英特尔纯小核至强Sierra Forrest Xeon 6 CPU参数
Intel Launches 144-core 'Sierra Forrest' Xeon 6 CPUs, Granite Rapids Follows in Q3
By Paul Alcorn published June 5, 2024
Pumping up the core counts.
(Image credit: Tom's Hardware)
Intel CEO Pat Geisinger announced the launch of the company's Xeon 6 'Sierra Forest' processors during his Computex 2024 keynote here in Taipei, Taiwan. The Xeon 6 lineup is comprised of two families. The E-core-powered Sierra Forest models with up to 144 cores are launching today, but 288-core models will follow next year. Intel's efficiency-focused 288-core chips will face off with AMD's looming density-focused 192-core EPYC Turin models when they come to market later this year. The flagship 6900P P-core Granite Rapids models, which can have up to 128 P-cores, will launch in Q3 of 2024. Early next year, Intel will offer Granite Rapids in models with lesser core counts. These models will face off with AMD's performance-focused Turin models that will arrive later this year with up to 128 cores. Intel will also expand its Xeon 6 roster further in Q1 of 2025 with network—and edge-optimized variants. Image 3 of 3

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)
Intel's Sierra Forest chips are the first to use the new Intel 3 process node, conferring power and performance benefits. The chips are geared for web and scale-out containerized microservices environments, networking, content delivery networks, cloud services, and AI workloads. Sierra Forrest is designed to provide the utmost performance consistency and have an extreme focus on power efficiency, accomplished with the Intel 3 process node and Crestmont microarchitecture.
The P-core-powered Granite Ridge chips are designed for latency-sensitive work and high single-core performance, making them a good fit for HPC, AI, and general workloads, among many other workloads outlined in the album above.
Image 4 of 4

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)
Both processors will drop into the Birch Stream platform, but Intel carves its P—and E-core offerings into two swim lanes. The Xeon 6700 series has up to either 144 E-cores or 86 P-cores and will slot into standard SP platforms that support up to 350W per CPU. Meanwhile, the 6900 series models with up to 288 E-cores or 128 P-cores will require AP platforms that support up to 500W per CPU.
The 6700E models come equipped with E-cores, as designated by the E suffix, and come in seven different models spanning from 64 cores to 144 cores. These chips have what we would typically consider to be low max turbo clocks that range from 2.6 GHz to 3.2 GHz, but this is optimum for this class of chip and helps boost performance efficiency, a key tenant of the Sierra Forest processors. Intel hasn't shared specs for the full roster of 6700P processors yet.
Image 8 of 11

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)
Despite using E-cores, Intel positions the 6700E models against AMD's P-core-powered EPYC processors. AMD's density-focused Bergamo platform comes with 12 memory channels, whereas the 6700E series supports eight memory channels to reduce cost. Intel shared a series of benchmarks to tout its performance advantages in several workloads, with many of the comparisons focusing on the generational performance improvements that can be expected by upgrading to newer models. As always, take vendor-provided benchmarks with some salt. Intel isn't shying away from third-party benchmarking — we have a server in our labs, so you can expect a full review with performance benchmarking soon.
Intel's radical improvements in power consumption are a key focus point, so the company compared Xeon 6's power efficiency against AMD's EPYC across a wide range of general compute, database, web, and media workloads. Depending on the workload, Xeon 6's power efficiency advantages ranged from 7% to 34%. Intel also highlighted performance disclosures from its partners, including eBay's findings of a 25% performance-per-watt improvement for its use case against a competitive solution, which would, of course, imply an EPYC processor.
Other highlights of Intel's performance claims include up to a 2x improvement in AI, a 2.3x improvement in HPC, and up to twice the general compute performance for its users updating over the prior Fifth-Gen Xeons. Naturally, those deltas become much larger when compared to Intel's five-year-old second-gen models, but this is the typical upgrade cadence for most server gear. In media transcode workloads, Intel claims a 2.6x improvement in performance-per-watt and a 4.2x improvement in overall performance over second-gen Xeons.
Intel also touts strong rack consolidation opportunities, saying that the 6700E series offers a 3:1 ratio over second-gen Xeon. That translates to replacing 200 racks of second-gen Xeon with only 66 racks of the 6700E series.
Image 21 of 23

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)

(Image credit: Tom's Hardware)
We've covered the Sierra Forrest architecture in depth here. The chips come in four different package types, two of which are designed for SP systems and the other two specifically for AP servers.
All models support DDR5-6400 in either eight- or 12-channel configurations, including MCR DIMM support for the P-core models, boosting throughput up to DDR5-8800. The chips also support either 88 or 136 PCIe 5.0 lanes, but this varies based on the number of sockets in the server. The standard SP servers support from 1S to 8S systems, while the AP platforms only scale to two sockets. Intel also offers full support for CXL 2.0, including support for Type 3 devices.
The E-core models launching today use the Crestmont architecture, while the P-core models leverage Redwood Cove. The chips are socket-compatible, but due to the differences between the core microarchitectures, the E-core models don't support AVX-512 or AMX. Intel also offers a full suite of security features and continues to offer its in-built QAT, DLB, DSA, and IAA accelerators.
Many of Intel's partners already have Xeon 6700E processors working in the field, but the company is now shipping to the broader market. We'll provide the full architectural breakdown and performance data in our upcoming review; stay tuned.
附DEEPL翻译:
英特尔首席执行官帕特·基辛格在台北举行的Computex 2024主题演讲中宣布推出至强6代“塞拉森林”处理器。该系列包含两大产品线。搭载E核的Sierra Forest型号今日首发,最高支持144核,而288核型号将于明年推出。英特尔这款注重能效的288核芯片,将在今年晚些时候上市时与AMD即将推出的高密度192核EPYC Turin型号展开竞争。
旗舰级6900P P核Granite Rapids型号(最高128个P核)将于2024年第三季度上市。明年初英特尔将推出核心数较低的Granite Rapids型号,这些型号将与AMD今年晚些时候推出的高性能Turin型号(最高128核)展开竞争。英特尔还计划在2025年第一季度进一步扩展至强6系列产品线,推出网络优化版与边缘计算优化版。
英特尔Sierra Forest芯片是首批采用全新英特尔3工艺节点的芯片,可带来功耗与性能优势。该系列芯片专为网络及横向扩展的容器化微服务环境、网络架构、内容分发网络、云服务和人工智能工作负载而设计。塞拉森林芯片通过英特尔3工艺节点与Crestmont微架构,致力于实现极致性能稳定性与卓越能效表现。
基于P核心的Granite Ridge芯片专为低延迟任务与高单核性能设计,适用于高性能计算、人工智能及通用工作负载等多种场景(详见上文图集)。
这两款处理器都将搭载于Birch Stream平台,但英特尔将其P核与E核产品划分为两条产品线。至强6700系列最高可配备144个E核或86个P核,适用于支持单CPU功耗高达350W的标准SP平台。而6900系列型号最高可配备288个E核或128个P核,需采用支持单CPU功耗高达500W的AP平台。
6700E型号如其E后缀所示均搭载E核,共推出七款型号,核心数从64核到144核不等。这些芯片的最大睿频通常被视为较低水平(2.6GHz至3.2GHz),但此设计正是该级别芯片的优化方案,有助于提升性能效率——这正是Sierra Forest处理器的核心理念。英特尔尚未公布完整的6700P处理器系列规格。
尽管采用E核架构,英特尔仍将6700E系列定位为AMD基于P核的EPYC处理器的竞争对手。AMD专注于密度优化的Bergamo平台配备12条内存通道,而6700E系列为降低成本仅支持8条内存通道。英特尔发布了一系列基准测试数据,宣称其在多种工作负载中具备性能优势,其中多数对比聚焦于升级至新款型号可获得的代际性能提升。一如既往,对厂商提供的基准测试数据需持保留态度。英特尔并未回避第三方测试——我们实验室已配备测试服务器,敬请期待即将发布的完整性能评测报告。
英特尔在功耗方面的突破性改进是关键焦点,因此该公司将至强6系列的能效与AMD EPYC处理器在广泛的一般计算、数据库、网络和媒体工作负载中进行了对比。根据不同工作负载,至强6系列的能效优势范围在7%至34%之间。英特尔还强调了合作伙伴的性能数据披露,包括eBay在实际应用场景中发现相较于竞争方案(显然指EPYC处理器)实现了25%的每瓦性能提升。
英特尔性能宣称的其他亮点包括:相较于第五代至强处理器,其AI性能最高提升2倍,HPC性能提升2.3倍,通用计算性能最高翻倍。当然,若与英特尔五年前的第二代型号相比,这些差距会更大——但这正是大多数服务器设备的典型升级节奏。在媒体转码工作负载方面,英特尔宣称相较第二代至强处理器,其性能功耗比提升2.6倍,整体性能提升4.2倍。
英特尔同时强调强大的机架整合能力,称6700E系列可实现3:1的整合率(相较第二代至强处理器)。这意味着仅需66个机架的6700E系列即可替代200个机架的第二代至强设备。
我们在此深入探讨了Sierra Forrest架构。该系列芯片提供四种封装类型,其中两种专为SP系统设计,另两种则专用于AP服务器。所有型号均支持八通道或十二通道的DDR5-6400内存配置,其中P-core型号还支持MCR DIMM技术,可将吞吐量提升至DDR5-8800。芯片还支持88或136条PCIe 5.0通道,具体数量取决于服务器插槽数量。标准SP服务器支持1S至8S系统,而AP平台仅支持双插槽扩展。英特尔同时全面支持CXL 2.0规范,包括对Type 3设备的兼容性。
今日发布的E核型号采用Crestmont架构,P核型号则基于Redwood Cove架构。两者插槽兼容,但因核心微架构差异,E核型号不支持AVX-512或AMX指令集。英特尔同时提供完整的安全特性套件,并延续其内置的QAT、DLB、DSA和IAA加速器。
众多合作伙伴已将至强6700E处理器投入实际应用,现正向更广泛市场供货。我们将在后续评测中提供完整的架构解析与性能数据,敬请期待。
作者声明本文无利益相关,欢迎值友理性交流,和谐讨论~
