Subvolts take Dataflow computing offers a new path for faster, more efficient AI inference, especially for LLMs. Oct 14, 2025 - Developers
- Learn about novel hardware architectures that can optimize AI model inference and agentic workflows.
- Business
- Understand how new hardware designs can lead to more efficient and cost-effective AI deployments for LLMs.
This is a solid interview with a leading researcher discussing innovative hardware architectures for AI inference.
Dataflow Computing for AI Inference with Kunle Olukotun - #751
Audio streams from Sam Charrington. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Sam Charrington interviews Kunle Olukotun about reconfigurable dataflow architectures for AI inference. Olukotun, a Stanford professor and Sambanova Systems co-founder, explains how these systems dynamically match AI model dataflow, improving LLM inference by reducing memory bottlenecks. The discussion covers efficient multi-model serving, agentic workflows, and future dynamic architectures.
This episode discusses a new way to build computers specifically for AI. Instead of following a fixed set of instructions, these computers change their structure to perfectly match the AI's task, making AI run much faster and more efficiently. This is important for handling complex AI like chatbots and agents.
This episode explores a novel approach to AI inference hardware, moving beyond traditional CPU/GPU designs. Listeners will learn how dataflow architectures can overcome memory bandwidth limitations critical for large language models and agentic systems. It is for researchers, hardware engineers, and AI developers interested in optimizing inference performance and exploring next-generation AI hardware. This work is relevant as frontier labs like Google Gemini and OpenAI push the boundaries of model size and complexity, requiring more efficient inference solutions.
- HostSam Charrington
- GuestKunle Olukotunprofessor of electrical engineering and computer science, Stanford University
- GuestKunle Olukotunco-founder and chief technologist, Sambanova Systems
- Guest: Kunle Olukotun
- Title: Dataflow Computing for AI Inference with Kunle Olukotun - #751
- Show: The TWIML AI Podcast
- Length: 58 minutes
- Architecture Focus: Reconfigurable dataflow architectures for AI inference
- Benefits: Reduces memory bandwidth bottlenecks, improves LLM inference performance, enables efficient multi-model serving and agentic workflows
In this episode, we're joined by Kunle Olukotun, professor of electrical engineering and computer science at Stanford University and co-founder and chief technologist at Sambanova Systems, to discuss reconfigurable dataflow architectures for AI inference. Kunle explains the core idea of building computers that are dynamically configured to match the dataflow graph of an AI model, moving beyond the traditional instruction-fetch paradigm of CPUs and GPUs. We explore how this architecture is well-suited for LLM inference, reducing memory bandwidth bottlenecks and improving performance. Kunle reviews how this system also enables efficient multi-model serving and agentic workflows through its large, tiered memory and fast model-switching capabilities. Finally, we discuss his research into future dynamic reconfigurable architectures, and the use of AI agents to build compilers for new hardware.
The complete show notes for this episode can be found at https://twimlai.com/go/751.
- dataflow graph
- Kunle Olukotun
- a diagram showing how data moves through a computer program or AI model
- AI inference
- Kunle Olukotun
- using a trained AI model to make predictions or generate outputs
- LLM
- Kunle Olukotun
- Large Language Model, like those powering chatbots
- agentic workflows
- Kunle Olukotun
- AI systems that can perform tasks autonomously and make decisions
53 min1 h 30Why Jev Is Changing How We Build With AI with Diogo Almeida - #779
WithDiogo Almeida
1 h 081 h 05From Voice Agents to AI Avatars with Alexander Smola - #777
WithAlexander Smola
Same guests, other shows
2 h 07AI:AM Highlights: Exploring the J-Space, AI Superforecasters, SambaNova's Chips, & LTX Video Gen
WithDan Schwarz, Zeev Farbman, Kunle Olukotun
Questions people ask
- What is dataflow computing for AI inference?
- It involves building computers that dynamically configure themselves to match the dataflow graph of an AI model, moving beyond traditional CPU/GPU instruction fetching.
- How does this architecture benefit LLM inference?
- It is well-suited for LLM inference, reducing memory bandwidth bottlenecks and improving overall performance.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sam Charrington.