Driving Superior watsonx Performance Through a Relationship with Intel

Co-Author: Edward Calvesbert, Vice President, Product Management, IBM watsonx Platform

IBM and Intel have had a long, robust relationship when it comes to Data and AI. As part of that relationship, IBM and Intel are committed to optimizing the IBM watsonx.data stack and delivering breakthrough performance through technological contributions to the Presto open-source community.

We have been working closely with Intel to optimize watsonx.data on the latest Intel processors. We are excited to announce that, based on testing conducted by Intel, the 5th gen Intel® Xeon® 8592+ processor can deliver up to 4.3X better query throughput from Presto C++ v0.286 and query optimizer on IBM watsonx.data than with Presto through AVX-512 optimizations.* This translates to better performance for our clients.

Edward Calvesbert, VP of Product Management for the watsonx platform, recently had the opportunity to connect with Intel to provide his thoughts on IBM watsonx, the strength of the IBM and Intel relationship, and our commitment to optimizing the performance of data workloads to enable AI everywhere. Let’s dive into his responses below.

On IBM watsonx.data – what is it?

“Watsonx.data is an open data lakehouse, a disruptive data management architecture for analytics workloads including data transformation and enrichment, business intelligence (BI), and, of course, artificial intelligence (AI). We have clients using watsonx.data in a variety of ways, for example:

We have several clients who recently lifted and shifted their data warehouse workloads from on-prem appliances to SaaS and are now facing rising costs and are looking to optimize their workloads, particularly their ingest and data transformation workloads, to save costs and also open up that data for new AI workloads.

We also have clients whose data lakes have become data swamps and are difficult to manage. They are modernizing to the Iceberg open table format, which provides ACID consistency and the ability to leverage multiple query engines and reduce the cost and risk of ETL.

We recently added a Milvus-based vector database to watsonx.data. This is helping clients quickly establish retrieval augmented generation or RAG applications across a variety of use cases like conversational access to enterprise knowledgebases ranging from technical and user documentation for customer support or HR and other corporate policies for employee self-help, to help drive productivity gains in a variety of front-office and back-office functions.”

On the importance of IBM watsonx.data for a CEO or CIO

“The open data lakehouse architecture is truly disruptive… Comprised of commodity COS, open table formats like Apache Iceberg, and a variety of query engines like Presto and Spark, a data lakehouse combines the flexibility and cost advantages of a data lake with the performance and functionality of a data warehouse.

Modernizing and unifying data architectures to a lakehouse can provide enterprises control over their data, stored in cost-effective storage, and accessible from anywhere and with ANY query engine that supports Iceberg.

By having multiple query engines accessing the same data at the same time, users get a choice of tooling, and enterprises can help enhance price performance, Spark for data transformation, Presto for interactive queries, and even a commercial data warehouse like Netezza or Db2 for high-performance dashboards, all while mitigating the cost and risk of ETL.”

On applying AI data to derive insights and business outcomes when so much enterprise data is still on-premises

“The data silos problem in the enterprise is pervasive, and you can’t solve it by creating a new data silo in the cloud… New insights are found in combining new data from various sources with existing enterprise data on customers, transactions, and assets. A lakehouse architecture augments existing repositories and can connect to data where it resides. It can also help unify and enrich data to make it easy to unlock those new insights.”

On IBM and Intel’s efforts to enable AI everywhere

“Watsonx is AI for business. IBM and Intel share similar guiding principles for serving our clients and the overall market. For watsonx, these are Open, Targeted, Trusted, and Empowering.

Open includes hybrid to run anywhere and on any infrastructure, and it also means open source, with a broad and diverse community of innovation and project governance.

Watsonx is targeted for business, which means we are helping customers solve a number of problems and enterprise considerations such as price performance, workload scheduling, optimization, and, of course, security and lifecycle management.

Trust is the license to operate in business; it takes a long time to build and can be lost in an instant. IBM and Intel have been earning our clients' trust for decades. In AI, trust is the process of selecting, customizing, deploying, and monitoring a model and structuring and tracking it to monitor adherence to corporate and regulatory policies. Use cases and applications that integrate AI must also be assessed for risk. Finally, watsonx is a platform that empowers organizations to customize, integrate, and co-locate their AI workloads with their enterprise data and applications to help drive real competitive advantage. In this rapidly evolving technology and vendor landscape, selecting the right partners that share your values and track record is an important consideration.”

On the results of the IBM plus Intel optimization work to run watsonx.data on Intel Xeon Scalable processors via open source contributions to the Presto query engine.

“For over a year, we have worked to optimize the performance of Presto 2.0, the next-generation of Presto being developed by Meta, IBM, Intel, and others that run Presto with Velox, an open-source C++ native acceleration library designed to be composable across compute engines. We’ve focused on the TPC-DS analytical workloads benchmark. This optimization happens at every layer in the stack, from the query optimizer to the query engine and the storage tier. Our results to date show massive improvements compared to traditional open-source Presto."

On IBM’s work with Intel® Gaudi® accelerators and why it’s important to watsonx

“AI workloads today are from a relatively small number of companies that are training foundation models and a much larger number of enterprises that are customizing those models with their data and integrating AI with their enterprise applications. This diversity of workloads across hybrid cloud environments and varying SLAs means that enterprises will need to decide where to run what in order to co-locate and scale AI. Our work with Intel Gaudi accelerators provides our joint clients the flexibility to optimize their training, tuning, and inferencing workloads to achieve optimal performance, which is critical for scaling AI across the enterprise.”

How can you get started with IBM watsonx.data?

Try watsonx.data yourself with a free trial.

*Based on Intel internal testing as of 04/15/24 on 1-node, 2x INTEL(R) XEON(R) PLATINUM 8592+, 64 cores, HT On, Turbo On, NUMA 2, Total Memory 2048GB (32x64GB DDR5 5600 MT/s [5600 MT/s]), BIOS 3B05.TEL4P1, microcode 0x21000161, 2x Ethernet Controller 10-Gigabit X540-AT2, 1x 1.7T INTEL SSDSC2KG019T8, 4x 3.5T INTEL SSDPF2KX038TZ, Red Hat Enterprise Linux 9.2 (Plow), 5.14.0-284.30.1.el9_2.x86_64 and a pre-release version of watsonx.data 2.0 (build 80) with query optimizer only applied to queries with performance benefits. Performance varies by use, configuration, and other factors. Learn more on the Performance Index site. 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. Your costs and results may vary. Intel technologies may require enabled hardware, software, or service activation.