Intel 4th Gen Xeon: Difference between revisions

From HPCWIKI
Jump to navigation Jump to search
No edit summary
Line 1,025: Line 1,025:
!Purpose
!Purpose
!Notes
!Notes
!
|-
|-
| rowspan="2" |Intel® AI Engine
| rowspan="2" |Intel® AI Engine
Line 1,037: Line 1,036:


processors while also offering significant performance gains
processors while also offering significant performance gains
|
|-
|-
|Intel® Advanced Vector Extensions 512 (Intel® AVX-512
|Intel® Advanced Vector Extensions 512 (Intel® AVX-512
Line 1,048: Line 1,046:


designed to speed up data preprocessin
designed to speed up data preprocessin
|
|-
|-
| rowspan="3" |Intel® Security Engines
| rowspan="3" |Intel® Security Engines
Line 1,070: Line 1,067:


application security and protect data confidentiality
application security and protect data confidentiality
|
|-
|-
|Intel® Trust Domain Extensions (Intel® TDX)
|Intel® Trust Domain Extensions (Intel® TDX)
Line 1,081: Line 1,077:


VMs on the platform.
VMs on the platform.
|
|-
|-
|Intel® Crypto Acceleration
|Intel® Crypto Acceleration
Line 1,090: Line 1,085:


impact on performance and user experience
impact on performance and user experience
|
|-
|-
| rowspan="3" |Intel® HPC Engines
| rowspan="3" |Intel® HPC Engines
Line 1,106: Line 1,100:


analytics, 3D modeling and scientific simulations
analytics, 3D modeling and scientific simulations
|
|-
|-
|Intel® Advanced Matrix Extensions (Intel® AMX)
|Intel® Advanced Matrix Extensions (Intel® AMX)
Line 1,115: Line 1,108:


dimensional register file to store large chunks of data
dimensional register file to store large chunks of data
|
|-
|-
|Intel® Data Streaming Accelerator (Intel® DSA
|Intel® Data Streaming Accelerator (Intel® DSA
Line 1,126: Line 1,118:


that cause CPU overhead in data-center-scale deployments
that cause CPU overhead in data-center-scale deployments
|
|-
|-
| rowspan="3" |Intel® Network Engines
| rowspan="3" |Intel® Network Engines
Line 1,140: Line 1,131:


server compared to software running on CPU cores without acceleration
server compared to software running on CPU cores without acceleration
|
|-
|-
|Intel® vRAN Boost
|Intel® vRAN Boost
Line 1,149: Line 1,139:


solution complexity and provides power savings for customers
solution complexity and provides power savings for customers
|
|-
|-
|Intel® Dynamic Load Balancer (Intel® DLB
|Intel® Dynamic Load Balancer (Intel® DLB
Line 1,164: Line 1,153:


the previous generation
the previous generation
|
|-
|-
| rowspan="3" |Intel® Analytics Engines
| rowspan="3" |Intel® Analytics Engines
Line 1,184: Line 1,172:


compared to the previous generation
compared to the previous generation
|
|-
|-
|Intel® Data Streaming Accelerator (Intel® DSA)
|Intel® Data Streaming Accelerator (Intel® DSA)
Line 1,197: Line 1,184:


latencies and increases throughput, delivering faster data processing
latencies and increases throughput, delivering faster data processing
|
|-
|-
|Intel® QuickAssist Technology (Intel® QAT)
|Intel® QuickAssist Technology (Intel® QAT)
Line 1,208: Line 1,194:


and a 6% reduction in backup storage capacity
and a 6% reduction in backup storage capacity
|
|-
|-
| rowspan="3" |Intel® Storage Engines
| rowspan="3" |Intel® Storage Engines
Line 1,226: Line 1,211:


compared to the previous generation
compared to the previous generation
|
|-
|-
|Intel® QuickAssist Technology (Intel® QAT)
|Intel® QuickAssist Technology (Intel® QAT)
Line 1,237: Line 1,221:


previous generation
previous generation
|
|-
|-
|Intel® Volume Management Device (Intel® VMD)
|Intel® Volume Management Device (Intel® VMD)
Line 1,250: Line 1,233:


and enables you to deploy next-generation storage with confidence
and enables you to deploy next-generation storage with confidence
|
|}
|}
== Reference ==
<references />

Revision as of 13:00, 29 March 2023

Intel has launched its 4th Gen Intel Xeon scalable processors (Sapphire Rapids) as of Jan'23, 4th Gen Xeon is purpose-built to improve data center performance, efficiency, security, and capabilities for AI, cloud computing, networks, edge computing, and supercomputers.

섬네일|369x369픽셀The new Xeon Models are available in six categories, including the

  • Max 9400,
  • Platinum 8000 - Platinum 8400 processors are designed for advanced data analytics, AI, and hybrid cloud data centers, offering high performance, platform capabilities, and workload acceleration, as well as enhanced hardware-based security and multi-socket (Sapphire Rapids platform supports up to 8 sockets[1]) processing
  • Gold 6000/Gold 5000 - optimized for data center and multi-cloud workloads. They offer enhanced memory speed, capacity, security, and workload acceleration
  • Silver 4000 - Multi purpose essential performance, improved memory speed, and power efficiency for entry-level data center computing, networking, and storage.
  • Bronze 3000
  • Max Series.


The new Optimized Power Mode can provide up to 20 percent socket power savings with less than 5 percent performance impact for specific workloads. Innovations in air and liquid cooling can further reduce total data center energy consumption 섬네일|367x367픽셀 Each series has a range of different models that can be drilled down to their target use case:

  • Performance general purpose
  • Mainline general purpose
  • Liquid cooling general purpose
  • Single socket general purpose (“Q” Series)
  • Long-life Use (IoT) general purpose (“T” Series)
  • IMDB/Analytics/virtualization optimized (“H” Series)
  • 5G/Networking optimized (“N” Series)
  • Cloud-optimized IaaS (“P”, “V” and “M” Series)
  • Storage and hyperconverged infrastructure optimized (“S” series)
  • HPC optimized (i.e., the Intel Xeon CPU Max series)

General Improvement

PCI Express Gen5 (PCIe 5.0)

4th Gen Intel Xeon Scalable processors have up to 80 lanes of PCIe 5.0—ideal for fast networking, high-bandwidth accelerators, and high-performance storage devices. PCIe 5.0 doubles the

I/O bandwidth from PCIe 4.0,9 maintains backward compatibility, and provides foundational slots for CX

DDR5

Improve compute performance by overcoming data bottlenecks with higher memory bandwidth. DDR5 offers up to 1.5x bandwidth improvement over DDR4,10 enabling opportunities to improve performance, capacity, power efficiency, and cost. 4th Gen Intel Xeon Scalable processors offer up to 4,800 MT/s (1 DPC) or 4,400 MT/s (2 DPC) with DDR5.

CXL

Reduce compute latency in the data center and help lower total cost of ownership (TCO) with CXL 1.1 for next-generation workloads. CXL is an alternate protocol that runs across the standard PCIe physical layer and can support both standard PCIe devices and CXL devices on the same link. CXL provides a critical capability to create a unified, coherent memory space between CPUs and accelerators, and it will revolutionize how data center server architectures will be built for years to come

4th Gen Intel® Xeon® Scalable Processors list [1]

SKU CORES BASE ALL CORE TURBO (GHz) Max TURBO (GHz) CACHE (MB) TDP (Watts) Maximum Scalability DDR5 Memory Speed UPI Links Enabled Default DSA Devices Default IAA Devices Default QAT Devices Default DLB Devices Intel SGX Enclave Capacity (Per Processor) Long Life Use Intel® On Demand Capable
2S PERFORMANCE GENERAL PURPOSE
8480+ 56 2.0 3.0 3.8 105 350 2S 4800 4 1 1 1 1 512GB ✓
8470 52 2.0 3.0 3.8 105 350 2S 4800 4 1 0 0 0 512GB ✓ ✓
8468 48 2.1 3.1 3.8 105 350 2S 4800 4 1 0 0 0 512GB ✓
8460Y+ 40 2.0 2.8 3.7 105 300 2S 4800 4 1 1 1 1 128GB ✓
8462Y+ 32 2.8 3.6 4.1 60 300 2S 4800 3 1 1 1 1 128GB ✓ ✓
6448Y 32 2.1 3.0 4.1 60 225 2S 4800 3 1 0 0 0 128GB ✓ ✓
6442Y 24 2.6 3.3 4.0 60 225 2S 4800 3 1 0 0 0 128GB ✓
6444Y 16 3.6 4.0 4.1 45 270 2S 4800 3 1 0 0 0 128GB ✓
6426Y 16 2.5 3.3 4.1 37.5 185 2S 4800 3 1 0 0 0 128GB ✓ ✓
6434 8 3.7 4.1 4.1 22.5 195 2S 4800 3 1 0 0 0 128GB ✓
5415+ 8 2.9 3.6 4.1 22.5 150 2S 4400 3 1 1 1 1 128GB ✓ ✓
2S MAINLINE GENERAL PURPOSE
8452Y 36 2.0 2.8 3.2 67.5 300 2S 4800 4 1 0 0 0 128GB ✓ ✓
6438Y+ 32 2.0 2.8 4.0 60 205 2S 4800 3 1 1 1 1 128GB ✓
6430 32 2.1 2.6 3.4 60 270 2S 4400 3 1 0 0 0 128GB ✓ ✓
5420+ 28 2.0 2.7 4.1 52.5 205 2S 4400 3 1 1 1 1 128GB ✓ ✓
5418Y 24 2.0 2.8 3.8 45 185 2S 4400 3 1 0 0 0 128GB ✓ ✓
4416+ 20 2.0 2.9 3.9 37.5 165 2S 4000 2 1 1 1 1 64GB ✓ ✓
4410Y 12 2.0 2.8 3.9 30 150 2S 4000 2 1 0 0 0 64GB ✓ ✓
LIQUID COOLED GENERAL PURPOSE(-Q)
8470Q 52 2.1 3.2 3.8 105 350 2S 4800 4 1 0 0 0 512GB ✓
6458Q 32 3.1 4.0 4.0 60 350 2S 4800 3 1 0 0 0 128GB ✓
SINGLE SOCKET GENERAL PURPOSE(-U)
6414U 32 2.0 2.6 3.4 60 350 1S 4800 0 1 0 0 0 128GB ✓
5412U 24 2.1 2.9 3.9 45 185 1S 4400 0 1 0 0 0 128GB ✓
3408U 8 1.8 1.9 1.9 22.5 125 1S 4000 0 1 0 0 0 64GB ✓
LONG - LIFE USE (IOT) GENERAL PURPOSE (-T)
4410T 10 2.7 3.4 4.0 26.25 150 2S 4000 2 1 0 0 0 64GB ✓ ✓
IMDB/ANALYTICS/VIRTUALIZATION OPTIMIZED (-H) – SOCKET SCALABLE
8490H 60 1.9 2.9 3.5 112.5 350 8S 4800 4 4 4 4 4 512GB
8468H 48 2.1 3.0 3.8 105 330 8S 4800 4 4 4 4 4 512GB
8460H 40 2.2 3.1 3.8 105 330 8S 4800 4 4 4 0 0 512GB
8454H 32 2.1 2.7 3.4 82.5 270 8S 4800 4 4 4 4 4 512GB
8450H 28 2.0 2.6 3.5 75 250 8S 4800 4 4 4 0 0 512GB
8444H 16 2.9 3.2 4.0 45 270 8S 4800 4 4 4 0 0 512GB
6448H 32 2.4 3.2 4.1 60 250 4S 4800 3 1 1 2 2 512GB
6418H 24 2.1 2.9 4.0 60 185 4S 4800 3 1 1 0 0 512GB ✓
6416H 18 2.2 2.9 4.2 45 165 4S 4800 3 1 1 0 0 512GB
6434H 8 3.7 4.1 4.1 22.5 195 4S 4800 3 1 1 0 0 512GB
5G / NETWORKING OPTIMIZED (-N)
8470N 52 1.7 2.7 3.6 97.5 300 2S 4800 4 4 0 4 4 128GB ✓ ✓
8471N 52 1.8 2.8 3.6 97.5 300 1S 4800 0 4 0 4 4 128GB ✓ ✓
6438N 32 2.0 2.7 3.6 60 205 2S 4800 3 1 0 2 2 128GB ✓ ✓
6428N 32 1.8 2.5 3.8 60 185 2S 4000 3 1 0 2 2 128GB ✓ ✓
6421N 32 1.8 2.6 3.6 60 185 1S 4400 0 1 0 0 0 128GB ✓ ✓
5418N 24 1.8 2.6 3.8 45 165 2S 4000 3 1 0 2 2 128GB ✓ ✓
5411N 24 1.9 2.8 3.9 45 165 1S 4400 0 1 0 2 2 128GB ✓ ✓
CLOUD OPTIMIZED IaaS (-P) / SaaS(-V) / Media(-M)
8468V 48 2.4 2.9 3.8 97.5 330 2S 4800 3 1 1 1 1 128GB ✓
8458P 44 2.7 3.2 3.8 82.5 350 2S 4800 3 1 1 1 1 512GB ✓
8461V 48 2.2 2.8 3.7 97.5 300 1S 4800 0 1 1 1 1 128GB ✓
6438M 32 2.2 2.8 3.9 60 205 2S 4800 3 1 1 0 0 128GB ✓
STORAGE & HYPERCONVERGED INFRASTRUCTURE (HCI) OPTIMIZED (-S)
6454S 32 2.2 2.8 3.4 60 270 2S 4800 4 4 0 4 4 128GB ✓
5416S 16 2.0 2.8 4.0 30 150 2S 4400 3 1 0 2 2 128GB ✓
HPC OPTMIZED (Intel® Xeon® CPU MAX Series)
9480 56 1.9 2.6 3.5 112.5 350 2S 4800 4 4 0 0 0 128GB
9470 52 2 2.7 3.5 105 350 2S 4800 4 4 0 0 0 512GB
9468 48 2.1 2.6 3.5 105 350 2S 4800 4 4 0 0 0 128GB
9460 40 2.2 2.7 3.5 97.5 350 2S 4800 3 4 0 0 0 128GB
9462 32 2.7 3.1 3.5 75 350 2S 4800 3 4 0 0 0 128GB
Y Supports Intel Speed Select Technology – Performance Profile 2.0 (Intel SST-PP)

built-in accelerators[2]

Integrated accelerators support today’s most demanding workloads spanning AI, security, HPC, networking, analytics and storage

Name Purpose Notes
Intel® AI Engine Intel® Advanced Matrix Extensions (Intel® AMX) Intel AMX is Intel’s next-generation advancement for deep-learning training and

inference on 4th Gen Intel Xeon Scalable processors. Ideal for workloads like natural

language processing, recommendation systems and image recognition, this new

feature extends the built-in AI acceleration capabilities of previous Intel Xeon Scalable

processors while also offering significant performance gains

Intel® Advanced Vector Extensions 512 (Intel® AVX-512 A continuing feature of Intel Xeon Scalable processors, Intel AVX-512 is a general-

purpose performance-enhancing accelerator with a wide range of uses. When it

comes to AI, Intel AVX-512 can accelerate machine learning workloads for training

and inferencing. 4th Gen Intel Xeon Scalable processors with Intel AVX-512 are also

designed to speed up data preprocessin

Intel® Security Engines Intel® Software Guard Extensions (Intel® SGX)


Intel SGX is the most researched, updated and deployed confidential computing

technology in data centers on the market today. This continuing feature of Intel Xeon

Scalable processors provides the foundation for confidential computing solutions

across edge and multi-cloud infrastructures.

Intel SGX offers a hardware-based security solution that is designed to prohibit

access to protected data in use via unique application-isolation technology. By

helping protect selected code and data from inspection or modification, Intel SGX

allows developers to run sensitive data operations inside enclaves to help increase

application security and protect data confidentiality

Intel® Trust Domain Extensions (Intel® TDX) Intel TDX is a new capability available through select cloud providers in 2023 that

offers increased confidentiality at the virtual machine (VM) level, enhancing privacy

and control over your data. Within an Intel TDX confidential VM, the guest OS and

VM applications are isolated from access by the cloud host, hypervisor and other

VMs on the platform.

Intel® Crypto Acceleration Intel Crypto Acceleration uses single instruction, multiple data (SIMD) techniques to

process more encryption operations in every clock cycle. This can help increase the

total throughput of applications that require strong data encryption, with minimal

impact on performance and user experience

Intel® HPC Engines Intel® Advanced Vector Extensions 512 (Intel® AVX-512)


With ultrawide 512-bit vector operations capabilities, Intel AVX-512 is especially

suited to handle the most demanding computational tasks commonly encountered in

HPC applications. It’s used by organizations across educational, financial, enterprise,

engineering and medical industries. By enabling users to run complex workloads

on existing hardware, Intel AVX-512 accelerates performance for tasks like financial

analytics, 3D modeling and scientific simulations

Intel® Advanced Matrix Extensions (Intel® AMX) In addition to accelerating AI workloads, Intel AMX is also designed to deliver

performance gains across popular HPC workloads. This new built-in accelerator

transforms large matrix math calculations into a single operation and uses a two-

dimensional register file to store large chunks of data

Intel® Data Streaming Accelerator (Intel® DSA Intel DSA is a new feature designed to optimize and speed up streaming data

movement and transformation operations common in networking, data-

processing-intensive applications and high-performance storage. Intel DSA

accelerates HPC workloads by offloading the most common data movement tasks

that cause CPU overhead in data-center-scale deployments

Intel® Network Engines Intel® QuickAssist Technology (Intel® QAT)


Intel QAT boosts performance to meet the demands of today’s networking workloads,

helping systems serve more clients. It can deliver significant workload acceleration for

cryptography, including symmetric and asymmetric encryption and decryption.

Intel QAT using RSA4K can increase client density on an open-source NGINX web

server compared to software running on CPU cores without acceleration

Intel® vRAN Boost Intel vRAN Boost is a new feature that eliminates the need for an external accelerator

card by integrating vRAN acceleration directly into the Intel 4th Gen Xeon Scalable

processor. By reducing vRAN component requirements, it reduces overall vRAN

solution complexity and provides power savings for customers

Intel® Dynamic Load Balancer (Intel® DLB Intel DLB is a new feature that enables the efficient distribution of network

processing across multiple CPU cores. It also restores the order of networking data

packets processed simultaneously on CPU cores.

With Intel DLB, customers can gain higher performance on packet forwarding

compared to software queue management on cores without acceleration.

Additionally, applications can achieve higher packet processing performance than

the previous generation

Intel® Analytics Engines Intel® In-Memory Analytics Accelerator (Intel® IAA)


Intel IAA is designed to accelerate database and analytics performance. By increasing

query throughput and decreasing the memory footprint for in-memory databases

and advanced analytics workloads, this new feature can provide faster data

movement and improve CPU core utilization by reducing dependency on CPU cores.

Intel IAA is ideal for in-memory databases, open-source databases and data stores

like RocksDB, Redis, Cassandra and MySQL. Customers using 4th Gen Intel Xeon

Platinum 8490H with Intel IAA will gain up to 3x higher RocksDB performance

compared to the previous generation

Intel® Data Streaming Accelerator (Intel® DSA) Intel DSA is a new feature designed to optimize and speed up streaming data

movement and transformation operations common in data-intensive applications,

driving better business outcomes. By offloading tasks like data movement, data

copying and error checking, Intel DSA enables the CPU to focus on business-

critical database functions or other analytics workloads. This reduces query

latencies and increases throughput, delivering faster data processing

Intel® QuickAssist Technology (Intel® QAT) Intel QAT can be used to accelerate database backups. When QAT was enabled

as a new feature in SQL Server 2022, it delivered more efficient and higher

performance applications.

With Intel QAT, SQL Server customers achieve up to 2.3x faster backup operations

and a 6% reduction in backup storage capacity

Intel® Storage Engines Intel® Data Streaming Accelerator (Intel® DSA)


Intel DSA works on the CPU — between DRAM, caches and processor cores — and

extends across I/Os to attached memory and storage, plus networked resources.

As Intel’s next-generation direct memory access (DMA) engine, it speeds transfers

between volatile and persistent memory and supports virtualized memory and I/Os.

Customers using 4th Gen Intel Xeon Platinum 8490H with Intel DSA will gain up to

1.6x higher IOPS and up to 37% latency reduction for large packet sequential read

compared to the previous generation

Intel® QuickAssist Technology (Intel® QAT) Intel QAT increases performance of storage workloads and applications by

accelerating cryptography and data compression/decompression. Customers

using 4th Gen Intel Xeon Platinum 8490H with Intel QAT will gain up to 95%

fewer cores and 2x higher level 1 compression throughput compared to the

previous generation

Intel® Volume Management Device (Intel® VMD) Intel VMD is a legacy feature that enables direct control and management of

NVMe SSDs from the PCIe bus without the need for additional hardware adapters.

It allows for a smoother and lower-cost transition to fast NVMe storage while

limiting the downtime of critical infrastructure. With benefits like bootable RAID,

robust surprise hot-plug and blink status LED, Intel VMD increases serviceability

and enables you to deploy next-generation storage with confidence

Reference

  1. ↑ https://linustechtips.com/topic/1480633-what-is-better-amd-epyc-9004-series-or-4th-gen-intel-xeon-processors/
  2. ↑ https://www.colfax-intl.com/downloads/4th-gen-xeon-acclerators-eguide-with-4th-gen-benchmarks.pdf
Retrieved from "http://wiki.hpcmate.com/index.php?title=Intel_4th_Gen_Xeon&oldid=416"