logo
منزل القضايا

Intel Diamond Rapids Xeon Gets Up to 1.6TB/s of Memory Bandwidth and 1.28GB of Cache Across 256 P-Cores

شهادة
الصين Beijing Qianxing Jietong Technology Co., Ltd. الشهادات
الصين Beijing Qianxing Jietong Technology Co., Ltd. الشهادات
زبون مراجعة
موظفو المبيعات في Beijing Qianxing Jietong Technology Co. ، Ltd محترفون وصبورون للغاية. يمكنهم تقديم الاقتباسات بسرعة. كما أن جودة المنتجات وتعبئتها جيدة جدًا. تعاوننا سلس للغاية.

—— 《Festfing DV LLC

عندما كنت أبحث عن وحدة المعالجة المركزية Intel CPU و Toshiba SSD بشكل عاجل ، أعطتني Sandy من Beijing Qianxing Jietong Technology Co.، Ltd الكثير من المساعدة وحصلت على المنتجات التي أحتاجها بسرعة. أنا حقا أقدرها.

—— كيتي ين

ساندي من بكين Qianxing Jietong Technology Co. ، Ltd هو بائع دقيق للغاية ، يمكنه تذكيرني بأخطاء التكوين في الوقت المناسب عندما أشتري خادمًا. المهندسون محترفون للغاية ويمكنهم إكمال عملية الاختبار بسرعة.

—— ستريلكين ميخائيل فلاديميروفيتش

نحن سعداء جدًا بتجربتنا في العمل مع شركة بكين تشيانشينغ جيتونغ. جودة المنتج ممتازة، والتسليم دائمًا في الموعد المحدد. فريق المبيعات لديهم محترف، صبور، ومفيد جدًا في الإجابة على جميع أسئلتنا. نحن نقدر حقًا دعمهم ونتطلع إلى شراكة طويلة الأمد. موصى به بشدة!

—— أحمد نافيد

الجودة: تجربة رائعة مع موردي. كانت ميكروتيك RB3011 مستخدمة بالفعل، لكنها كانت في حالة جيدة جدا وكل شيء يعمل بشكل مثالي. التواصل كان سريعا وسلاسة،وكل مخاوفي تمت معالجتها بسرعةمُزود موثوق به جداً

—— جيران كوليسيو

ابن دردش الآن

Intel Diamond Rapids Xeon Gets Up to 1.6TB/s of Memory Bandwidth and 1.28GB of Cache Across 256 P-Cores

November 17, 2026

Intel shared its high-performance computing roadmap for SC26, centered on Diamond Rapids, its next-generation Xeon server platform. Intel states the design can scale up to 256 performance cores, delivering as much as 1.6TB/s of memory bandwidth and a maximum 1.28GB of last-level cache. This overview was presented by Jeff McVeigh, vice president and general manager of Data Center Strategic Programs, and also covers Intel GPU D-series, HPC software stacks, silicon photonics, and quantum computing progress. Intel has not yet announced a release timeline for Diamond Rapids.


أحدث حالة شركة حول Intel Diamond Rapids Xeon Gets Up to 1.6TB/s of Memory Bandwidth and 1.28GB of Cache Across 256 P-Cores  0


Diamond Rapids CPU Architecture
Intel detailed this 18A-P design at Hot Chips, noting it combines greater compute density with improved memory bandwidth, larger cache capacity, and expanded CPU acceleration blocks. Intel Advanced Performance Extensions (APX) target single-threaded performance, while what Intel calls a unified memory fabric enables the quoted 1.6TB/s bandwidth figure. Intel has not disclosed the memory channel count, memory clock speed behind this metric, or whether the rating applies per socket.


The architecture also supports Intel AVX 10.2 and upgraded Intel Advanced Matrix Extensions (Intel AMX). Intel markets this next-gen Xeon as the only x86 server CPU with integrated matrix compute built for HPC and AI workloads. For core-count context, the highest-end AMD EPYC Venice SP7 chips we evaluated provide 256 Zen 6c cores per socket.


McVeigh framed this roadmap around hybrid systems where CPUs coordinate workflows and run parallel communication libraries, while accelerators handle AI workloads. He wrote that the future of HPC will be “open, heterogeneous and balanced.”


Intel GPU D-series and HPC Software
This air-cooled PCIe card merges AI acceleration with double-precision (FP64) calculation, supporting data formats ranging from FP4 and MXFP4 up through FP64. The Intel-branded GPU D270 carries 160GB of LPDDR5X and 1.5TB/s memory bandwidth, and partner-built variants scale to 480GB. Intel pitches this product to research facilities primarily for its large memory footprint; keeping bigger working datasets local to the card reduces data partitioning and host-to-accelerator transfers that slow scientific pipelines. The company also positions the card for persistent AI agents that write code and launch compute jobs.


أحدث حالة شركة حول Intel Diamond Rapids Xeon Gets Up to 1.6TB/s of Memory Bandwidth and 1.28GB of Cache Across 256 P-Cores  1


Intel is running two live demos of the GPU D270 — the D-series part previously codenamed Crescent Island — at its SC26 booth. The first is a seismic imaging mini-application from Veridien, ported from HIP and CUDA to SYCL in under one month. Intel reports 90% of its kernels operated reliably without Intel-specific tuning. The second demo splits an inference pipeline: D270 boards manage prefill tasks while SambaNova SN50 hardware handles decoding, with Xeon 6 and Xeon 6+ processors handling orchestration.


Intel also published benchmark comparisons against NVIDIA’s RTX PRO 6000 Blackwell Server Edition. For FP64 financial workloads, it claims up to 5.9x throughput gain on European Monte Carlo, 5.6x on European Black-Scholes, 4.6x on binomial pricing, and 1.92x on American Black-Scholes. The D270 runs at 350W, versus 600W for the NVIDIA card. On inference, Intel cites up to 29% lower latency for Drex v2.0 and 1.12x higher output token throughput for Qwen3-30B-A3B under FP8.


These results come from the vendor using pre-production silicon; we have not executed or validated these tests. Intel’s financial benchmark setup paired eight NVIDIA cards against one D270, and it did not clarify if results were normalized per card. The NVIDIA inference testing was completed on an AWS virtual instance.


On the software side, Intel says it will keep investing in Intel MPI, Fortran support, compiler optimizations, and oneAPI, so existing parallel codebases continue operating smoothly on mixed CPU-GPU systems. The team is also working on upstream integration with AI frameworks, inference runtimes and deployment platforms. Its stated objective is that standard models and toolchains can run on new hardware such as the D-series immediately upon deployment.


Optical Connectivity and Quantum Development
Intel also highlighted its silicon photonics efforts, which use light signals to transmit data over longer distances at what it describes as a fraction of the power draw of copper cabling. Intel did not share performance figures or a product schedule for this technology.


For quantum computing, Intel reaffirmed its planned acquisition of Classiq, which will merge Classiq’s automated quantum algorithm synthesis platform with Intel’s silicon spin-qubit hardware and chip manufacturing capacity. Intel expects the transaction to close in the first half of 2027, subject to regulatory approvals.


Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!

تفاصيل الاتصال
Beijing Qianxing Jietong Technology Co., Ltd.

اتصل شخص: Ms. Sandy Yang

الهاتف :: 13426366826

إرسال استفسارك مباشرة لنا (0 / 3000)