Academic Databases
Explore Our Semantic Databases
STEM & Tech
Math, Physics, CS & Engineering...
Economics
Macro, Micro, Econometrics...
Business
Management, Marketing, Finance...
Health & Life Sciences
Biology, Medicine, Neuroscience...
Humanities & Social Sciences
Psychology, Law, Philosophy, History...
General Studies
General & interdisciplinary studies...
Explore Topics
Features

AMD CEO Lisa Su on High Performance Computing Vision

Added: 2026-08-15

Deep Dive

Basic Computer Architecture: Fundamental differences between Central Processing Units (CPUs) and Graphics Processing Units (GPUs) and how they process workloads.

MCS-012 ( Computer Organization and Assembly Language Programming)Block01 Unit-1 THE BASIC COMPUTER

MCS-012 ( Computer Organization and Assembly Language Programming)Block01 Unit-1 THE BASIC COMPUTER

42.5K

The basic computer architecture consists of interconnected components: memory for storing data and instructions, Central Processing Unit (CPU) with Arithmetic Logic Unit (ALU), Control Unit, and various registers, and input-output systems. Von Neumann architecture, designed in 1945 by John von Neumann, is also known as basic computer architecture. It uses a single memory system for both instructions and data, which creates the Von Neumann bottleneck problem. This architecture, consisting of a control unit, ALU, memory, registers, and I/O devices, remains the foundation for most modern computers including laptops and PCs.

Click to watch →

Complete COA in One Video | Coa Rgpv Exam True Engineer Limitless Learning #rgpv

Complete COA in One Video | Coa Rgpv Exam True Engineer Limitless Learning #rgpv

37K

A desktop computer consists of fundamental units: Input Unit (receives data), Control Unit (manages operations), Arithmetic Logic Unit (ALU) (performs arithmetic and logical operations), Registers (temporary data storage), Memory Unit (data storage), and Output Unit (presents results). The ALU handles addition, subtraction, multiplication, division, and logical operations like AND, OR, NOT. Registers are groups of flip-flops that store binary data, categorized as General Purpose (store data or addresses) and Special Purpose (store instruction addresses, program counters, memory addresses). Bus organization enables data transfer between CPU and components through Address Bus (carries memory addresses), Data Bus (transfers actual data), and Control Bus (carries control signals).

Click to watch →

I Made a 32-bit Computer Inside Terraria

I Made a 32-bit Computer Inside Terraria

4.9M

At a high level, computers consist of four main components: (1) A control unit that fetches instructions from memory, (2) An Arithmetic Logic Unit (ALU) that performs calculations, (3) Registers that store temporary data during computation, and (4) Memory that stores both programs and data. The computer repeatedly fetches instructions, executes them using the ALU, and stores results in registers or memory.

Click to watch →

Dica 02 - Arquitetura Básica - Parte 1

Dica 02 - Arquitetura Básica - Parte 1

44K

A computer's basic architecture consists of four main components: input peripherals (such as keyboard, mouse, webcam, and scanner) that send commands and data to the CPU, the CPU (Central Process Unit) that executes programs and processes data, output peripherals (such as monitor, printer, and speakers) that display processing results, and memory devices that store processed data. The CPU is considered the brain of the computer as it performs the essential processing and storage functions. Some devices like touch screens, multifunctional printers, and modems serve as mixed peripherals that handle both input and output operations simultaneously.

Click to watch →

Colour Maximite 2 (A Modern BASIC Computer) - Unboxing & First Impressions

Colour Maximite 2 (A Modern BASIC Computer) - Unboxing & First Impressions

40.1K

A boot to basic computer is a single-board computer that boots directly to a BASIC programming environment without running an operating system first. Unlike modern computers that load an OS and then a programming environment, these systems boot straight to a command line that executes BASIC commands and provides editing capabilities. This design philosophy prioritizes immediate programming access over system complexity.

Click to watch →

Deep Dive

Introduction to High-Performance Computing (HPC): The concept of parallel processing and how supercomputers scale to solve complex scientific calculations.

What is HPC? An introduction to High-Performance Computing

What is HPC? An introduction to High-Performance Computing

52.3K

High-Performance Computing (HPC) combines computational resources into clusters or supercomputers to solve complex problems requiring more processing power than typical desktop computers, evolving from single-system supercomputers to distributed memory architectures using message-passing interfaces that enable multiple commodity servers to work together over high-speed networks, making advanced computational capabilities accessible to organizations of all sizes through public clouds, private clouds, or hybrid solutions.

Click to watch →

What is HPC? High Performance Computing | HPC Introduction with Rake Diagram Connectivity #hpc

What is HPC? High Performance Computing | HPC Introduction with Rake Diagram Connectivity #hpc

7.9K

High-Performance Computing (HPC) is a computational approach that combines multiple computers (or nodes) to process complex data and perform intensive calculations at significantly higher speeds than standard desktop computers or workstations. HPC systems utilize parallel processing, where multiple compute nodes (including CPU nodes and GPU nodes) work simultaneously on a single task, enabling users to complete tasks that would otherwise take weeks or months in just minutes. This technology is essential for running large-scale software applications, scientific simulations, and data-intensive workloads that exceed the capabilities of conventional computing hardware.

Click to watch →

3.1 Introduction to HPC

3.1 Introduction to HPC

294

High Performance Computing (HPC) is a computing approach that uses interconnected small computers to solve computationally heavy tasks faster than single computers, particularly when programs take too long to process or datasets are too large to fit on one machine. An HPC cluster consists of three main components: the login node where users prepare source code and manage workflows (shared by multiple users simultaneously), compute nodes with CPUs, RAM, disk space, and optionally GPUs where actual programs run, and a scheduler (like SLAM) that allocates resources, monitors jobs, and manages a queue of pending jobs waiting for computational resources. The key principle of HPC is parallel computing, where code is written to utilize multiple nodes simultaneously to achieve faster execution.

Click to watch →

HPC Thusuok ding chun awm i ti ve am??

HPC Thusuok ding chun awm i ti ve am??

14K

HPC (High-Performance Computing) refers to the use of supercomputers and parallel processing systems to solve complex computational problems that require significant processing power, such as scientific simulations, data analysis, and large-scale modeling.

Click to watch →

Introduction to HPC Computing A Practical Tutorial, Marco Verdicchio, SURFsara

Introduction to HPC Computing A Practical Tutorial, Marco Verdicchio, SURFsara

18.6K

High-performance computing (HPC) involves using supercomputers composed of many interconnected CPUs and specialized hardware (like GPUs or FPGAs) to solve complex computational problems through either scaling up (moving to more powerful machines) or scaling out (distributing work across many processors). Users access HPC systems remotely via SSH to login nodes, then prepare job scripts containing UNIX commands and submit them to batch schedulers (like SLURM) which manage compute node allocation. Essential skills include mastering basic UNIX/Linux commands (pwd, ls, cd, cp, scp, mkdir), understanding bash scripting for automating workflows, and learning how to run parallel jobs using techniques like task parallelism with background execution or more advanced methods like MPI for distributed memory parallelism.

Click to watch →

Deep Dive

Data Center Foundations: An understanding of cloud infrastructure, virtualization, and the hardware demands of enterprise-level data centers.

We Are Just Having A Private Meeting 😮🛸🎥❌✔

We Are Just Having A Private Meeting 😮🛸🎥❌✔

57.1K

Large-scale data center construction requires robust foundation systems including steel pilings driven into the ground to support the building's weight. Underground tanks are often installed for water supply or other utilities. The construction process involves pouring concrete into corrugated flooring to create strong structural elements. These foundational systems ensure the building can withstand operational loads and environmental conditions.

Click to watch →

5 Stocks Making People RICH Off the AI Bubble (Not AI Companies)

5 Stocks Making People RICH Off the AI Bubble (Not AI Companies)

107.3K

AI depends on complex infrastructure that remains invisible to consumers but absolutely essential for operations. Vertiv (VRT) provides cooling systems, server racks, and turnkey solutions for data centers, managing the thermal challenges of high-performance computing environments. Arista Networks (ANET) supplies the networking equipment enabling communication between servers and GPUs. These companies represent the 'always and forever' infrastructure that persists regardless of technological changes. The broader lesson is that successful investing requires shifting from a consumer mindset (using products) to an owner mindset (understanding what enables products to function), identifying companies that provide irreplaceable services behind the scenes.

Click to watch →

Data Center Fundamentals Course – What is a Data Center

Data Center Fundamentals Course – What is a Data Center

10.3K

A data center is a physical facility containing servers, storage systems, and networking equipment that supports information technology operations, serving as the backbone of the digital economy; the DCFC course covers its history, relation to business, standards (TIA-942), site selection, facility areas, power infrastructure (including generators and UPS systems), lighting, cooling systems, ICT and network infrastructure, security measures, fire safety protocols, and monitoring/reporting systems, preparing participants for a career in data center operations and management.

Click to watch →

DATA CENTER Infrastructure Design for Aspiring Engineers @sangelsoft #creative sessions #datacenter

DATA CENTER Infrastructure Design for Aspiring Engineers @sangelsoft #creative sessions #datacenter

9.1K

Data centers are physical facilities housing computer systems for storing, processing, managing, and distributing large amounts of data. They consist of networked computers, storage infrastructure, and cooling systems. Key infrastructure elements include raised access floors for cable and utility distribution, proper cable pathway sizing with 50% capacity buffer, and strategic routing of power and telecommunications cabling to cold and hot aisles respectively. Grounding and bonding ensure safety by minimizing current flow and directing electromagnetic interference to ground. Fire protection relies on Vesta aspirating smoke detection systems that sample air continuously for rapid fire response.

Click to watch →

What is a Data Center? – Data Center Fundamentals

What is a Data Center? – Data Center Fundamentals

106.5K

A data center is fundamentally a facility that houses digital infrastructure, ranging from small closet setups to massive $150+ million purpose-built facilities. The industry represents a unique commercial real estate class that is smaller than traditional sectors but highly capital intensive—building costs are approximately 10 times higher than equivalent office buildings. Location selection involves evaluating power availability, infrastructure access, labor markets, and critical risk factors including flood plains, hurricane zones, tornado-prone areas, and earthquake fault lines. Organizations invest substantial capital in these facilities, necessitating thorough due diligence on all location-related risks.

Click to watch →

Deep Dive

Understanding FPGAs: What Field Programmable Gate Arrays are and their role in specialized hardware acceleration compared to fixed-silicon ASICs.

What's an FPGA?

What's an FPGA?

282.5K

An FPGA (Field Programmable Gate Array) is a configurable integrated circuit that executes code by synthesizing hardware description language programs into physical hardware blocks composed of registers and logic gates (AND, OR, NOR, NOT, XOR, XNOR), where look-up tables are programmable and interconnected through a switching fabric to create a reconfigurable computing device.

Click to watch →

What is an FPGA (Field Programmable Gate Array)? | FPGA Concepts

What is an FPGA (Field Programmable Gate Array)? | FPGA Concepts

294.4K

An FPGA (Field Programmable Gate Array) is a reprogrammable integrated circuit composed of configurable logic blocks (CLBs) containing lookup tables, multiplexers, full adders, and flip-flops, interconnected through reconfigurable pathways with input/output blocks and specialized functional blocks like multipliers and block RAM, offering advantages such as faster parallel signal processing, reprogrammable hardware structure eliminating the need for remanufacturing, and cost-effectiveness compared to application-specific integrated circuits (ASICs), making them versatile for applications in aerospace, defense, medical devices, automotive systems, and wireless communications.

Click to watch →

¿Que son los FPGAs? - Hackeando Tec

¿Que son los FPGAs? - Hackeando Tec

29.2K

FPGAs (Field Programmable Gate Arrays) are semiconductor devices composed of configurable logic blocks (CLBs) connected by programmable interconnections, allowing them to be reprogrammed for various applications; they consist of CLBs (the basic logic units containing LUTs and flip-flops), I/O blocks for interfacing with external signals, RAM memory columns, and clock managers, making them versatile for applications ranging from aerospace and defense to automotive, medical, and artificial vision systems.

Click to watch →

What is an FPGA?

What is an FPGA?

6.3K

An FPGA (Field Programmable Gate Array) is a reconfigurable electronic device composed of configurable logic blocks (CLBs) containing flip-flops, lookup tables, and multiplexers connected through an interconnect matrix, which allows users to program custom circuitry using HDLs like Verilog or VHDL; unlike microcontrollers that process sequentially and ASICs that are fixed after manufacturing, FPGAs enable parallel processing and field reprogramming, making them ideal for applications requiring high performance in a small package with future upgradeability, though they typically have higher power consumption and cost compared to microcontrollers.

Click to watch →

FPGA in HFT Systems Explained | Why Reconfigurable Hardware Beats CPUs

FPGA in HFT Systems Explained | Why Reconfigurable Hardware Beats CPUs

53.4K

Inside each CLB, there's a tiny component called a Lookup Table (LUT). A LUT functions like a very small hard-coded table that maps inputs to outputs. For software engineers, a LUT is similar to a pre-filled hashmap or function table. In an FPGA, the LUT behaves like an XOR gate. Instead of hardwiring a logic function like an AND or OR gate, FPGAs use LUTs to simulate behavior by storing a truth table. This means a single LUT can be reprogrammed to act like any small logic function.

Click to watch →

Event Kickoff

Post-Pandemic Era

Data Evolution

Acquisition Strategy

Industry Frontier

Semiconductor Crunch

Event Kickoff

0:08

Playing Section

1
Summit launched by host, setting the stage for keynote.
2
AMD CEO Lisa Su's vision and leadership take center stage.
3
High-performance computing's crucial role is highlighted as core focus.

Deep Dive

Heterogeneous Computing Architectures: Deep dive into how CPUs, GPUs, and adaptive SoCs (like Xilinx FPGAs) are integrated using high-speed interconnect technologies.

Архитектура и эмуляция PlayStation 3

Архитектура и эмуляция PlayStation 3

62.4K

Heterogeneous computing is an architectural approach where a system contains multiple specialized processors with different architectures and clock speeds, all connected through shared data buses and coordinated by a central processor. This approach allows each component to handle specific tasks efficiently while the central processor manages overall system coordination. The PlayStation 1 used this concept with its Geometry Transformation Engine (GTE) as a specialized coprocessor for geometry calculations, separate from the main MIPS processor.

Click to watch →

Nvidia is already reducing RTX 50 GPU production? - Tech News Sept 7

Nvidia is already reducing RTX 50 GPU production? - Tech News Sept 7

59.8K

Modern processors increasingly incorporate heterogeneous computing architectures combining different types of processing cores. The leaked Nvidia N1X CPU features 20 total cores divided between ARM Cortex X925 performance cores and A725 efficiency cores, while its GPU specifications rival the RTX 5070 with 48 streaming multiprocessors and 6,144 CUDA cores. This integration of CPU and GPU capabilities on a single chip enables diverse workloads from high-performance computing to graphics rendering, though power limits typically constrain such designs to mobile and handheld devices.

Click to watch →

Design Exciting Chips in the Classroom, Presented By Prof Bora Nikolic

Design Exciting Chips in the Classroom, Presented By Prof Bora Nikolic

1.5K

Modern machine learning accelerators implement heterogeneous computing architectures with multiple processor cores and specialized accelerators. The Barely ML chip featured four RISC-V Rocket cores each equipped with custom sparse matrix accelerators, plus a Saturn5 vector processor. Dense matrix computations were implemented near L2 cache banks using DMA-accessible arithmetic units. This architecture demonstrates mapping different ML workloads—dense versus sparse linear algebra—to appropriate computational resources for optimal performance, showing how to combine multiple accelerator types on a single chip.

Click to watch →

Перевод презентации Radeon VII и Ryzen 3000 AMD CES 2019 - Lisa Su Keynote

Перевод презентации Radeon VII и Ryzen 3000 AMD CES 2019 - Lisa Su Keynote

122.2K

Heterogeneous computing combines different types of processors (such as CPUs, GPUs, and specialized accelerators) within a single system to maximize performance for diverse computational tasks. This approach allows systems to leverage the strengths of each processor type for specific workloads, providing optimal performance across a wide range of applications from cloud computing to local device processing.

Click to watch →

PS5 Easier To Develop For Says Crytek Engineer - Analysis | Nvidia Next Gen GPU Spotted

PS5 Easier To Develop For Says Crytek Engineer - Analysis | Nvidia Next Gen GPU Spotted

20.9K

This segment covers AMD's Exascale Heterogeneous Processor (EHP) concept combining high-performance CPUs, GPUs, and high-bandwidth memory for server computing. The architecture has evolved through bifurcated GPU designs (CDNA for data centers, DNA for gaming) and Infinity Fabric improvements (CNA, CNA2). Key innovations include dynamic memory management patents allowing concurrent memory allocation modification during kernel execution, and GPU interconnect enabling data sharing between multiple GPUs and system memory. Nvidia's parallel development maintains separate architectures for data center (Volta) and gaming (Turing) applications, with Ampere architecture still unconfirmed and launch potentially delayed to Q4 due to manufacturing challenges.

Click to watch →

Deep Dive

AI and Deep Learning Hardware Accelerators: Investigating the specific silicon architectures optimized for training and deploying large language models and neural networks.

691: A.I. Accelerators: Hardware Specialized for Deep Learning — with Ron Diamant

691: A.I. Accelerators: Hardware Specialized for Deep Learning — with Ron Diamant

3.4K

AI accelerators like AWS's Trainium and Inferentia chips achieve significantly higher energy efficiency (99% vs. 50% in traditional GPUs/CPU) by using systolic array architectures that minimize instruction overhead, while also employing different form factors optimized for either training (high-bandwidth interconnect for scale-up workloads) or inference (compute density for scale-out deployment), demonstrating how specialized hardware design can address the growing computational demands of deep learning through architectural innovation rather than incremental improvements.

Click to watch →

Fundamentals of Circuit Design for AI/ML - Charlotte Frenkel | ESSERC 2024 | Circuit Insights (6)

Fundamentals of Circuit Design for AI/ML - Charlotte Frenkel | ESSERC 2024 | Circuit Insights (6)

2.6K

This section introduces the speaker's research background in hardware accelerators for AI and machine learning, including neuromorphic engineering and brain-inspired computing. It covers the two main design schools: traditional machine learning requiring feature engineering with human insight, and modern deep learning using end-to-end blackbox models that derive parameters from data. The section explains how GPUs catalyzed the deep learning revolution in 2012 by providing superior parallel performance over CPUs. It introduces the three key design principles for AI hardware: parallelism, data reuse, and memory hierarchy. The deep learning network architecture is explained, including input/output units, weighted sums, and nonlinearities like ReLU. The distinction between inference (making predictions) and training (adjusting weights) is clarified.

Click to watch →

ECE Colloquium: Bill Dally: Deep Learning Hardware

ECE Colloquium: Bill Dally: Deep Learning Hardware

3K

Deep learning requires three ingredients: data (fuel), algorithms (air), and hardware (spark). Since AlexNet in 2012, AI hardware demand has increased by two orders of magnitude every two years, and one order of magnitude per year since 2017. NVIDIA GPUs evolved from Kepler (4 TFlops FP32) to Hopper (4 PFlops), achieving 1,000x improvement. The key insight is that hardware developers are gating AI progress, creating continuous pressure to improve performance as researchers want to run larger models on more data.

Click to watch →

VisionFive RISC-V Linux SBC

VisionFive RISC-V Linux SBC

234.3K

The Vision 5 includes several AI-focused hardware accelerators: a 600 MHz Tensilica VP6 Vision DSP/VPU, a 500 MHz neural network engine, and an 800 MHz NVDLA (NVIDIA Deep Learning Accelerator). These components make the board targeted toward AI applications and the emerging AIoT (Artificial Intelligence of Things).

Click to watch →

Unboxing the Tenstorrent Grayskull AI Accelerator!

Unboxing the Tenstorrent Grayskull AI Accelerator!

57K

Modern AI hardware development requires a dual-approach strategy that provides both high-level accessible entry points (like the Buddha compiler for quick model deployment) and low-level bare-metal programming interfaces (for fine-grained hardware control), enabling developers to choose their preferred abstraction level while maintaining forward compatibility across hardware generations.

Click to watch →

Deep Dive

Semiconductor Industry Dynamics & Strategy: Analyzing the competitive landscape (e.g., AMD, Intel, NVIDIA) and the strategic value of mergers and acquisitions in tech.

권영화 교수 ‖ '최신 반도체 이슈' 반도체는 앞으로 2~3년은 계속 좋다고 생각합니다 (풀버전) / 머니특강

권영화 교수 ‖ '최신 반도체 이슈' 반도체는 앞으로 2~3년은 계속 좋다고 생각합니다 (풀버전) / 머니특강

10.9K

The semiconductor industry is recovering with SK Hynix and Samsung expected to exit losses over 2-3 years. Unlike mature industries like steel, semiconductors continue growing due to continuous innovation. Memory semiconductors are recovering first, with system semiconductors following. Samsung's counter-cyclical investment strategy provides competitive advantage, while SK Hynix faces capacity expansion delays. TSMC has 35 years of foundry experience while Samsung's began in 2017, creating substantial gaps in capacity, workforce, and ecosystem maturity. Korea's semiconductor design market share is only 3.4% compared to America's 67%, indicating significant weakness in design capabilities. Two key growth drivers are artificial intelligence and automotive semiconductors, with AI experiencing explosive growth and automotive semiconductors growing at 10%+ annually. Vehicle semiconductor usage is growing dramatically: traditional vehicles use ~300 chips, electric vehicles use 1,000+ chips, and autonomous vehicles may require 3,000-10,000 chips. There are significant technology gaps between manufacturers: Samsung focuses on advanced processes (3nm, 2nm) while competitors focus on mature processes. Japan is positioning itself as a strategic alternative to Taiwan and Korea due to geopolitical concerns. China's semiconductor strategy focuses on mature processes (20nm+) due to equipment restrictions, with the mature process market being larger than advanced processes. Korea imports approximately 80% of semiconductor equipment, with advanced processes requiring equipment from American and Japanese companies. NVIDIA dominates the AI semiconductor market with 80-90% market share, with the market still in early stages and projected 10x growth. The semiconductor industry follows 3-4 year cycles of boom and bust, with new applications like AI driving sustained growth beyond traditional cycles.

Click to watch →

반도체 말고 '이것' 사라! 역대급 하락장 속에서 찾아낸 새로운 부의 이동 [판도]

반도체 말고 '이것' 사라! 역대급 하락장 속에서 찾아낸 새로운 부의 이동 [판도]

7.2K

The semiconductor industry faces memory price constraints limiting pricing flexibility, leading to increased long-term supply contracts (5-10 years) for revenue stability. China's semiconductor industry is in early development stages without full cyclical experience, with government subsidies supporting domestic capabilities. Hyperscalers like Amazon have demonstrated data center investments can generate strong returns (AWS growth of 36%), improving market confidence. Vertical integration strategies provide cost efficiencies similar to Tesla's battery approach. South Korea faces challenges in AI positioning despite strong semiconductor manufacturing capabilities.

Click to watch →

【米国株】半導体株トップ１０！

【米国株】半導体株トップ１０！

89.4K

The semiconductor industry operates as a highly competitive 'war of the nations' where companies with weak financials face acquisition threats. The industry structure includes fabless companies (Apple, NVIDIA, AMD, Google, Amazon, Facebook, Microsoft) that design chips but outsource manufacturing to foundries like TSMC. TSMC dominates global semiconductor manufacturing, controlling the supply chain for most major companies. ARM provides CPU designs that many companies license, creating a complex ecosystem where companies compete for talent, technology, and market share through strategic acquisitions and in-house development.

Click to watch →

2나노 로드맵 발표한 삼성전자!! 그 뒤에는 미국 IBM이 있다! (Feat. 파운드리에서 중요한 설계자산)

2나노 로드맵 발표한 삼성전자!! 그 뒤에는 미국 IBM이 있다! (Feat. 파운드리에서 중요한 설계자산)

91.5K

The semiconductor industry is characterized by intense competition between major players like Samsung Electronics and TSMC, with TSMC holding dominant market position through its foundry-only strategy. Samsung is responding by investing heavily in advanced process technology, including GAA transistor technology for 3nm and 2nm nodes, with plans to begin mass production by 2025. The industry's evolution has been shaped by strategic outsourcing decisions, such as IBM's transition from in-house manufacturing to partnering with Samsung Electronics and Hynix, which catalyzed the growth of the Korean semiconductor industry. TSMC's extensive IP library, containing over 37,000 designs, provides a significant competitive advantage. The industry's growth is driven by emerging technologies including AI, autonomous vehicles, 5G communications, and the metaverse, while geopolitical factors like US-China trade tensions create uncertainty affecting market dynamics.

Click to watch →

지금부터 '이 종목' 투자 안하면 다 털립니다 삼성전자 하이닉스 그리고 에코프로 이후 미친듯이 오를 '이 종목'은?ㅣ주식클라쓰 하창봉 대표

지금부터 '이 종목' 투자 안하면 다 털립니다 삼성전자 하이닉스 그리고 에코프로 이후 미친듯이 오를 '이 종목'은?ㅣ주식클라쓰 하창봉 대표

9.5K

The semiconductor industry experiences significant dynamics driven by technology evolution, strategic partnerships, and market positioning. NVIDIA's GTC26 conference highlights next-generation technology including Blackwell and Rubin architectures, with revenue targets reaching $1 trillion by 2027. Korean semiconductor companies like Samsung and SK Hynix face challenges including foundry financial losses and competitive pressures. Strategic positioning involves ADR listings for improved market valuation, substrate industry growth from device complexity, and equipment market opportunities. The industry requires careful balance between technological advancement, financial management, and global market access.

Click to watch →

Deep Dive

Exascale Computing Case Studies: Studying real-world deployments of modern supercomputers (such as the Frontier supercomputer) to see how these hardware visions are realized in practice.

What is Exascale Computing?

What is Exascale Computing?

22K

Exascale computing represents the next generation of supercomputing technology, capable of performing quintillions (10^18) of operations per second, enabling unprecedented problem-solving for national security, precision medicine, climate research, and scientific discovery through advanced multiphysics simulations that can complete complex calculations in hours rather than weeks.

Click to watch →

World's Most Powerful Supercomputers

World's Most Powerful Supercomputers

204.6K

The transition to exascale computing represents a quantum leap in supercomputing capability. Fugaku (Fujitsu, Japan) achieved 422 PetaFLOPS in May 2020, nearly tripling Summit's performance. It dominated multiple benchmarks simultaneously (Graph 500, HPL-AI, HPCG). Frontier (HPE/Cray, Oak Ridge) became the world's first exascale computer in June 2022, achieving 1.102 ExaFLOPS (1 quintillion operations/second), 10 times more powerful than Summit. This progression demonstrates how nations compete for computational supremacy through increasingly ambitious performance targets.

Click to watch →

Lori Diachin, Exascale Computing Project Director, Speaking to ECP's Industry and Agency Council

Lori Diachin, Exascale Computing Project Director, Speaking to ECP's Industry and Agency Council

250

ECP funded applications spanning molecular dynamics, lattice QCD, power grid modeling, cancer research, and wind energy, representing 62 codes over 10 million lines of software. Success was measured through four Key Performance Parameters: 50x performance improvement targets (10/11 applications succeeded), capability applications demonstrating GPU use (9/10 Office of Science, 3/4 NNSA), software integration scores (66.5/68 points achieved), and vendor milestone completion (100%). The E4S software stack provides over 115 products with multi-layer testing ensuring compatibility. Key innovations included AMRex as an integration nexus, EXO Wind's high-fidelity turbine simulations, and algorithmic redesigns for GPU efficiency in reactor simulations, demonstrating how foundational software enables broad scientific impact.

Click to watch →

Erik Draeger on Exascale Computing Project Applications Lessons Learned

Erik Draeger on Exascale Computing Project Applications Lessons Learned

155

Exascale computing projects require strategic foundations including diverse application portfolios spanning multiple scientific domains, starting from established codes rather than writing new software, categorizing projects into performance-focused (achieving 50x+ gains on existing hardware) and capability-focused (building entirely new scientific capabilities), and planning for hardware diversity including multiple GPU vendors rather than assuming single-platform dominance.

Click to watch →

HW News - Intel & Apple Lawsuit, Linux Laptops, & AMD RDNA2 for GPUs

HW News - Intel & Apple Lawsuit, Linux Laptops, & AMD RDNA2 for GPUs

251.6K

The Aurora supercomputer represents Intel's entry into exascale computing, combining Sapphire Rapids CPUs with Ponte Vecchio accelerators in a Multi-Chip Module (MCM) design. This architecture uses Intel's advanced packaging technology and EMIB interconnect fabric to scale to thousands of execution units while reducing manufacturing costs. The system will support over 200 nodes, 230 petabytes of storage, and 10 petabytes of memory. Ponte Vecchio will debut PCIe 5.0 support through Intel's XE Link interface, enabling memory coherency between GPU dies and CPUs. This infrastructure demonstrates how modern supercomputers integrate multiple processing technologies to achieve exascale performance for HPC and AI workloads.

Click to watch →

Event Kickoff

Post-Pandemic Era

Data Evolution

Acquisition Strategy

Industry Frontier

Semiconductor Crunch

Event Kickoff

0:08

Playing Section

1
Summit launched by host, setting the stage for keynote.
2
AMD CEO Lisa Su's vision and leadership take center stage.
3
High-performance computing's crucial role is highlighted as core focus.

93.6K views8K13:54@SixFiveMediaOriginal Release: 2021-06-17

High performance computing is essential for modern digital transformation, requiring heterogeneous architectures that combine CPUs, GPUs, ASICs, and FPGAs to optimize different workloads efficiently; the semiconductor industry faces a positive cycle of increasing demand driven by compute's pervasiveness in all aspects of life, while strategic acquisitions like AMD's planned Xilinx purchase enable broader portfolio offerings and accelerated innovation across data centers, cloud, PCs, and edge computing.

[00:00:08]hello and welcome to the 6-5 summit i'm shelley kramer one of the founding partners of future research and on behalf of my team here at futurem and the team at moore insights and strategy welcome we're thrilled to have you in this keynote conversation more insights patrick morehead sits down with amd's president and ceo lisa sue and their conversation today is all about high performance computing and the role that hbc plays in our lives from the data center to the edge this technology is essential and i can't wait to hear what lisa has to share about amd's vision for the future let's go check it out lisa thank you so much for coming to the 6'5 summit for the second year in a row you're doing the big keynote for semiconductors and devices thank you so much it's great to be here pat thanks for having me this year absolutely it's been very busy year for amd so maybe the best place to start is maybe catch us up on some of the highlights and what you've been up to for for the last year yeah so uh pat it has been an incredibly busy year i think for all of us and uh you know we're always focused on bringing our new technologies to market and you know really focused on high performance computing as you know our cornerstone um and it's been a busy year you know lots of new products um you know we launched new products um with our zen 3 architectures and our rdna 2 architecture and you know across all of our markets so you know pcs gaming data center um so yeah it's been an incredibly busy time yeah it there doesn't go a week where i don't see some big home run announcement or something going on with amd it's exciting exciting place to be and it's fun to say i used to work there so right here exactly right here uh so the theme of this year's summit is is the roaring 20s and we're not out of covet yet but boy we are slowly inching we see the light at the end of the tunnel uh can you look into your crystal ball and talk about how are you looking for the opportunities of some people call the new new normal uh what comes next but we certainly are in a new period i've seen road maps particularly on sas properties and collaboration get pulled in years and i know that that has to have an impact there's a huge growth in the cloud but what are you seeing uh for this next new normal yeah i mean i i think you're you're right there's been just so much um acceleration of the pace of um of innovation and again the way you know i look at it i actually i would call it hopefully the better normal not even the the new normal but the the better normal is you know we've um we've now lived in a world where everybody realizes how important um semiconductors and and compute are um to everything that we do right whether you're a business or you know um you know you're a person or you're you know all of these applications um need more chips need more um computing and and so what we really see is you know this vicious positive cycle right so that this the cycle is we've now um you know adopted compute every aspect of our lives and um and that's true also in terms of digital transformation and businesses um we've worked out some kinks this year frankly it hasn't been the you know the technology hasn't been perfect but now it's better and it's more capable and we're going to make it even accelerate the use of technology going forward so yeah i think it's a incredible time to to be in the business yeah it is uh funny even 10 years ago when people are saying software is eating the world i would say well software has to run on something and it is incredible when you align semiconductors uh software in the cloud and and i'd like to double click in on the data center and uh under you're in the middle of all of that and i'm wondering what are you seeing some of the major trends the next three to five years in the data center and how uh is amd prepared to take advantage of that well look the data is extremely strategic to us and it's i think strategic to everyone and you know we look at the data center as you know sort of an evolving environment that is increasingly more and more heterogeneous so you know the idea that you know one size doesn't fit all you know frankly with as much data that is out there and as much as you want to do with it you actually need to optimize for various workloads and so you need cpus and gpus and asics and fpgas and you know from an amd standpoint you know our job is to really provide the foundational technology but then also the ability to mix and match the ip as you need in the data center and work very closely with our customers to bring that innovation to the market yeah the other thing that i'm seeing too is um it's not just in a huge building i mean it's on the edge it's in the mid-edge and we get into these debates on on what is the edge what is the data center edge but i think that's all goodness because uh you have to build this out to be able to take advantage of all that data that's being created on the edge so we're seeing a lot out there well actually not a lot but a few people designing their own ships leveraging uh arm ecosystem out there what are your thoughts on on that maybe how that plays out what does that mean and how does that relate to amd well you know i think again i think in the data center it's all about workload optimization so when you think about what the big cloud providers are doing or large enterprises it's trying to they're really trying to optimize their compute for the data that they're processing or the workload that they're trying to uh to handle and so you know i think there's a place for arm in the data center but it's more of a conversation of what are you trying to do like what is it you know what workload are you trying to handle um you know x86 um it really really handles the highest performance general purpose workloads um there's a lot of you know accelerated computing growth so you know you have gpus and other accelerators um and then there are you know some arm-based um you know compute instances as well and you know our goal is actually to pick the best compute for uh for each workload so that that's really the way we think about it yeah heterogeneity is is is important and one of the things that i always try to explain to people is hey don't don't confuse instruction set with uh performance or or even wattage i mean it really comes down to who's developing it uh you know how much you're willing to invest in and i think we see that that playing out every day speaking of heterogeneity amd announced its intention to acquire xilinx and xilinx brings a lot of really interesting ip and products to the table what does it mean for the data center what do people need to think about when it comes to xilinx and amd and the data center well we are really excited about the acquisition of xilinx and you know i think it's a phenomenal team that we're going to bring as a part of amd and it's really the idea of you know let us have the broadest portfolio of sort of high-performance computing elements and we want to put them together in the most efficient way so xilinx brings just leadership in adaptive computing um you know their fpga leadership you know our goal is to make sure that we can integrate that with our cpus and gpus and system solutions for people and when you look at the data center that's exactly what people want right they want to be able to choose these different types of compute and you know i'm also excited about some of the other markets that xilinx you know have a capabilities in you know it's like 5g and communications and automotive and industrial so those are all areas that we think we can you know take our high performance computing technology and really extend it um into a broader set of customers yeah i am uh many times having to explain the xylinks business on what they do and they might be known for fpgas but their heterogeneous designs they have i think are really super exciting i can't wait to see when the deal goes through how amd integrates all of that together yeah i mean if you think about it we'll have the full spectrum of you know general purpose you know cpus straight to you know gpus asics fpgas really in in a um in a total system solution that we can provide and work with customers on so yeah i'm excited as well yeah and even not to mention some asics that are uh slid in there in those designs as well so the full the full uh baileywick so let's talk about a little about what's going on in the semiconductor industry people are waking up and saying hey semiconductor is really important i can't buy my car or i can't get my chromebook uh for for my kid or my um or my pc uh or even uh you know a ps5 or something like that and and it really put a spotlight on on two things i think the first of all is supply okay and also there's been a lot of talk about where ships should be built where they shouldn't be built um what are what are your viewpoints on those well look i think there's no question that semiconductors have you know really become so much more pervasive uh they're in you know everything i mean basically everything and um you know what the pandemic has unveiled is really sort of the demand for semiconductors far exceeded what any of us as an industry predicted and so you know we do have this sort of imbalance between demand and supply that being said i mean you know you know our colleagues in the business are working extraordinarily hard to get more capacity online and that's significant capital investments that's doing all kinds of engineering work to you know qualify and bring more supply to bear so you know we will work through this like we have worked through other other times but no question there's there's significantly high demand and it's it's put a spotlight on the industry you know i think um on your broader question of you know what does it say about the industry i think as semiconductors have now become so essential it's the right conversation right you you want to have um you know geographic diversity you want to be able to ensure that you have assurance of supply and so you know i think it's a it's a great thing that we're talking about you know how do we as an industry and as a country invest more from a research and development standpoint as well as an entrepreneur manufacturing standpoint i love it if nothing else that it reinforces something that i've been seeing for over 30 years is that semiconductors are strategic and and whether you're a business whether you're an oem or an odm you can use semiconductors to get a strategic competitive advantage and that's why i'm i even like to see people doing their own designs and it because more competition and i know amd knows this more competition is is better completely agree when it comes to designs or even when it comes to manufacturing more is better and i think it will rise and i can't wait to see what the next five to ten years uh hold uh for us i agree i agree with that um so i'm going to give you the last word here uh and you know we have your investors we have your biggest customers we have your biggest ecosystem partners actually the biggest ecosystem partners participating uh in in the show as well any final last words any words of wisdom you'd like to share with them well you know i'll say that it's an incredible time to be in the semiconductor business um you know i feel um extremely fortunate and humbled that all of our uh technology is you know powering some of these uh you know incredibly important applications and what i would say you know to um you know to your viewers is look this is a chance for us to partner and accelerate the pace of innovation um and that that really comes when people put um all of their you know sort of technologies and know-how together and so that's very much what we're about is accelerating the pace of innovation in the high performance computing market so you know look forward to seeing um folks in person sooner rather than later and thanks for having us those are great words of wisdom and i feel i feel good that they're very consistent lisa i mean as long as we've known each other you've been very consistent on amd what amd wants to be and what amd wants to do and how it's going to do it and i appreciate that and i think your your customers your ecosystem partners appreciate that as well and i'm i'm excited i'm a technology optimist and i really believe that the technology can help us solve our hardest problems that are out there and we need more high performance computing uh to get there fantastic thank you so much pat yeah so lisa thank you so much uh for being a part of the show and you're gonna you made it so much better thanks so this is pat morehead with more insights and strategy signing off from the 6'5 summit 2021 with amd ceo lisa sue you

Up Next

Gen. Jack Keane: Why Israel Must Dismantle Hamas

@FoxBusiness

340.3K views•2023-11-10

IFS Therapy Demonstration: Complete Session with Unburdening

@IFSCA

95.9K views•2021-01-13

FastAPI vs Flask vs Django: Choosing the Right Python Web Framework

@TechWithTim

302.5K views•2024-05-26

Game of Thrones Opening Credits: A Cinematic Analysis

@gameofthrones

46.3M views•2011-04-18

Related Study Plans & Knowledge Roadmaps

Structured learning paths in General & Interdisciplinary Studies

Browse General & Interdisciplinary Study Plans

general-studies

Containerization: Evolution, Logistics & Trade

01Ancient Trade & Bulk Cargo: Amphorae to Barrels

02The Industrial Revolution & Cardboard Packaging

03The Containerization Revolution: Malcom McLean

+ 2 more learning modules

Explore Roadmap

general-studies

Textile Tech: Evolution, Labor & Global Impact

01Pre-Industrial Textiles: Handloom & Cottage Industry

02Mechanization: The Spinning Jenny to the Power Loom

03The Factory System & Global Labor Transformation

+ 2 more learning modules

Explore Roadmap

general-studies

Pharmacology: Botany, Patents & Biopiracy

01Foundations of Ethnobotany & Traditional Medicine

02The Birth of Synthetic Chemistry: Aspirin & Penicillin

03The Battle Against Malaria: Quinine to Artemisinin

+ 3 more learning modules

Explore Roadmap

general-studies

Writing Surfaces: Papyrus, Parchment & Paper

01Ancient Foundations: Clay, Stone, and Egyptian Papyrus

02Parchment, Vellum, and the Rise of the Codex

03The Chinese Rag Paper Revolution and the Silk Road

+ 3 more learning modules

Explore Roadmap