Executive Summary
Artificial Intelligence (AI) applications are critical to businesses in boosting productivity, innovation, and delivering greater business value. Generative AI (GenAI) offers a range of benefits across industries and applications. However, it’s important to consider potential data challenges across data warehouses and data lakes, on-prem or in the cloud, which can slow adoption of AI and increase complexity and expense.
IBM watsonx.data, part of IBM’s Enterprise GenAI solutions, empowers business to leverage their data for greater insight and innovation in their business operations. Whether data is on-premises or in the cloud, watsonx.data offers a fast and flexible fit-for-purpose data store.
In addition to leveraging an open data lakehouse architecture and integrating IBM’s own intelligent query processing optimizations drawn from decades of delivering high-performance data solutions, IBM and Intel worked together to accelerate watsonx.data on Intel® processors. These optimizations have been proven to deliver higher performance on 4th Gen and 5th Gen Intel® Xeon® Scalable processors. Based on Intel 10 TB TPC-DS benchmarking, they take advantage of Intel® Accelerator Engines to boost watsonx.data query performance by up to 2.7X on 5th Gen Intel Xeon Scalable processors and 2.17X on 4th Gen Intel Xeon Scalable processors compared to 2nd Gen Intel Xeon Scalable processors.1 This level of performance delivers faster time to insight and enables quicker intelligent decision making. Faster queries also enable more work to be done on the hardware, enhancing the value of the solution while reducing the overall Total Cost of Ownership (TCO).
Data Challenges for AI
Enterprise strategies are maturing due to integration of AI tools into business operations and analytics. According to Gartner, more than 80 percent of enterprises will have used and adopted GenAI-enabled applications by 2026. 2
But AI feeds on data—lots of data—and AI is only as good as your data. Today, enterprises face real challenges with their data, due to the following:
- Data is in more locations: on-prem, cloud, applications, and silos, which create significant challenges; 82 percent of enterprises are inhibited by data silos. 3
- Data is in more formats: documents, images, and video. Such variety slows down the creation of quality data; 80 percent of time is spent on data cleaning, integration, and preparation. 4
- Data quality suffers because it ends up being stale and inconsistent; 82 percent of enterprises say data quality is a barrier on their data integration projects. 3
- Data curation and cleaning for AI and Business Intelligence may require different tools to act on the same data to achieve desired results.
Data lakehouse solutions have emerged to help address these challenges, but often these 1st-generation lakehouse designs do not meet the various diverse needs of AI and analytic workflows and workloads. For example, they are limited to a single query engine, which complicates supporting some types of workloads. Additionally, some solutions are available on cloud only; they do not support multi-cloud, hybrid-cloud, or on-premises deployments. 1st-generation solutions are limited in their ability to reduce cost and complexity.
Harness the power of AI and accelerate applications across your entire enterprise. Discover how AI powered by Intel Xeon can work for you.
IBM watsonx.data – A Fit-For-Purpose Data Store
IBM watsonx.data is a 2nd-generation open data lakehouse, delivering capabilities not part of 1st-generation solutions. watsonx.data is a powerful and flexible data store designed for a wide variety of enterprise Business Intelligence (BI), analytics, and AI workloads. Benefits of watsonx.data include the following:
- Reduced TCO – watsonx.data is built on an open source architecture with open source tools and technologies, allowing customers to cost-effectively scale their analytics and AI across all their data, wherever it’s located.
- Flexibility and Fit-for-Purpose Queries – watsonx.data supports multiple query engines, including Presto and Apache Spark, enabling a broader set of users to tailor data queries to their needs and workloads.
- Optimized Performance – watsonx.data is optimized for Intel® architecture and technologies, delivering exceptional query performance and price-performance.
IBM watsonx.data is part of the IBM watsonx platform. The platform is designed to enable enterprises to create and customize predictive machine learning and GenAI models with their data for deep understanding of their business operations and environment. With the platform, companies can store sensitive data in one interface and use it for leveraging AI and ML. IBM watsonx also includes governance capabilities to direct, manage, and monitor the AI activities of your organization.
Optimized for Analytics and GenAI Performance
Intel and IBM have a long history of deep collaboration to optimize IBM software on Intel processors and accelerators. That relationship continues with optimizations for watsonx. data to deliver the performance and reliability IBM customers expect and trust from IBM.
Intel has contributed optimizations for the native Presto query engine’s source code. But new Intel optimizations include enhancements on components within Presto.
Presto’s native worker component is built on a Java Virtual Machine layer. Presto C++ is a native implementation of that layer that takes advantage of Intel® Advanced Vector Extensions 512 (Intel® AVX-512). Integrating optimized Presto C++ significantly accelerates queries in Presto when run on Intel processors, including the 4th Gen Intel Xeon Scalable processors and 5th Gen Intel Xeon Scalable processors.
Additionally, IBM leveraged their experience with designing and optimizing database query engines for their products, such as Db2, to integrate an intelligent query optimizer in watsonx.data. The query optimizer dynamically creates a plan for each query to execute the fastest search of the data, accelerating results from large datasets.
A Data Lakehouse Built on Open Source
Using an open source architecture (Figure 1), watsonx. data offers the flexibility afforded by open source tools and easily scales to accommodate expanding data. Intel contributes to many open source projects, such as Presto, Presto C++, OpenJDK, and Apache Spark, enabling highly performant software. Intel’s optimizations take advantage of the many Intel® technologies and Intel Accelerator Engines built into Intel Xeon Scalable processors and Intel® Xeon® CPU Max Series processors.
Figure 1. watsonx.data is built on an open source architecture with Intel optimizations built in.
Multiple Query Engines for the Right Data in the Right Project
watsonx.data supports multiple analytics engines, including open source Apache Spark and Presto. Presto is an open-source, distributed SQL query engine, designed for analytics against data of any size. Presto is fast, reliable, and efficient at scale. Apache Spark is an open source, unified analytics engine for large-scale data processing and machine learning pipelines. With Spark, clients can run data engineering, data wrangling, data science, and machine learning workloads with data parallelism and fault tolerance. Both Presto and Spark are optimized for Intel architecture, taking advantage of Intel Accelerator Engines, such as Intel AVX-512.
Additionally, watsonx.data supports external analytics and query engines that are compatible with the Apache Iceberg open table format. IBM’s Db2 and Netezza are being enhanced to read from and write to the Iceberg format so they can then participate in the lakehouse architecture and share the same data.
Finally, watsonx.data can be deployed as containerized software on-premises and is available as SaaS on IBM Cloud and AWS, offering multiple deployment options according to enterprise needs.
Up to 2.7X Higher Analytics Throughput on Intel Technologies
Using two different analytics workloads—a Decision Support workload and IBM Big Data Insights (BDI)—Intel engineers benchmarked query throughput-per-node of IBM watsonx.data. Throughput was measured across three different hardware configurations to evaluate the gen-togen performance across three generations of Intel Xeon Scalable processors. These included 2nd Gen Intel Xeon Scalable processor (baseline), 4th Gen Intel Xeon Scalable processor, and 5th Gen Intel Xeon Salable processor. Configurations are described in the footnotes below.1
Benchmarks (Figure 2) reveal for Decision Support, that 5th Gen Intel Xeon Scalable processors deliver up to 2.7X higher throughput compared to 2nd Gen Intel Xeon Scalable processors. 4th Gen Intel Xeon Scalable processors provide up to 2.17X higher throughput.1 For IBM Big Data Insights, the 5th Gen Intel Xeon Scalable processor delivered up to 2.41X higher throughput, while 4th Gen Intel Xeon Scalable processor achieved 2.03X higher throughput.1
Additionally, the benchmarks also assessed the value of software optimizations to Presto C++ and impact on throughput of the query optimizer in watsonx.data (Figure 3). The open source Presto analytics engine provided a baseline measurement. Against this baseline, optimized Presto C++ with query optimizer runs queries over 4X faster on the two latest generations of Intel Xeon Scalable processors.1 The following charts show the results.
Figure 2. Decision support and IBM BDI throughput benchmark.1
Figure 3. Software optimizations benchmark.1
Conclusion
As enterprises increasingly leverage the power and benefits of GenAI to gain insight and enhance operations, the speed of queries and analysis from their data becomes paramount. IBM watsonx.data is designed to not only accelerate analytics on massive data lakehouse architectures, but also to deliver greater flexibility to use fit-for-purpose analytics engines for the right application.
Compute systems running 4th Gen Intel Xeon Scalable processors and 5th Gen Intel Xeon Scalable processors accelerate IBM watsonx.data queries and analytics over 2X faster than servers built on previous 2nd Gen Intel Xeon Scalable processor-based systems. Performance enhancements to watsonx.data comes not only from improvements in the latest generation of Intel Xeon Scalable processors, but also due to the following:
- Intel software optimizations in Presto C++
- IBM’s intelligent query optimizer in watsonx.data
The level of throughput improvements from the newest hardware and optimized software mean that enterprises can do more with less. IT departments can build compute solutions or cloud instance clusters with fewer nodes to accomplish more analysis and achieve new insights using IBM watsonx.
About Bhuma, Inc.
“One of the unique value propositions for IBM watsonx.data is the flexibility to deploy the lakehouse anywhere—IBM Cloud or customer’s private cloud This is particularly valuable for some of our large regulated and security conscious customers.
We are thrilled to have partnered with IBM and deployed watsonx.data solutions for AWS cloud, powered by the new Intel technology, to our financial customers to give the best performance, cost, flexibility—and total governance.” – Srini Gurrapu, Founder and CEO, Bhuma, Inc.
Bhuma is a strategic IBM watsonx ecosystem partner that helps organizations rapidly build real-time data apps, analytics, and GenAI agents purpose built for modern lakehouses such as watsonx and Presto.