{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T06:20:51.496Z","headline":"AMD 发布 Instinct MI455X AI 加速器：台积电 2nm、3200 亿晶体管、432GB HBM4","description":"AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。其 OCP MXFP4 峰值算力达 40.26 PFLOPS，是前代的 4 倍。全新 Helios 机架架构将 72 个 GPU 显存整合为统一相干域，对标英伟达 NVL72。","url":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/","mainEntityOfPage":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/","datePublished":"2026-07-23T18:29:44.000Z","dateModified":"2026-07-23T18:29:44.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://www.ithome.com/0/980/873.htm","https://aihot.virxact.com/items/cmrxv8um402v1roxpc56792h5"],"canonicalUrl":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。 Aioga 将其归入「产品更新」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/","dateCreated":"2026-07-23T18:29:44.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"IT之家 source article","url":"https://www.ithome.com/0/980/873.htm","datePublished":"2026-07-23T18:29:44.000Z","provider":{"@type":"Organization","name":"IT之家","url":"https://www.ithome.com/0/980/873.htm"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrxv8um402v1roxpc56792h5","datePublished":"2026-07-23T18:29:44.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrxv8um402v1roxpc56792h5"}}],"aggregationSource":"IT之家（RSS）","originalPublisher":{"name":"IT之家","url":"https://www.ithome.com/0/980/873.htm"},"article":{"id":"cmrxv8um402v1roxpc56792h5","slug":"cmrxv8um402v1roxpc56792h5","url":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/","title":"AMD 发布 Instinct MI455X AI 加速器：台积电 2nm、3200 亿晶体管、432GB HBM4","title_en":"AMD 展示最强 AI 加速器 MI455X：台积电 2nm 工艺、3200 亿晶体管、432GB HBM4","summary":"AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。其 OCP MXFP4 峰值算力达 40.26 PFLOPS，是前代的 4 倍。全新 Helios 机架架构将 72 个 GPU 显存整合为统一相干域，对标英伟达 NVL72。","source":"IT之家（RSS）","sourceUrl":"https://www.ithome.com/0/980/873.htm","aiHotUrl":"https://aihot.virxact.com/items/cmrxv8um402v1roxpc56792h5","publishedAt":"2026-07-23T18:29:44.000Z","category":"产品更新","score":53,"selected":false,"articleBody":["IT之家：https://www.ithome.com/ 7 月 24 日消息，今天（7 月 24 日）在美国旧金山召开的年度盛会 Advancing AI 2026 上，AMD 董事会主席及首席执行官苏姿丰发表主题演讲， 并公布了下一代 AI 加速器 Instinct MI455X 的更多细节。","工艺方面，Instinct MI455X GPU 拥有 3200 亿个晶体管，采用先进封装技术构建。其设计包含 4 个加速器复合芯粒（XCD），通过混合键合（hybrid bonding）堆叠在 Fabric 与缓存芯粒（FCD）之上。而 2 个 FCD 均使用台积电的 CoWoS-L 技术封装，并连接 6 个 HBM4 堆和 2 个 I/O Die。IT之家附上相关图片如下：","全新 Helios 机架级架构首次将 72 个 MI455X GPU 的显存整合进一个统一的相干域，这是 AMD 对标英伟达 NVL72 的关键一步。结合 CDNA 5 中改进的张量数据搬运单元与新增的多播读取支持，AMD 专注于提升芯片数据移动效率。","苏姿丰表示通过这种 chiplet 设计，AMD 公司在 XCD 上使用台积电最先进的 2N GAA（环绕栅极）打造，而最有效地提升功耗和性能；而 FCD 和 I/O Die 部分采用台积电的 N3P 工艺制造。","CDNA 5 架构方面，AMD 现在将 CDNA 5 加速器复杂芯片（Accelerator Complex Die）的基本构建单元称为“工作组处理器”（Work Group Processor），不再叫“计算单元”（Compute Unit）。","MI455X 的 WGPs 数量与前代 MI355X 一致，维持在 256 个。架构的关键变革之一是波前（wavefront）宽度从 64 位调整至 32 位，AMD 称此举旨在改善指令延迟与寄存器压力，并提升与不同 AI tensor tile 交互的灵活性。","CDNA 5 的片上缓存层级也经过深度重构，不再使用前代大容量无限缓存，转而配备每 FCD 96MB、总计 192MB 的共享 L2 缓存。","计算性能方面，MI455X 的理论峰值算力大幅提升。针对低精度推理格式，其 OCP MXFP4 性能可达 40.26 PFLOPS，理论上比前代 MI355X 快 4 倍。","与英伟达 Rubin 对比，其 FP32 性能为 315 TFLOPS，在部分场景下明显高于英伟达的 Rubin。AMD 还改进了超越数学单元，其吞吐量比 CDNA 4 翻倍，能加速 softmax 等关键 AI 函数。","内存系统是本代产品的显著优势。MI455X 总计配备 12 堆栈 HBM4，提供 432GB 容量与 23.3 TB/s 带宽。","这远超英伟达本周详述的首款 Rubin GPU 配置，后者初代规格为 288GB HBM4 与最高 22 TB/s 带宽。在 72-GPU 的 Helios 机架规模下，AMD 方案可聚合 31.1 TB 显存，较英伟达 Vera Rubin 的 20.7 TB 总容量高出 50%。"],"articleImages":[{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/c00801f9-c6de-47c0-a1c8-ae376862eaa6.png","alt":"AMD Instinct MI455X Accelerator","afterParagraph":0,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/b43c4965f18164bd.png"},{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/0a8466cc-c3ed-432f-8142-d46ccb82856a.jpg@s_2,w_820,h_461","alt":"AMD Instinct MI455X GPU","afterParagraph":1,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/1aade708725be443.jpg"},{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/2cf63e3f-91a5-43dd-98a1-c595a5bc6c0b.jpg@s_2,w_820,h_461","alt":"AMD Instinct MI455X GPU","afterParagraph":1,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/6e1901a207a5647d.jpg"},{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/669f6397-6888-4edf-9880-00980a951ede.jpg@s_2,w_820,h_461","alt":"AMD MI455X GPU","afterParagraph":2,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/da83fb443a161b83.jpg"},{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/21e1814e-a172-4bfe-9346-2311cd19e838.jpg@s_2,w_820,h_461","alt":"AMD CDNA 5","afterParagraph":5,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/38b11bbfcbe6a941.jpg"},{"sourceUrl":"https://img.ithome.com/newsuploadfiles/2026/7/f2c54681-96c3-43d7-b12e-895013f9afad.jpg@s_2,w_820,h_461","alt":"AMD CDNA 5 memory hierarchy","afterParagraph":8,"url":"/media/articles/cmrxv8um402v1roxpc56792h5/7e138985743017b4.jpg"}],"mediaStatus":"ok","articleBodyZh":[],"translationStatus":"","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。 Aioga 将其归入「产品更新」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：产品与工具类动态的价值取决于它是否解决明确场景、能否进入工作流，以及交付、价格和数据安全是否可接受。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察产品是否开放使用、用户反馈、定价、集成能力和后续版本更新。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-07-28T06:29:10.581Z","sourceHash":"00a07aef4335189e","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["产品更新","IT之家（RSS）"],"translations":{"zh-CN":{"title":"AMD 发布 Instinct MI455X AI 加速器：台积电 2nm、3200 亿晶体管、432GB HBM4","summary":"AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。其 OCP MXFP4 峰值算力达 40.26 PFLOPS，是前代的 4 倍。全新 Helios 机架架构将 72 个 GPU 显存整合为统一相干域，对标英伟达 NVL72。","category":"产品更新","source":"IT之家","aggregationSource":"IT之家（RSS）","pageTitle":"AMD 发布 Instinct MI455X AI 加速器：台积电 2nm、3200 亿晶体管、432GB HBM4 - Aioga AI资讯","description":"AMD 在 Advancing AI 2026 上公布 Instinct MI455X，采用台积电 2nm 工艺，集成 3200 亿晶体管与 432GB HBM4 内存。其 OCP MXFP4 峰值算力达 40.26 PFLOPS，是前代的 4 倍。全新 Helios 机架架构将 72 个 GPU 显存整合为统一相干域，对标英伟达 NVL72。","url":"https://www.aioga.com/news/cmrxv8um402v1roxpc56792h5/"},"en":{"title":"AMD Releases Instinct MI455X AI Accelerator: TSMC 2nm, 32 Billion Transistors, 432GB HBM4","summary":"AMD announced the Instinct MI455X at Advancing AI 2026, using TSMC's 2nm process, integrating 320 billion transistors and 432GB of HBM4 memory. Its OCP MXFP4 peak performance reaches 40.26 PFLOPS, four times that of the previous generation. The new Helios rack architecture integrates the memory of 72 GPUs into a unified coherent domain, comparable to NVIDIA's NVL72.","category":"Products","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD Releases Instinct MI455X AI Accelerator: TSMC 2nm, 32 Billion Transistors, 432GB HBM4 - Aioga AI News","description":"AMD announced the Instinct MI455X at Advancing AI 2026, using TSMC's 2nm process, integrating 320 billion transistors and 432GB of HBM4 memory. Its OCP MXFP4 peak performance reach...","url":"https://www.aioga.com/en/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:01:49.937Z"},"ja":{"title":"AMD、Instinct MI455X AIアクセラレーターを発表：TSMC 2nm、3200億トランジスタ、432GB HBM4","summary":"AMD は Advancing AI 2026 で Instinct MI455X を発表し、TSMC の 2nm プロセスを採用し、3,200 億個のトランジスタと 432GB の HBM4 メモリを統合しました。その OCP MXFP4 のピーク演算性能は 40.26 PFLOPS で、前世代の 4 倍です。新しい Helios ラックアーキテクチャは 72 GPU のビデオメモリを統合して単一の整合ドメインにまとめ、NVIDIA の NVL72 に相当します。","category":"製品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD、Instinct MI455X AIアクセラレーターを発表：TSMC 2nm、3200億トランジスタ、432GB HBM4 - Aioga AIニュース","description":"AMD は Advancing AI 2026 で Instinct MI455X を発表し、TSMC の 2nm プロセスを採用し、3,200 億個のトランジスタと 432GB の HBM4 メモリを統合しました。その OCP MXFP4 のピーク演算性能は 40.26 PFLOPS で、前世代の 4 倍です。新しい Helios ラックアーキテクチャは...","url":"https://www.aioga.com/ja/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:02:06.521Z"},"ko":{"title":"AMD, Instinct MI455X AI 가속기 출시: TSMC 2nm, 3200억 트랜지스터, 432GB HBM4","summary":"AMD는 Advancing AI 2026에서 Instinct MI455X를 발표했으며, TSMC 2nm 공정을 사용하고 3,200억 개의 트랜지스터와 432GB HBM4 메모리를 통합했습니다. OCP MXFP4의 최대 연산 성능은 40.26 PFLOPS로 이전 세대의 4배입니다. 새로운 Helios 랙 아키텍처는 72개의 GPU 메모리를 통합하여 단일 일관 성 영역으로 구성하며, NVIDIA NVL72와 대응합니다.","category":"제품 업데이트","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD, Instinct MI455X AI 가속기 출시: TSMC 2nm, 3200억 트랜지스터, 432GB HBM4 - Aioga AI 뉴스","description":"AMD는 Advancing AI 2026에서 Instinct MI455X를 발표했으며, TSMC 2nm 공정을 사용하고 3,200억 개의 트랜지스터와 432GB HBM4 메모리를 통합했습니다. OCP MXFP4의 최대 연산 성능은 40.26 PFLOPS로 이전 세대의 4배입니다. 새로운 Helios 랙 아키텍처는 72개의...","url":"https://www.aioga.com/ko/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:02:46.868Z"},"es":{"title":"AMD lanza el acelerador de IA Instinct MI455X: TSMC 2nm, 32 mil millones de transistores, 432GB HBM4","summary":"AMD anunció el Instinct MI455X en Advancing AI 2026, utilizando el proceso de 2 nm de TSMC, integrando 320 mil millones de transistores y 432 GB de memoria HBM4. Su potencia de cálculo máxima OCP MXFP4 alcanza 40,26 PFLOPS, cuatro veces la de la generación anterior. La nueva arquitectura de chasis Helios integra la memoria de 72 GPU en un dominio coherente unificado, en comparación con la NVL72 de NVIDIA.","category":"Productos","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD lanza el acelerador de IA Instinct MI455X: TSMC 2nm, 32 mil millones de transistores, 432GB HBM4 - Aioga Noticias de IA","description":"AMD anunció el Instinct MI455X en Advancing AI 2026, utilizando el proceso de 2 nm de TSMC, integrando 320 mil millones de transistores y 432 GB de memoria HBM4. Su potencia de cál...","url":"https://www.aioga.com/es/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:02:43.501Z"},"fr":{"title":"AMD lance l'accélérateur AI Instinct MI455X : TSMC 2nm, 320 milliards de transistors, 432 Go de HBM4","summary":"AMD a annoncé l'Instinct MI455X lors de l'Advancing AI 2026, utilisant le procédé 2 nm de TSMC, intégrant 320 milliards de transistors et 432 Go de mémoire HBM4. Sa puissance de calcul maximale OCP MXFP4 atteint 40,26 PFLOPS, soit quatre fois celle de la génération précédente. La toute nouvelle architecture de cadre Helios intègre 72 GPU avec mémoire en un domaine cohérent unifié, concurrençant le NVL72 de NVIDIA.","category":"Produits","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD lance l'accélérateur AI Instinct MI455X : TSMC 2nm, 320 milliards de transistors, 432 Go de HBM4 - Aioga Actualités IA","description":"AMD a annoncé l'Instinct MI455X lors de l'Advancing AI 2026, utilisant le procédé 2 nm de TSMC, intégrant 320 milliards de transistors et 432 Go de mémoire HBM4. Sa puissance de ca...","url":"https://www.aioga.com/fr/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:03:24.855Z"},"de":{"title":"AMD veröffentlicht Instinct MI455X AI-Beschleuniger: TSMC 2nm, 320 Milliarden Transistoren, 432GB HBM4","summary":"AMD hat auf der Advancing AI 2026 die Instinct MI455X vorgestellt, die im 2-nm-Prozess von TSMC gefertigt ist, 320 Milliarden Transistoren und 432 GB HBM4-Speicher integriert. Ihre OCP MXFP4-Spitzenleistung erreicht 40,26 PFLOPS, was das Vierfache der Vorgängergeneration ist. Die neue Helios-Rack-Architektur integriert 72 GPU-Speicher zu einem einheitlichen kohärenten Bereich und konkurriert mit Nvidias NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD veröffentlicht Instinct MI455X AI-Beschleuniger: TSMC 2nm, 320 Milliarden Transistoren, 432GB HBM4 - Aioga KI-News","description":"AMD hat auf der Advancing AI 2026 die Instinct MI455X vorgestellt, die im 2-nm-Prozess von TSMC gefertigt ist, 320 Milliarden Transistoren und 432 GB HBM4-Speicher integriert. Ihre...","url":"https://www.aioga.com/de/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:03:22.605Z"},"pt-BR":{"title":"AMD lança acelerador de IA Instinct MI455X: TSMC 2nm, 320 bilhões de transistores, 432GB HBM4","summary":"A AMD anunciou o Instinct MI455X no Advancing AI 2026, utilizando o processo de 2nm da TSMC, com 32 bilhões de transistores e 432GB de memória HBM4. Seu poder de computação máximo OCP MXFP4 chega a 40,26 PFLOPS, quatro vezes o da geração anterior. A nova arquitetura de rack Helios integra 72 GPUs em uma memória unificada e coerente, competindo com a NVL72 da NVIDIA.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD lança acelerador de IA Instinct MI455X: TSMC 2nm, 320 bilhões de transistores, 432GB HBM4 - Aioga Notícias de IA","description":"A AMD anunciou o Instinct MI455X no Advancing AI 2026, utilizando o processo de 2nm da TSMC, com 32 bilhões de transistores e 432GB de memória HBM4. Seu poder de computação máximo...","url":"https://www.aioga.com/pt-BR/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:04:03.780Z"},"ru":{"title":"AMD выпустила ускоритель AI Instinct MI455X: TSMC 2 нм, 320 миллиардов транзисторов, 432 ГБ HBM4","summary":"AMD на конференции Advancing AI 2026 представила Instinct MI455X, изготовленный по 2-нм технологии TSMC, с интеграцией 320 миллиардов транзисторов и 432 ГБ памяти HBM4. Его пиковая производительность по стандарту OCP MXFP4 достигает 40,26 PFLOPS, что в 4 раза выше предыдущей модели. Новая архитектура стоек Helios объединяет 72 GPU памяти в единое когерентное пространство, конкурируя с NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD выпустила ускоритель AI Instinct MI455X: TSMC 2 нм, 320 миллиардов транзисторов, 432 ГБ HBM4 - Aioga Новости ИИ","description":"AMD на конференции Advancing AI 2026 представила Instinct MI455X, изготовленный по 2-нм технологии TSMC, с интеграцией 320 миллиардов транзисторов и 432 ГБ памяти HBM4. Его пиковая...","url":"https://www.aioga.com/ru/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:04:03.259Z"},"ar":{"title":"أصدرت AMD مُسرِّع AI Instinct MI455X: تصنيع TSMC بتقنية 2nm، و320 مليار ترانزستور، و432GB HBM4","summary":"أعلنت شركة AMD في مؤتمر Advancing AI 2026 عن معالج Instinct MI455X، الذي يستخدم تقنية TSMC بحجم 2 نانومتر، ويضم 320 مليار ترانزستور وذاكرة HBM4 بسعة 432 جيجابايت. تصل قدرة الحوسبة القصوى له في OCP MXFP4 إلى 40.26 بيتافلوب، وهو أربعة أضعاف الجيل السابق. كما أن البنية الجديدة للخادم Helios تدمج ذاكرة 72 وحدة GPU لتشكل نطاقًا متسقًا واحدًا، معادلةً بذلك NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"أصدرت AMD مُسرِّع AI Instinct MI455X: تصنيع TSMC بتقنية 2nm، و320 مليار ترانزستور، و432GB HBM4 - Aioga أخبار الذكاء الاصطناعي","description":"أعلنت شركة AMD في مؤتمر Advancing AI 2026 عن معالج Instinct MI455X، الذي يستخدم تقنية TSMC بحجم 2 نانومتر، ويضم 320 مليار ترانزستور وذاكرة HBM4 بسعة 432 جيجابايت. تصل قدرة الحوسبة...","url":"https://www.aioga.com/ar/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:04:44.989Z"},"hi":{"title":"AMD ने Instinct MI455X AI एक्सेलेरेटर लॉन्च किया: TSMC 2nm, 3200 करोड़ ट्रांजिस्टर, 432GB HBM4","summary":"AMD ने Advancing AI 2026 में Instinct MI455X की घोषणा की, जो TSMC के 2nm प्रक्रिया का उपयोग करता है, इसमें 3200 अरब ट्रांजिस्टर और 432GB HBM4 मेमोरी एकीकृत है। इसका OCP MXFP4 सर्वोच्च गणना शक्ति 40.26 PFLOPS तक जाती है, जो पिछले संस्करण का 4 गुना है। नया Helios रैक आर्किटेक्चर 72 GPU मेमोरी को एक एकीकृत सुसंगत डोमेन में जोड़ता है, और यह NVIDIA NVL72 के बराबर है।","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD ने Instinct MI455X AI एक्सेलेरेटर लॉन्च किया: TSMC 2nm, 3200 करोड़ ट्रांजिस्टर, 432GB HBM4 - Aioga AI समाचार","description":"AMD ने Advancing AI 2026 में Instinct MI455X की घोषणा की, जो TSMC के 2nm प्रक्रिया का उपयोग करता है, इसमें 3200 अरब ट्रांजिस्टर और 432GB HBM4 मेमोरी एकीकृत है। इसका OCP MXFP4 सर्वो...","url":"https://www.aioga.com/hi/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:04:55.014Z"},"it":{"title":"AMD lancia l'acceleratore AI Instinct MI455X: TSMC 2nm, 320 miliardi di transistor, 432GB HBM4","summary":"AMD ha annunciato Instinct MI455X all'Advancing AI 2026, utilizzando il processo a 2nm di TSMC, integrando 320 miliardi di transistor e 432GB di memoria HBM4. La sua potenza di calcolo massima OCP MXFP4 raggiunge i 40,26 PFLOPS, quattro volte quella della generazione precedente. La nuova architettura a rack Helios integra la memoria di 72 GPU in un dominio coerente unificato, in concorrenza con la NVL72 di NVIDIA.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD lancia l'acceleratore AI Instinct MI455X: TSMC 2nm, 320 miliardi di transistor, 432GB HBM4 - Aioga Notizie IA","description":"AMD ha annunciato Instinct MI455X all'Advancing AI 2026, utilizzando il processo a 2nm di TSMC, integrando 320 miliardi di transistor e 432GB di memoria HBM4. La sua potenza di cal...","url":"https://www.aioga.com/it/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:05:32.953Z"},"nl":{"title":"AMD lanceert Instinct MI455X AI-versneller: TSMC 2nm, 320 miljard transistors, 432GB HBM4","summary":"AMD kondigt op Advancing AI 2026 de Instinct MI455X aan, geproduceerd met TSMC 2nm-technologie, geïntegreerd met 320 miljard transistoren en 432GB HBM4-geheugen. Het OCP MXFP4 piekvermogen bereikt 40,26 PFLOPS, vier keer dat van de vorige generatie. De gloednieuwe Helios-rackarchitectuur integreert het geheugen van 72 GPU's in één uniforme coherente domein en is vergelijkbaar met de NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD lanceert Instinct MI455X AI-versneller: TSMC 2nm, 320 miljard transistors, 432GB HBM4 - Aioga AI-nieuws","description":"AMD kondigt op Advancing AI 2026 de Instinct MI455X aan, geproduceerd met TSMC 2nm-technologie, geïntegreerd met 320 miljard transistoren en 432GB HBM4-geheugen. Het OCP MXFP4 piek...","url":"https://www.aioga.com/nl/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:05:33.448Z"},"tr":{"title":"AMD, Instinct MI455X AI hızlandırıcısını duyurdu: TSMC 2nm, 320 milyar transistör, 432GB HBM4","summary":"AMD, Advancing AI 2026 etkinliğinde Instinct MI455X'i duyurdu; bu ürün, TSMC'nin 2nm süreci kullanılarak üretilmiş olup, 320 milyar transistör ve 432GB HBM4 belleği entegre ediyor. OCP MXFP4 ile doruk hesaplama gücü 40,26 PFLOPS'a ulaşıyor ve bu, selefinin dört katı. Yeni Helios raf mimarisi, 72 GPU belleğini tek bir uyumlu alan altında birleştiriyor ve Nvidia NVL72 ile kıyaslanıyor.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD, Instinct MI455X AI hızlandırıcısını duyurdu: TSMC 2nm, 320 milyar transistör, 432GB HBM4 - Aioga AI Haberleri","description":"AMD, Advancing AI 2026 etkinliğinde Instinct MI455X'i duyurdu; bu ürün, TSMC'nin 2nm süreci kullanılarak üretilmiş olup, 320 milyar transistör ve 432GB HBM4 belleği entegre ediyor....","url":"https://www.aioga.com/tr/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:06:14.581Z"},"vi":{"title":"AMD ra mắt bộ tăng tốc AI Instinct MI455X: TSMC 2nm, 320 tỷ bóng bán dẫn, 432GB HBM4","summary":"AMD đã công bố Instinct MI455X tại Advancing AI 2026, sử dụng công nghệ 2nm của TSMC, tích hợp 320 tỷ bóng bán dẫn và 432GB bộ nhớ HBM4. Hiệu năng đỉnh OCP MXFP4 đạt 40,26 PFLOPS, gấp 4 lần thế hệ trước. Kiến trúc giá đỡ Helios hoàn toàn mới hợp nhất 72 GPU vào một miền nhất quán, đối đầu với NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD ra mắt bộ tăng tốc AI Instinct MI455X: TSMC 2nm, 320 tỷ bóng bán dẫn, 432GB HBM4 - Tin tức AI Aioga","description":"AMD đã công bố Instinct MI455X tại Advancing AI 2026, sử dụng công nghệ 2nm của TSMC, tích hợp 320 tỷ bóng bán dẫn và 432GB bộ nhớ HBM4. Hiệu năng đỉnh OCP MXFP4 đạt 40,26 PFLOPS,...","url":"https://www.aioga.com/vi/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:06:14.148Z"},"id":{"title":"AMD merilis akselerator AI Instinct MI455X: TSMC 2nm, 320 miliar transistor, 432GB HBM4","summary":"AMD mengumumkan Instinct MI455X di Advancing AI 2026, menggunakan proses 2nm TSMC, mengintegrasikan 320 miliar transistor dan memori 432GB HBM4. Peak computing power OCP MXFP4 mencapai 40,26 PFLOPS, empat kali lipat dari generasi sebelumnya. Arsitektur rak Helios baru menggabungkan 72 GPU memory menjadi satu domain yang kohesif, bersaing dengan NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD merilis akselerator AI Instinct MI455X: TSMC 2nm, 320 miliar transistor, 432GB HBM4 - Berita AI Aioga","description":"AMD mengumumkan Instinct MI455X di Advancing AI 2026, menggunakan proses 2nm TSMC, mengintegrasikan 320 miliar transistor dan memori 432GB HBM4. Peak computing power OCP MXFP4 menc...","url":"https://www.aioga.com/id/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:06:51.712Z"},"th":{"title":"AMD เปิดตัวตัวเร่ง AI Instinct MI455X: TSMC 2nm, 320 พันล้านทรานซิสเตอร์, 432GB HBM4","summary":"AMD ประกาศ Instinct MI455X ในงาน Advancing AI 2026 ใช้เทคโนโลยี 2nm ของ TSMC รวมทรานซิสเตอร์ 320 พันล้านตัวและหน่วยความจำ HBM4 ขนาด 432GB กำลังประมวลผลสูงสุด OCP MXFP4 ของมันถึง 40.26 PFLOPS เป็น 4 เท่าของรุ่นก่อน สถาปัตยกรรมชั้นวาง Helios ใหม่รวมหน่วยความจำ GPU 72 ตัวเป็นโดเมนสอดคล้องเดียว เปรียบเทียบกับ NVL72 ของ NVIDIA","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD เปิดตัวตัวเร่ง AI Instinct MI455X: TSMC 2nm, 320 พันล้านทรานซิสเตอร์, 432GB HBM4 - ข่าว AI Aioga","description":"AMD ประกาศ Instinct MI455X ในงาน Advancing AI 2026 ใช้เทคโนโลยี 2nm ของ TSMC รวมทรานซิสเตอร์ 320 พันล้านตัวและหน่วยความจำ HBM4 ขนาด 432GB กำลังประมวลผลสูงสุด OCP MXFP4 ของมันถึง 40...","url":"https://www.aioga.com/th/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:07:00.393Z"},"pl":{"title":"AMD wprowadza akcelerator AI Instinct MI455X: TSMC 2nm, 320 miliardów tranzystorów, 432GB HBM4","summary":"AMD ogłosiło Instinct MI455X na Advancing AI 2026, wykorzystujący proces technologiczny 2 nm TSMC, integrujący 320 miliardów tranzystorów oraz 432 GB pamięci HBM4. Jego szczytowa wydajność obliczeniowa OCP MXFP4 osiąga 40,26 PFLOPS, co jest czterokrotnością poprzedniej generacji. Nowa architektura szaf Helios integruje 72 GPU z pamięcią w jeden spójny obszar, konkurując z NVIDIA NVL72.","category":"产品更新","source":"IT之家（RSS）","aggregationSource":"IT之家（RSS）","pageTitle":"AMD wprowadza akcelerator AI Instinct MI455X: TSMC 2nm, 320 miliardów tranzystorów, 432GB HBM4 - Aioga Wiadomości AI","description":"AMD ogłosiło Instinct MI455X na Advancing AI 2026, wykorzystujący proces technologiczny 2 nm TSMC, integrujący 320 miliardów tranzystorów oraz 432 GB pamięci HBM4. Jego szczytowa w...","url":"https://www.aioga.com/pl/news/cmrxv8um402v1roxpc56792h5/","contentTranslated":true,"sourceHash":"55aca367a929542b","translatedAt":"2026-07-26T21:07:48.741Z"}}}}