Hardware Podcasts

AMD's MI300 and the Future of Accelerated Compute

Oxide and Friends · 12 Dec 2023 · 1h 14m · Episode 34

What this episode covers

George Cozma of Chips and Cheese and Jordan Cozma of Storage Review join Bryan Cantrill and Adam Leventhal of Oxide Computer Company to analyze AMD's MI300 series, particularly the MI300A APU. The group compares AMD's unified memory architecture and open-source ROCm software stack against NVIDIA's Grace Hopper and CUDA ecosystem, discussing the implications for accelerated compute, data center power density, and the future of AI hardware.

Most relevant to

large enterprises

Topics

Mentioned

Companies
AMD, NVIDIA, Oxide Computer Company, Chips and Cheese, Storage Review
Products
MI300, MI300A, Grace Hopper
Also
CUDA, ROCm

Related episodes from other shows

Other shows on the same ground, closest first.

  • The Circuit · 42m

    Episode 47: Event Season Insights (Marvell, AMD, Intel)

    Ben Bajarin and Jay Goldberg analyze recent semiconductor events, focusing on Marvell's data center strategy, AMD's MI300 X AI accelerator, and Intel's Meteor Lake client silicon. They discuss Marvell's positioning as a data infrastructure company and its semi-custom business, while noting AMD's strong execution with Microsoft and Meta as customers for its MI300 X, though its ROCm software stack lags behind Nvidia's CUDA. The hosts also examine Intel's Meteor Lake chiplet SOC, debating whether it represents a generational leap or merely a competitive response to Apple's M-series chips, and express skepticism about the near-term consumer value of AI PCs.

    Chips, FPGAs & semiconductorsAI & machine learning
    7 chapters

  • The Circuit · 33m

    Episode 23: Emerging Competitive Dynamics in Silicon Thanks AI

    Ben Bajarin and Jay Goldberg discuss the emerging competitive dynamics in the semiconductor industry driven by AI, debating whether AI spending is additive to data center capital expenditure or cannibalizes traditional cloud processor budgets. They analyze AMD's recent AI and Datacenter event, highlighting the MI300X chip and its modular chiplet architecture as a strategic response to NVIDIA's dominance in training workloads. The hosts also examine the shift toward Total Cost of Ownership (TCO) and performance-per-watt metrics in the 'cloud native' segment, where AMD and Intel are competing with ARM-based designs like those from Ampere and Amazon Graviton.

    Chips, FPGAs & semiconductorsPricing, margins & unit economics
    6 chapters

  • The Circuit · 45m

    Episode 89: Advancing AI, thoughts on AMD's Event, Nvidia, and Chain of Reasoning AI

    Ben Bajarin and Jay Goldberg discuss AMD's recent AI event, highlighting the launch of the MI325X accelerator and the strategic shift toward vertical integration in AI systems, where vendors like AMD and NVIDIA provide complete rack solutions. They analyze the competitive landscape, noting AMD's progress against NVIDIA's Blackwell and the growing importance of CPU performance in AI workloads due to 'agentic' or 'chain of reasoning' models. The hosts also explore the implications for hyperscalers, who are increasingly designing their own silicon (like Google's Axion and Amazon's Graviton) to optimize total cost of ownership, and touch on the scaling challenges and sensor costs in autonomous vehicle technology from Tesla and Waymo.

    Chips, FPGAs & semiconductorsAI & machine learning
    10 chapters

  • EE Times Current · 46m

    2020 Mega-Merger #3 | An AI Ecosystem (Almost) from Scratch

    Tirias Research analyst Kevin Krewell explains that AMD's $35 billion acquisition of Xilinx creates a diversified powerhouse, combining AMD's CPU/GPU strength with Xilinx's FPGA and Versal technologies to target data center, edge, and embedded markets. Meanwhile, IBM Research VP Mukesh Khare and Synopsys VP Arun Venkatachar discuss the IBM Research AI Hardware Center, a public-private partnership aimed at building an end-to-end AI ecosystem from devices to software. They detail their goal of improving AI compute efficiency by 1,000 times in 10 years through innovations in heterogeneous integration, chiplets, and analog AI, leveraging Synopsys' EDA tools and the broader ecosystem including Samsung and Applied Materials.

    Chips, FPGAs & semiconductorsAI & machine learning
    5 chapters

  • The Circuit · 37m

    Episode 122: Apple's WWDC, AMD AI Event, Semiconductors 2.0

    Ben Bajarin and Jay Goldberg analyze Apple's WWDC, noting its strategic pivot to an AI platform for developers by exposing small language models and APIs, while highlighting the organizational risk of internal stalemate and OpenAI's default partnership. They discuss AMD's AI event, emphasizing Lisa Su's confirmation of the MI300/400 roadmap, the shift from evaluation to production workloads with major customers like Meta and OpenAI, and the competitive necessity of full-rack networking solutions like Helios to challenge Nvidia's dominance.

    Connected products & IoTChips, FPGAs & semiconductors
    6 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Oxide and Friends. The episode itself belongs to its publisher.