Credits : 03
Teaching Segments : 01 - 06 (27 Jul – 13 Nov 2026)
Instructor : Abhijit Das (abhijit.das@cse.iith.ac.in)
Teaching Assistants (TAs) : TBD
Lecture Schedule : 3 hours a week, Q slot (Mon: 16:00, Thu: 14:30)
Lecture Venue : CS-LH-2, Dept. of CSE, IIT Hyderabad
Meeting Hours : By prior appointment (Tue: 15:00 – 16:00); meetings immediately after lectures may also be possible
Meeting Venue : CS-407, Dept. of CSE, IIT Hyderabad
Prerequisite : CS2323: Computer Architecture or equivalent
Moodle will be used for course management. Students are requested to self register here.
Foundations: Introduction to ML systems, ML workload characterisation, roofline model and arithmetic intensity, benchmarking and evaluation
Hardware Platforms: GPU architecture and memory hierarchy, ML accelerator design and dataflows
Memory and Interconnects: On-chip memory and data reuse for ML accelerators, interconnects for ML accelerators, distributed ML and collective communication
Design Optimisation: Design space exploration for ML accelerators, HW/SW co-optimisation for multi-model ML workloads
Advanced Topics: Hardware for LLM-based systems, emerging hardware trends
Guest Lectures: Experts from industry and academia
Machine Learning Systems by Vijay Janapa Reddy, The MIT Press, 2026 (Version 0.7.1)
Data Orchestration in Deep Learning Accelerators by Tushar Krishna, Hyoukjun Kwon, Angshuman Parashar, Michael Pellauer and Ananda Samajdar, Synthesis Lectures on Computer Architecture, 2020 (Edition 01)
Curated research papers from premier venues: ISCA, MICRO, HPCA, TC, TCAD, etc.
| Week | Date | Lecture | Topic | Materials | References | Remarks | |
| 1 | 27 Jul 2026 | 0 | Course Overview | CS_6123_L0.pdf |
Course project (CP) : 40%
Paper reading and presentation (PRP) : 15%
Mid-sem exam : 10-15%
End-sem exam : 25-30%
Class participation : 01-05%
A 1% penalty per additional late hour on the CP submission, up to a maximum penalty of 50%.
The use of GenAI tools is allowed and, in some contexts, encouraged for the CP and PRP, with proper attribution.
27-07-2026 : Course website launched!