Built-in Accelerators and Why You Should Use Them

By Anita Tsuchiya

Intel-sponsored

Multiple times a day, you make strategic business decisions based on what your data tells you about your , current market trends, and future projections. And you are not alone. The demand for data-driven insights has reached unprecedented levels in businesses of all sizes. To meet these needs, your organization needs high-performance servers and cloud-computing platforms capable of handling workloads associated with artificial intelligence (AI), high-performance computing (HPC), secure web servers, data analytics, and network storage. At the same time, you must strike a balance between achieving optimal performance, maintaining robust security, and optimizing your total cost of ownership (TCO). This paradox underscores the need to find innovative solutions that can answer both imperatives in the complex landscape of today’s IT infrastructures.

In this Intel-sponsored article, we suggest there are two key steps you can follow to achieve the right balance between performance and expense: choose the right hardware for your needs and match it to the right workloads. General-purpose hardware can serve you well for certain workloads, but for modern ones such as AI and predictive analytics, you’ll get better performance from processors built to handle those demanding tasks.

How to Use Intel® Accelerator Engines

4th Gen Intel® Xeon® Scalable processors include built-in Intel® Accelerator Engines, which can not only boost server performance but also help lower TCO. Let’s look at how these built-in accelerators can deliver superior overall performance, performance per core, and performance per watt for some of today’s most popular workloads.

Accelerate AI

Processor: 4th Gen Intel Xeon Scalable processor Built-in accelerator: Intel® Advanced Matrix Extensions (Intel® AMX) Example workloads: Natural language processing (NLP), recommender system, and image classification Results:

• Up to 10x higher training and inference performance, compared to 3rd Gen Intel Xeon Scalable processors.[i]
• Up to 4x higher inference performance, compared to 4th Generation AMD EPYC™ processors using the same number of cores.[ii]

Accelerate HPC

Processor: 4th Gen Intel Xeon Scalable processor Built-in accelerator: Intel AMX Example workloads: Scientific simulations; financial analytics; 3D modeling and analysis; and image, audio, and video processing Results:

• Up to 9x higher performance per watt, compared to 3rd Gen Intel Xeon Scalable processors.[iii]
• Up to 9x higher performance per watt, compared to non-accelerated workloads.[iv]

Accelerate Data Compression and Encryption

Processor: 4th Gen Intel Xeon Scalable processor Built-in accelerator: Intel® QuickAssist Technology (Intel® QAT) Example workloads: Web server handshakes, data compression/decompression, and cryptography Results:

• Up to 47% fewer cores used to support the same number of NGINX® key handshake connections per second, compared to 3rd Gen Intel Xeon Scalable processors.[v]
• Up to 2x higher compression throughput while using 95% fewer cores, compared to 3rd Gen Intel Xeon Scalable processors.[vi]

Accelerate Data Analytics

Processor: 4th Gen Intel Xeon Scalable processor

Built-in accelerator: Intel® In-Memory Analytics Accelerator (Intel® IAA)

Example workloads: Database, database management system (DBMS), online analytical processing (OLAP), and online transaction processing (OLTP)

Results:

• Up to 3x higher RocksDB performance, compared to 3rd Gen Intel Xeon Scalable processors.[vii]
• Up to 52% lower TCO, compared to 3rd Gen Intel Xeon Scalable processors.[viii]

Accelerate Data Transformation

Processor: 4th Gen Intel Xeon Scalable processor

Built-in accelerator: Intel® Data Streaming Accelerator (Intel® DSA)

Example workloads: Network storage, software-defined networking (SDN), video streaming, and content delivery network (CDN)

Results:

• Up to 1.95x higher virtual switch (vSwitch) bi-directional throughput, compared to a software-only vSwitch solution.[ix]
• Up to 1.6x higher input/output operations per second (IOPS) and 37% lower sequential-read latency, compared to 3rd Gen Intel Xeon Scalable processors.[x]

Learn More

Explore the Intel Accelerator Engines portfolio.

Read about other use cases that can benefit from built-in accelerators:

• Accelerate AI and HPC using Intel AMX.
• Accelerate HPC, storage, analytics, network, and security using Intel QAT.
• Accelerate analytics using Intel IAA.
• Accelerate analytics, storage, and networking using Intel DSA.

[i] See [A16] and [A17] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[ii] See [A222] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[iii] See [E1] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[iv] See [E2] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[v] See [N15] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[vi] See [N16] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[vii] See [D1] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[viii] See [E8] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[ix] See [W8] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.

[x] See [N18] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.