--- title: Modern Computer Architecture & Organization • Jim Ledin... description: An interview-style book club discussion with Jim Ledin, author of Modern Computer Architecture and Organization (3rd edition), covers how the book's new... canonical: https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026 | daily.dev og:description: An interview-style book club discussion with Jim Ledin, author of Modern Computer Architecture and Organization (3rd edition), covers how the book's new... og:url: https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug og:image: https://api.daily.dev/og/posts/kjjSVi8uG.png og:image:alt: Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026 og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026 **[GOTO Conferences](https://daily.dev/sources/goto)** · 26 min read · 0 upvotes · 0 comments ## Summary An interview-style book club discussion with Jim Ledin, author of Modern Computer Architecture and Organization (3rd edition), covers how the book's new chapters on GPUs and LLMs were written to focus on durable architectural principles rather than fast-moving specifics. Topics include why GPUs turned out well-suited for AI workloads due to matrix multiplication similarities with 3D rendering, the difference between TPUs and GPUs, why GPT-2 was chosen as a simpler case study to explain transformer architecture, memory bandwidth as the current bottleneck for AI hardware (including the DDR5 price spike driven by high-bandwidth memory demand), data center scale engineering challenges, and advice for software developers on understanding pipelining and CPU caches to write more performant code. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why are GPUs well suited for running large language models even though they were designed for graphics rendering? GPUs turned out to be well suited for large language models because both workloads rely heavily on matrix multiplication. Roughly 80-90% of LLM processing comes down to multiplying and accumulating small matrices, typically 16x16, which is exactly the operation GPU hardware was built to accelerate for rendering 3D graphics, lighting, and object transforms, even though GPUs also carry graphics-specific hardware like color shading that LLMs never use. _Anyone comparing GPUs against TPUs for AI workloads can track deeper hardware breakdowns like this on daily.dev._ ### What is the difference between GPU training and inference workloads for a model like GPT-2? Training and inference overlap significantly because training repeatedly runs the inference process itself, then feeds the output back through the network to adjust weights. GPU architecture is generally better aligned with training's dual forward-and-backward passes, while specialized hardware such as TPUs is more often used for pure inference, meaning answering user queries or running agentic tasks after the model is already trained. _Developers weighing GPU versus TPU choices for their AI pipeline can follow this kind of architecture explainer on daily.dev._ ### Why has DDR5 RAM gotten so much more expensive recently? DDR5 memory prices have spiked because manufacturers are shifting production capacity toward high-bandwidth memory (HBM) for AI accelerators like Nvidia's Blackwell architecture and TPUs, which is far more profitable than consumer memory. A 32GB DDR5 stick that cost $69 a year earlier reportedly rose to $440, driven by demand for the memory bandwidth needed to move huge tensors in and out of memory during LLM processing. _Engineers budgeting hardware costs around AI-driven memory shortages can keep tabs on stories like this via daily.dev._ ## Similar posts on daily.dev - [Modern Microprocessors](https://daily.dev/posts/modern-microprocessors-ky8ipxuod) · Hacker News · 0 upvotes · 0 comments - [Why AI Needs GPUs and TPUs: The Hardware Behind LLMs](https://daily.dev/posts/why-ai-needs-gpus-and-tpus-the-hardware-behind-llms-6l5t8bybp) · ByteByteGo · 1 upvotes · 0 comments - [Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC](https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac) · The Developing Dev · 34 upvotes · 1 comments - [Architecting memory and storage in the AI era](https://daily.dev/posts/architecting-memory-and-storage-in-the-ai-era-j0cdvelkz) · MIT Technology Review · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#architecture](https://daily.dev/tags/architecture), [#nvidia](https://daily.dev/tags/nvidia), [#gpu](https://daily.dev/tags/gpu) [View this post on daily.dev](https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026","url":"https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug"},"datePublished":"2026-08-20T12:04:21.025Z","dateModified":"2026-09-14T09:09:50.590Z","description":"An interview-style book club discussion with Jim Ledin, author of Modern Computer Architecture and Organization (3rd edition), covers how the book's new...","image":"https://i.ytimg.com/vi/u7k1l_qkQH8/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/u7k1l_qkQH8/sddefault.jpg","isAccessibleForFree":true,"articleSection":"GOTO Conferences","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"GOTO Conferences","logo":"https://media.daily.dev/image/upload/s--rax7fnoD--/f_auto/v1704628081/logos/goto.jpg","url":"https://daily.dev/sources/goto"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,architecture,nvidia,gpu","timeRequired":"PT26M","video":{"@type":"VideoObject","name":"Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026","description":"An interview-style book club discussion with Jim Ledin, author of Modern Computer Architecture and Organization (3rd edition), covers how the book's new...","thumbnailUrl":"https://i.ytimg.com/vi/u7k1l_qkQH8/sddefault.jpg","uploadDate":"2026-08-20T12:04:21.025Z","duration":"PT26M","url":"https://api.daily.dev/r/kjjSVi8uG","embedUrl":"https://www.youtube.com/embed/u7k1l_qkQH8"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"GOTO Conferences","item":"https://daily.dev/sources/goto"},{"@type":"ListItem","position":3,"name":"Modern Computer Architecture & Organization • Jim Ledin & Maciej «MJ» Jedrzejewski • GOTO 2026"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/modern-computer-architecture-organization-jim-ledin-maciej-mj-jedrzejewski-goto-2026-kjjsvi8ug#faq","mainEntity":[{"@type":"Question","name":"Why are GPUs well suited for running large language models even though they were designed for graphics rendering?","acceptedAnswer":{"@type":"Answer","text":"GPUs turned out to be well suited for large language models because both workloads rely heavily on matrix multiplication. Roughly 80-90% of LLM processing comes down to multiplying and accumulating small matrices, typically 16x16, which is exactly the operation GPU hardware was built to accelerate for rendering 3D graphics, lighting, and object transforms, even though GPUs also carry graphics-specific hardware like color shading that LLMs never use. Anyone comparing GPUs against TPUs for AI workloads can track deeper hardware breakdowns like this on daily.dev."}},{"@type":"Question","name":"What is the difference between GPU training and inference workloads for a model like GPT-2?","acceptedAnswer":{"@type":"Answer","text":"Training and inference overlap significantly because training repeatedly runs the inference process itself, then feeds the output back through the network to adjust weights. GPU architecture is generally better aligned with training's dual forward-and-backward passes, while specialized hardware such as TPUs is more often used for pure inference, meaning answering user queries or running agentic tasks after the model is already trained. Developers weighing GPU versus TPU choices for their AI pipeline can follow this kind of architecture explainer on daily.dev."}},{"@type":"Question","name":"Why has DDR5 RAM gotten so much more expensive recently?","acceptedAnswer":{"@type":"Answer","text":"DDR5 memory prices have spiked because manufacturers are shifting production capacity toward high-bandwidth memory (HBM) for AI accelerators like Nvidia's Blackwell architecture and TPUs, which is far more profitable than consumer memory. A 32GB DDR5 stick that cost $69 a year earlier reportedly rose to $440, driven by demand for the memory bandwidth needed to move huge tensors in and out of memory during LLM processing. Engineers budgeting hardware costs around AI-driven memory shortages can keep tabs on stories like this via daily.dev."}}]} ```