Utkrisht Patesaria

Product Engineer

Bengaluru, Karnataka, India5 yrs 5 mos experience

Key Highlights

Expert in GPU performance modeling and optimization.
Proven track record in GNN training pipeline enhancements.
Recognized for outstanding customer service in cloud environments.

Stackforce AI infers this person is a Semiconductor and AI specialist with a strong focus on performance optimization.

Contact

Skills

Core Skills

Computer ArchitectureGpgpuOperating Systems

Other Skills

CC++C/C++CUDAComputer NetworkingDeep LearningGraph NetworksLinuxLinux System AdministrationSimulatorVirtualization

About

Utkrisht is a member of AMD’s pre-silicon team for MI SoC, where he focuses on developing performance models for the next generation of data-center GPUs. He recently earned his graduate degree from the Indian Institute of Science, Bangalore, under the supervision of Prof. Arkaprava Basu in the Computer Systems Lab. He primarily worked on the CPU-GPU memory subsystem interface, in turn optimizing the GNN training pipeline. Previously, he worked at Nutanix as an SRE, troubleshooting issues across the entire cloud stack.

Experience

Amd

Sr. Silicon Design Engineer

Aug 2025 – Present · 7 mos · Bengaluru, Karnataka, India · Hybrid

Part of AMD’s Data Center Group, working on the pre-silicon performance-modeling for the next generations of MI(Instinct) GPU SoCs.
Dissecting the entire GPU pipeline and identifying potential bottlenecks for various kernels, primarily targeting variants of GEMMs - coming with architectural level decisions (eg. cache modelling) to mitigate potential hotspots and improve end-end performance.
Improving performance projections of various LLM workloads during inference/training, on MI SoCs by breaking them into layer/operator level by co-relating their performance to specific kernels.

Computer ArchitectureGPGPU

Indian institute of science (iisc)

Graduate Researcher

Dec 2022 – Jul 2025 · 2 yrs 7 mos · Bengaluru, Karnataka, India

For his thesis, Utkrisht optimized a GNN training pipeline by leveraging CPU–GPU heterogeneity. He built a runtime system on top of the Deep Graph Library (DGL) and implemented the necessary optimizations in the C/C++ backend of PyTorch. His work also incorporated custom shared-memory libraries for both CPU and GPU and explored software-based GPU caching mechanisms to accelerate each stage of the pipeline. Earlier in the project, he investigated out-of-memory graph processing techniques on GPUs using unified memory, graph partitioning, and zero-copy (UVA) methods.

Computer ArchitectureOperating Systems

Nutanix

System Reliability Engineer - I

Jan 2020 – Apr 2022 · 2 yrs 3 mos · Pune, Maharashtra, India

++ Recognised with a spot award for customer obsession in Q4FY21
++ Awarded the POD start performer in Q4FY21 for filing the most number of JIRA tickets and Knowledge articles in 1 quarter.
++ Troubleshot 650+ issues ranging over hardware, networking, virtualization, and storage with a positive customer sentiment