-level data capacity with triple-replica massive storage, achieving data availability of 99.9999999% (9 nines), completely solving the scalability issues of traditional databases.
In terms of performance acceleration, TDSQL-C deeply integrates Intel's QuickAssist Technology (QAT) hardware acceleration engine, specifically optimizing for cold data compression scenarios. Tencent Cloud Senior Product Manager Chen Hao stated that compared to pure software compression solutions, the QAT engine reduces storage space by 60%, lowers compression latency by 50%, and increases data throughput by 100%, while freeing up CPU resources for core business processing
Intel's Chief Software Engineer, Xie Yi, explained during a live broadcast: &#34;The 'compute-to-storage' logic of QAT aligns closely with the cold data processing needs of TDSQL-C, as it can solve the storage cost issues of massive data through compression, while also leveraging hardware acceleration to avoid latency losses during data retrieval, achieving a win-win in 'cost and efficiency.'&#34;
In addition, TDSQL-C utilizes the large cache advantage of Intel Xeon 6 processors to achieve tiered storage of hot, cold, and warm data, further enhancing data access efficiency, ultimately achieving a processing capacity of over one million QPS, with performance in large table scenarios improved threefold compared to traditional cloud databases.
In terms of elastic design, TDSQL-C's Serverless architecture has also innovated based on the resource scheduling capabilities of Intel Xeon 6 processors. For instance, a new &#34;recovery perceiver&#34; has been added to the access layer to ensure that user connections remain uninterrupted during instance pause/resume; the management layer monitors hardware resources and business loads in real-time, dynamically adjusting CPU and memory allocation, supporting cross-machine elasticity and cluster-level expansion (up to 15 RO nodes can be expanded to handle sudden traffic).
Unlike traditional elastic solutions, TDSQL-C adopts a &#34;real-time resource reservation&#34; model, reserving resource space of 1-10 CCU (Compute Resource Units) in advance for users, achieving &#34;instant full load&#34; through dynamic adjustment of the buffer pool, thus avoiding the scaling delay issues of traditional solutions. This elastic capability, combined with the cost advantages of Intel Xeon 6's efficient cores, can help users reduce costs by 60% in tidal flow scenarios, while also supporting a hybrid deployment of &#34;fixed specification master nodes + Serverless slave nodes,&#34; balancing business stability and cost optimization.
<h2>From Collaborative Innovation to Future Evolution: Continuous Deepening of Computing Power Value</h2>
The collaboration between Intel® Xeon® 6 processors and Tencent Cloud is not merely a simple &#34;hardware supply + software adaptation,&#34; but a full-link collaboration of &#34;demand - technology - product - implementation.&#34; This collaboration not only addresses the current computing power pain points in the industry but also provides long-term support for enterprise digitization through a clear future evolution direction. From the cases that have been implemented, the cooperation has already created tangible value for various industries.
In terms of cost reduction and efficiency enhancement, the &#34;efficient cores + hardware acceleration&#34; capability of Intel Xeon 6 processors has been transformed into quantifiable cost optimization through Tencent Cloud products—for example: a 30% cost reduction brought by the E-Core model S9, and a 60% storage compression of TDSQL-C; in terms of scenario adaptation, the dual-core architecture allows Tencent Cloud to customize solutions for different industries: internet companies use S9e/S9Pro to handle high concurrency, traditional manufacturing enterprises use S9 to control steady-state costs, and financial institutions leverage TDSQL-C's QAT acceleration to balance data security and processing efficiency, truly achieving &#34;on-demand matching and precise empowerment.&#34;
More importantly, this cooperation model has formed a virtuous cycle of &#34;demand-driven technology iteration&#34; — Tencent Cloud provides hardware optimization suggestions to Intel based on feedback from enterprise customers (such as the adaptation requirements of the QAT engine for database compression); Intel then iterates processor functions based on these demands, which are validated through Tencent Cloud products, ultimately driving the improvement of computing efficiency across the industry.
As trends such as AI large models, green data centers, and industrial internet deepen, both parties have clarified the next generation of cooperation paths, continuously amplifying the value of the dual-core architecture. On the hardware side, Intel will launch the Intel Xeon 6 Plus processor (Clearwater Forest), maintaining the energy efficiency core advantage while increasing memory bandwidth to 8000MT/s, adopting 18A advanced process technology, and adding features such as Intel® Software Guard Extensions (SGX) and SM3/SM4 encryption acceleration, further enhancing &#34;high performance + high security + low carbon&#34; capabilities; on the software side, Tencent Cloud will simultaneously iterate CVM and TDSQL-C: CVM will adapt to VRDMA networks to improve cross-node communication efficiency, while launching a new generation of CVM models in conjunction with Intel's next-generation chips; TDSQL-C will collaborate with Intel to continuously explore more value of databases in AI scenarios.
<h2>Conclusion</h2>
If the core of digital transformation is computing power transformation, then the key to computing power transformation lies in &#34;precisely matching business needs&#34; and &#34;efficiently coordinating industrial resources.&#34; The Intel Xeon 6 processor, with its &#34;performance core + energy efficiency core&#34; architecture breakthrough, provides the hardware foundation to meet diverse needs; Tencent Cloud, through software innovations in CVM and TDSQL-C, transforms hardware capabilities into perceivable values of &#34;performance improvement, cost reduction, and enhanced flexibility&#34; for enterprises.
This deep collaboration between &#34;hardware manufacturers + cloud service providers&#34; not only resolves the industry's dilemmas of &#34;performance vs. cost&#34; and &#34;flexibility vs. stability,&#34; but also proves that only by rooting technological innovation in actual business needs can the true value of computing power be unleashed. For enterprises, choosing Tencent Cloud products based on Intel Xeon 6 processors means opting for a computing power solution that is &#34;more business-savvy, more cost-effective, and more sustainable&#34;; for the industry, this collaborative model will upgrade computing power supply from &#34;scale coverage&#34; to &#34;high-quality adaptation.&#34; In the era of intelligent computing power, the &#34;dual-core advancement&#34; of Intel and Tencent Cloud will undoubtedly continue to drive technological innovation, injecting inexhaustible momentum into the digital transformation of various industries.
<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null"/> https://static001.geekbang.org/infoq/c2/c2c955283b1a208d9e557afa7e2e4682.webp<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/article/null"/>
Click to watch the full video～

INTC

The digital transformation of industries and the explosion of GenAI technology have driven the demand for computing power from linear growth to exponential leaps. Enterprises face challenges such as the difficulty of accurately matching computing power supply with scenarios and balancing energy efficiency. Cloud vendors and hardware manufacturers need to collaborate on innovation; precise hardware adaptation and deep software collaboration are essential to unleash the potential of computing power

- The article discusses the challenges and innovations in the industry regarding computing power, driven by the increasing demand for AI and cloud services.  
- It highlights the emergence of a "dual-core architecture" with Intel's new processors tailored to meet diverse performance needs, enabling companies to optimize costs and efficiency.  
- Collaborative innovations by Tencent Cloud and Intel provide solutions that align computing power with business requirements, supporting businesses in their digital transformation journey.

Goodbye to "one-size-fits-all": How does "demand differentiation" reconstruct computing power supply?