Fall 2026 Graduate Seminar
CS 395T: Hardware-Software Co-design for Cloud and Datacenters
Course Details
- Instructor: Jovan Stojkovic (jovan.stojkovic@utexas.edu)
- Schedule: Tuesday/Thursday 9:30-11:00AM
- Location: GDC 5.304
- Format: Lectures, paper reading & discussions, research-oriented term projects
- Office Hours: TBD
Prerequisite
Background in computer architecture and operating/distributed systems.
Course Format
This is a graduate-level seminar. Each class covers 2-3 research papers. Students are expected to read the assigned papers before class. Classes will consist of instructor- or student-led paper presentations followed by group discussion.
Course Description
Modern cloud infrastructure is defined by a fundamental tension: applications demand ever-lower latency, higher throughput, and greater efficiency, yet the end of Dennard scaling and the slowdown of Moore's Law mean hardware alone cannot deliver these gains. The path forward requires co-design, rethinking software and hardware together, each informed by the constraints and opportunities of the other.
This seminar traces that co-design process end to end. We begin by understanding what datacenters actually spend their time doing. We will profile warehouse-scale workloads to identify where cycles and energy go. With that empirical foundation, we will study two dominant cloud execution paradigms, microservices and serverless computing, examining how each has driven innovations in both software systems and hardware. We then turn to a recurring theme across all datacenter workloads: the "datacenter tax", pervasive overheads like serialization, compression, and RPCs, and the accelerator architectures designed to reclaim those lost cycles. Moving forward, as AI rapidly emerges as a defining class of datacenter workload, we examine the infrastructure built to serve recommendation and large language models, and how these workloads reshape system design from chip architecture to cluster scheduling. We will also look at using the ML itself as a tool for managing datacenter systems. Finally, we will conclude by studying the increasingly critical challenge of power and energy management at scale, the physical reality that bounds everything we can build.
Throughout, students will see how workload characterization motivates system design, how software abstractions shape hardware opportunities, and how physical constraints feedback into both.
Grading
| Paper Presentations | 20% |
| Class Participation | 5% |
| Paper Reviews | 15% |
| Midterm Project Report + Presentation | 20% |
| Final Project Report + Presentation | 40% |
Academic Integrity
The university provides a Canvas page with policies and resources relevant to all courses. You can refer to it as you navigate your time at UT. This course also has the following additional policies.
Use of Generative AI Tools
The use of generative AI tools (e.g., ChatGPT, Copilot, or similar systems) is not permitted for paper reviews and summaries in this course. This includes any use for drafting, paraphrasing, or grammar correction. Submissions must reflect your own original thinking and writing; minor grammatical errors are preferable to AI-assisted text.
For project reports, the use of generative AI tools is allowed only with full transparency. Students must clearly specify whether AI tools were used, which tools were used, and how they contributed (e.g., brainstorming, editing, code assistance). Failure to disclose such use will be treated as a violation of academic integrity policies.
A Notice of Academic Accommodations from Disability and Access (D&A)
If you are a student with a disability, or think you may have a disability, and need accommodations please contact Disability and Access (D&A). You may refer to D&A’s website for contact and more information: http://disability.utexas.edu/. If you are already registered with D&A, please deliver your accommodation letter to me as early as possible in the semester so we can discuss your approved accommodations.