GPU Compute Architect - Leading a team creating a novel end to end (GPU/CPU/interconnect) system simulation methodology including simulating concurrency between GPU streams and other components and projecting overall AI/HPC application performance. - Innovated and developed a new framework to create system level projections and identified platform bottlenecks - Proved the methodology on multiple classes of real world applications and using results for pathfinding HW features - Developed equations for high level GPU architectural model for thread dispatch, latencies, and cache features. - Driving the productization of this concept including managing collaboration between various architecture, API and tool teams, defining technical interfaces and models, obtaining funding, planningtimelines/resources and execution, hiring engineers to build team from ground-up. https://www.ixpug.org/images/docs/Annual_Conference_2023/IXPUG_Annual_Conference_2023_paper_1056.pdf- Enabling several large-scale customer contracts for multi-node GPU based systems by providing performance projections and evaluating different system configuration options.- Analyzing pre and post silicon performance for relevant AI and HPC workloads including compute and communication overlap, power analysis and optimization.- Performance and architecture analysis of the GPU with a focus on Compute, specially GPU scheduling and concurrency features.- Defining hardware features and pathfinding in future GPU architectures based on findings from workload performance analysis.- Focused on HPC/AI. Previously focused on Ray Tracing/VR. Workload analysis, root causing performance bottlenecks, and iteratively improving SW design as well as HW features.

Graphics/Compute API Capture Tool - Lead engineer and primary customer contact on multi-OS (Windows/OS X) multi-API on OpenGL/DX11/OpenCL 1.2 capture tool. Also developed an OpenGL ES on Android prototype.- Scheduled and coordinated customer requests for enabling a wide variety of workloads and influenced the tool’s overall direction through this effort.- Developed different hooking mechanisms to capture complex applications on multiple platforms. Implemented features such as logging formats, capturing unique kernels in a kernel range (OpenCL), and OpenGL/OpenCL resource sharing to enable multi-API applications.- Enabled testing of different workloads for OpenCL which increased test coverage by over 3x. Automated the generation of the user manual as well as parts of the release process and incorporated these changes into the Python-based infrastructure.- Received a Division Recognition Award in Q3 2015 for “5x Perf Gain on iGfx for DaVinci Resolve & a “Post-Production” Software Win for IA and Gen” . Video from Intel marketing below:

Cedar Trail and Clover Trail (Atom Z2760) DX9/10 driver development - Fixed several blocking bugs gating production drivers as well as issues that improved WHQL scores, including issues with GDI Blits, texture size calculations and compressed textures.- Optimized performance of the DX9/10 device drivers by analyzing and replacing shaders to study application and benchmark speeds.Developed and improved tracing tools to analyze and optimize apps that used Windows 8 metro-style tile rendering. Also enabled these traces to be part of CI which significantly improved the driver quality.

Larrabee IA-based many core architecture drivers and software-based graphics rendering pipeline - Implemented and tested features such as texture sampling from the vertex shaders, fixed issues like corner cases in the blit copy code in the OpenGL and DirectX device drivers as well as the software pipeline for Intel’s high-throughput many-core architecture.- Enabled workloads and benchmarks by mapping artifacts to different parts of the pipeline, such as the geometry shader, stream-out and contributed significantly to developing a highly optimized WHQL compliant software graphics pipeline.

Implemented major features in the OpenGL driver for Intel’s high-throughput many-core architecture.

ABAP/4 Programming: Development of Reports, SAP Scripts, Dialog Programs, User Exits, Interfaces.Worked to develop and upgrade SAP systems for clients like Rio Tinto, Daimler Chrysler and PSEG.