Tim Kaldewey


Performance Architect, IBM Watson Group

51 Astor Place
New York, NY 10003
tkaldew(at)us.ibm.com (corporate)
tim(at)kaldewey.com (personal)

My resume and CV

About Me

My interests revolve around performance and efficiency of AI workloads, from a parallel algorithms and architecture perspective. While a lot of my work these days happens behind the scenes of Watson, some achivements have made it into the public realm. For example, I led the technical team for a CTO initiative to deliver a 40x efficiency improvement for Watson's question answering capability. I also led the efforts to leverage hardware acceleration for Watson offerings, which we first presented at at Supercomputing 2015 .

In 2014 I joined the IBM Watson team as their performance architect, leading a team of engineers focused on optimizing efficiency of Watson offerings across the hardware/software stack. I started my IBM career in 2010 when I joined the Database Technologies Group at IBM Almaden Research. During that time I had the opportunity to "play" with several emerging parallel architectures and evaluate their use for data-intensive applications. The outcomes were a thorough understanding of parallel architectures and their application to large scale software systems. Since 2011 I am also an Adjunct Professor in the Computer and Information Science department at the University of Pennsylvania where I teach lectures on GPU and systems programming. Before my time at IBM I was a member of Oracle's Special Projects Team and prior to getting my PhD I worked at SAP, Lufthansa, and Software AG.

I received my Ph.D. in Computer Science from the Baskin School of Engineering at the University of California Santa Cruz. My advisor was Prof. Scott Brandt and my thesis is on "Predictable High-Performance Data Management - Leveraging System Resource Characteristics to Improve Performance and Predictability". I did my undergraduate work in Business Informatics (Wirtschaftsinformatik) at Technische Universität Darmstadt in Germany. I received my Master's degree in Network Engineering from Eurecom, a research institute funded by Telecom Paris in France and EPFL in Switzerland.

Selected Publications

Michael Gschwind, Tim Kaldewey, David K. Tam.  "Optimizing Efficiency of Deep Learning Through Accelerator Virtualization."   IBM Journal of Research and Development   under review.

Tim Kaldewey, Guy Lohman, Rene Mueller, Peter Volk.  "GPU Join Processing Revisited."   Eighth International Workshop on Data Management on New Hardware
(DaMoN '12).
  Scottsdale, AZ, 2012. 

Tim Kaldewey, Sandeep Tata, Eugene Shekita.  "Clydesdale: Structured Data Processing on MapReduce."   15th International Conference on Extending Database Technology (EDBT '12).   Berlin, GERMANY, 2012. 

Tim Kaldewey, Andrea Di Blas.   "Large Scale GPU Search."   GPU Computing Gems - Jade Edition.  Chapter 1. Morgan Kaufmann Publishers, Waltham, MA, 2011.

Changkyu Kim, Jatin Chhugani, Nadathur Satish, Eric Sedlar, Anthony Nguyen, Tim Kaldewey, Victor Lee, Scott Brandt, Pradeep Dubey.   "FAST: Fast Architecture Sensitive Tree Search on Modern CPUs and GPUs."   2010 ACM SIGMOD/PODS Conference (SIGMOD '10).   Indianapolis, IN, 2010.
Best Paper Award

Tim Kaldewey.   "Programming Video Cards for Database Applications."   USENIX ;login.   Volume 34, Issue 4, 2009.

Tim Kaldewey, Theodore Wong, Richard Golding, Anna Povzner, Scott Brandt.   “Virtualizing Disk Performance.”   14th IEEE Real-Time and Embedded Technology Applications Symposium (RTAS '08).   Saint Louis, MO, 2008.
Best Paper Award

All Publications