--- title: Turing Award Winner: TPUs vs GPUs vs CPUs, Computer... description: Turing Award winner David Patterson explains the RISC vs CISC debate from the 1980s, covering why RISC won (3-4x speedup, simpler compiler optimization, energy... canonical: https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC | daily.dev og:description: Turing Award winner David Patterson explains the RISC vs CISC debate from the 1980s, covering why RISC won (3-4x speedup, simpler compiler optimization, energy... og:url: https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac og:image: https://api.daily.dev/og/posts/GZywPABac.png og:image:alt: Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC 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. # Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC **[The Developing Dev](https://daily.dev/sources/developingdev)** · 51 min read · 34 upvotes · 1 comments ## Summary Turing Award winner David Patterson explains the RISC vs CISC debate from the 1980s, covering why RISC won (3-4x speedup, simpler compiler optimization, energy efficiency), how ARM became dominant across mobile and cloud, and why CISC's sophisticated instructions were rarely used by compilers. He then explains the architectural differences between CPUs, GPUs, and TPUs: CPUs are general-purpose, GPUs evolved from graphics with multithreaded architectures and became useful for ML due to fast floating-point, and TPUs were purpose-built by Google for matrix multiplication in neural networks — achieving 30x better inference than GPUs and 80x over CPUs. Patterson also discusses Moore's Law slowing down, Dennard scaling ending around 2005, the shift to domain-specific architectures, NVIDIA's CUDA moat, and the MLPerf benchmarking effort. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did RISC architectures ultimately win out over CISC designs? RISC won because compiler algorithms improved enough to efficiently handle simple instructions and allocate registers, eliminating the need for a microcode interpreter that CISC required. Early measurements showed RISC needed 30-40% more instructions than CISC but ran them 4-5 times faster, yielding a net 3-4x speedup, while compilers rarely used the sophisticated CISC instructions architects had designed. _Engineers comparing instruction set trade-offs can find grounded architecture history on daily.dev._ ### How much faster was Google's original TPU compared to contemporary GPUs and CPUs? Google's first TPU, debuted in 2016, was roughly 30 times better at inference than the contemporary GPU and 80 times better than a CPU. It achieved this using a dedicated large matrix multiply unit, no hardware cache, and a novel narrow floating-point format called bfloat16 designed for machine learning's range needs rather than scientific precision. _Developers weighing GPU versus TPU choices for machine learning workloads follow hardware comparisons like this on daily.dev._ ### Why is it hard for startups to compete with NVIDIA in machine learning hardware benchmarks like MLPerf? Startups struggle because NVIDIA's advantage comes not just from GPU architecture but from its extensive, tailored software libraries built by a large engineering team, often called the CUDA moat. Google has built its own libraries relying more on compilers with fewer engineers, but startups lack the resources to tailor libraries the way NVIDIA does, making MLPerf results hard for them to match. _Teams evaluating AI accelerator options beyond NVIDIA can track these hardware and software trade-offs on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@trevorsuna** · 2 upvotes > Domain-specific hardware gains always come with a software-ecosystem tax. CUDA’s moat shows that tooling, libraries, debuggers, and developer familiarity can matter as much as raw architecture. Accelerators win when workloads are stable enough to justify specialization; CPUs remain the hedge against workloads that keep changing. ## Similar posts on daily.dev - [CPU vs GPU vs TPU vs QPU: The Complete 2026 Guide to Modern Processors](https://daily.dev/posts/cpu-vs-gpu-vs-tpu-vs-qpu-the-complete-2026-guide-to-modern-processors-utp5fgiot) · Medium · 0 upvotes · 0 comments --- Tags: [#architecture](https://daily.dev/tags/architecture), [#gpu](https://daily.dev/tags/gpu) [View this post on daily.dev](https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac) ```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":"Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC","url":"https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac"},"datePublished":"2026-07-13T13:07:32.585Z","dateModified":"2026-09-14T07:25:25.113Z","description":"Turing Award winner David Patterson explains the RISC vs CISC debate from the 1980s, covering why RISC won (3-4x speedup, simpler compiler optimization, energy...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4f3e74f556441a4b97ac098646389ee6?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4f3e74f556441a4b97ac098646389ee6?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The Developing Dev","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":"The Developing Dev","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/c19307d4b9a74379871c33d6f72a6b10","url":"https://daily.dev/sources/developingdev"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":34},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"architecture,gpu","timeRequired":"PT51M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The Developing Dev","item":"https://daily.dev/sources/developingdev"},{"@type":"ListItem","position":3,"name":"Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac","comment":[{"@type":"Comment","text":"Domain-specific hardware gains always come with a software-ecosystem tax. CUDA’s moat shows that tooling, libraries, debuggers, and developer familiarity can matter as much as raw architecture. Accelerators win when workloads are stable enough to justify specialization; CPUs remain the hedge against workloads that keep changing.","datePublished":"2026-07-16T01:40:11.140Z","url":"https://daily.dev/posts/GZywPABac#c-ekJcP5q2g","author":{"@type":"Person","name":"Trevor Suna","url":"https://daily.dev/trevorsuna","image":"https://media.daily.dev/image/upload/s--dZ7gXxpp--/f_auto/v1784081551/avatars/avatar_EMoP47rpuw8DNjhp6R1b6?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/turing-award-winner-tpus-vs-gpus-vs-cpus-computer-architecture-risc-vs-cisc-gzywpabac#faq","mainEntity":[{"@type":"Question","name":"Why did RISC architectures ultimately win out over CISC designs?","acceptedAnswer":{"@type":"Answer","text":"RISC won because compiler algorithms improved enough to efficiently handle simple instructions and allocate registers, eliminating the need for a microcode interpreter that CISC required. Early measurements showed RISC needed 30-40% more instructions than CISC but ran them 4-5 times faster, yielding a net 3-4x speedup, while compilers rarely used the sophisticated CISC instructions architects had designed. Engineers comparing instruction set trade-offs can find grounded architecture history on daily.dev."}},{"@type":"Question","name":"How much faster was Google's original TPU compared to contemporary GPUs and CPUs?","acceptedAnswer":{"@type":"Answer","text":"Google's first TPU, debuted in 2016, was roughly 30 times better at inference than the contemporary GPU and 80 times better than a CPU. It achieved this using a dedicated large matrix multiply unit, no hardware cache, and a novel narrow floating-point format called bfloat16 designed for machine learning's range needs rather than scientific precision. Developers weighing GPU versus TPU choices for machine learning workloads follow hardware comparisons like this on daily.dev."}},{"@type":"Question","name":"Why is it hard for startups to compete with NVIDIA in machine learning hardware benchmarks like MLPerf?","acceptedAnswer":{"@type":"Answer","text":"Startups struggle because NVIDIA's advantage comes not just from GPU architecture but from its extensive, tailored software libraries built by a large engineering team, often called the CUDA moat. Google has built its own libraries relying more on compilers with fewer engineers, but startups lack the resources to tailor libraries the way NVIDIA does, making MLPerf results hard for them to match. Teams evaluating AI accelerator options beyond NVIDIA can track these hardware and software trade-offs on daily.dev."}}]} ```