AMD EPYC formerly Opteron

2022: Vulnerability in AMD processors leading to data leakage through third-party channels

A group of researchers from Graz Technical University Austria (), previously known for developing MDS attacks , NetSpecter, Throwhammer, and ZombieLoad, revealed information about the method of third-party channel attack (CVE-2021-46778) on processor scheduler queues AMD used to schedule instruction execution in different execution blocks. CPU This became known on August 12, 2022. The attack, called SQUIP, allows you to determine what is used in computing in another process or virtual machine data or organize a hidden channel between communications processes or machines that virtual allows you to exchange data bypassing the system mechanisms for delimiting access. More. here

2020: Use in IBM Cloud Infrastructure

AMD announced that IBM Cloud is expanding its global infrastructure with second-generation AMD EPYC processors that will enable open source servers. This became known on April 2, 2020. Read more here.

2019: AMD compares Epyc to Xeon and talks about Intel underperformance

In late March 2019, AMD posted materials on its website on the benefits of its Epyc server processors over Xeon. The company reported that Intel underestimates the performance of its solutions so that customers buy more expensive products.

AMD says that companies buying Xeon processors are forced to pay the so-called "Intel tax." It represents "the mark-up on Intel processors you have to pay to get the functionality and performance you need."

AMD says its Epyc server chips are faster and cheaper than Xeon

These chips are stuffed with imposed built-in performance constraints that affect real-world performance. You have to buy a more expensive processor to get the performance you need, AMD said in a statement.

Customers are forced to buy a new Intel Xeon processor just to add more RAM for cloud and virtual workloads, AMD continues. Customers must pay a lot for Gold and Platinum processors to maximize memory performance.

In addition, according to AMD representatives, in order to get more than 28 cores in the system, Intel forces users to buy a different processor, and the same applies when more I/O capabilities are required. All these shortcomings are not in Epyc chips, they are cheaper and faster, AMD said.

This document compares prices and performance between Intel Xeon processors and AMD Epyc processors, showing the benefits of the latter. However, the white paper only compares using a single test - SPECrate2017_fp_base, which is a standard computationally intensive performance test. The document also does not consider a wider sample of Intel Xeon processors for comparison.

AMD compared Epyc and Xeon processors

Intel has long been the undisputed leader in the server processor market, but AMD, which actually returned to this market in 2017 after the release of the Epyc series, is actively working to strengthen its position. There are already fruits.

According to a study conducted by the Spiceworks IT portal in March 2019, 16% of companies choose AMD server processors, and in the next two years this figure may grow to 21%. More than 500 IT executives in different countries were interviewed.^[1]

According to AMD CEO Lisa Su, the company met the goal of increasing its market share of server chips to 4-6% by the end of 2018. Industry sources of DigiTimes expect AMD to exceed 10% by the end of 2020 against 3.2% in the fourth quarter of 2018.

AMD released an Intel-critical document days before the start of the Intel Data-Centric Innovation Summit 2019, which is expected to announce several Intel server products, including new representatives of the Intel family.

Andrew Kretzer, director of sales and marketing at Bold Data Technology (selling computers assembled according to individual configurations), agreed with AMD's arguments. In particular, he supported a point in which AMD points out the complexity of Intel's range. AMD compared it to "a maze in which 42 different processors and five different series were lost."

At Intel, you can quickly find what you buy and what you need instead of getting what you need, AMD said.

According to Andrew Kretzer, Intel not only artificially complicates the product portfolio and sells them at very high prices, but also greatly complicates the process of buying processors for potential customers.

We've had to create a dedicated tool to help customers understand what they're buying and what they need when it comes to Intel Xeon processors. This tool is becoming more complex with each new Xeon family, "Kretzer said.

However, in his view, excessive complexity is not necessarily an advantage for AMD, as companies often prefer to buy older Xeon products such as Broadwell.

Earlier, AMD has already been attacked, with the help of which the chipmaker is trying to challenge Intel's dominance in the server processor market. In July 2018, AMD posted advertisements around San Jose International Airport that read: "Xeon is not Epyc. Xeon steers... just like dinosaurs. "

Intel tax: Company undercuts processor performance so customers buy more expensive

The giants of the processor market do not miss the opportunity to prick competitors. So, the CEO Mobileye (part of Intel) Amnon Shashua (Amnon Shashua) in March 2019 published a blog post in which he criticized Nvidia the Safety Force Field (SFF) unmanned control system for copying Mobileye Responsibility Sensitive Safety (RSS) technology.

In our opinion, SFF is just a second-rate version of our RSS, painted green and black. If there are any innovations in this development, then only in the fact that it has a different name, - wrote Shashua.

In January 2019, Nvidia CEO Jensen Huang called the AMD Radeon VII graphics card "mediocre," explaining his opinion that the product has "disgusting performance and carries nothing new." Huang also said that AMD FreeSync technology does not work.

2018

Critical Vulnerability Detection

In March 2018, information security experts announced the discovery of more than a dozen critical vulnerabilities and backdoors in AMD Ryzen and EPYC processors, which AMD "owes" to its suppliers. All vulnerabilities are contained in the secure code execution environment AMD Secure Processor, used in both EPYC and Ryzen. Vulnerabilities allow attackers who have already penetrated the system to install malware that will be very difficult to detect and fix. Read more here.

EPYC 7000 Series Delivery on PowerEdge Platforms

On February 7, 2018, the company AMD announced that the EPYC 7000 series processors are shipped in three platforms. Dell EMC PowerEdge

PowerEdge R6415, R7415, and R7425 are scalable single and dual processor servers designed to support high-performance workloads, including virtualized storage area networks (VSANs), hybrid cloud applications, dense virtualization, and big data analytics. AMD EPYC's single-processor configuration enables Dell EMC PowerEdge platforms to demonstrate up to 20% lower total cost of ownership in single-socket, 4-node VSAN configurations, the company claims.

Dell EMC PowerEdge leverages the capabilities of the AMD EPYC 7000 series to create a variety of single and dual processor servers with 32 to 64 cores with up to 4 TB of memory and 12 to 24 NVMe drives optimized for database and analytics. EPYC also supports high throughput and works with dense GPU/FPGA installation on these Dell EMS PowerEdge platforms for HPC applications.

EPYC's single-processor capabilities enable Dell EMC to create single-processor servers that can meet the needs of most client workloads. Dell EMC's AMD EPYC portfolio of solutions will have advantages for a wide range of workloads, including dense virtualization, hybrid cloud applications,, program defined storage systems, and CAD other memory and I/O demanding applications, said AMD's senior Forrest Norrod vice president and general manager of Database and Embedded Solutions.

In addition to creating the PowerEdge R7425 multi-feature dual-processor platform, EPYC allows you to create competitive single-processor solutions, allowing you to use this category of servers in new segments. According to AMD, in single-processor configurations, PowerEdge R6415 and R7415 have 32 processing cores and are optimized for heavy workloads and applications with multi-threaded architecture. In addition to the available TCO of all three platforms, they also deliver high I/O speeds with up to 128 PCIe interfaces and up to 342 Gb/s of memory bandwidth.

As of February 7, 2018, Dell EMC PowerEdge R6415, R7415, and R7425 servers with AMD EPYC processors are already available for order.

2017

AMD brings EPYC server processors to market

On June 21, 2017, AMD announced the replacement of Opteron 14nm server processors with EPYC processors. As part of the line of 8-, 16-, 24- and 32-core models of the 7000 series for single-processor and dual-processor servers (nodes).

EPYC, (2017)

The company said its products provide greater floating point computing performance, memory bandwidth and I/O controllers. Positioning on one- and two-socket systems, according to the creators, will help meet the needs of 90-95% of the server market^[2].

All processors are based on one basic structure: four Zeppelin crystals in a multi-chip package (MCP). Each Zeppelin crystal is equipped with eight cores and is divided into two CPU Complex units. The processors support the Infinity Fabric interconnect, which is used on all AMD chips to communicate between two processors in two-socket systems. Both processors offer only 64 PCI Express lines each, the remaining 64 lines are allocated to the interconnect. In total, two processors offer 128 PCI Express lines that can be used arbitrarily.

The frequencies of the EPYC 7000 products are low, due to the presence of system logic units in the crystal structure. At the head of the family is a 32-core EPYC processor 7601 with a frequency of 2.2/3.2 GHz, 64 MB of L3 cache and heat dissipation up to 180 W. Some CPUs have a configurable TDP of 155/170 W - thus, system integrators will have a choice between power saving modes and maximum server performance.

The EPYC 7601 can raise the frequency for all cores from the base to 2.7 GHz and up to 3.2 GHz for 12 cores or less at the same time. The EPYC 7401 model with 24 cores for eight or less cores supports operation at a frequency of 3 GHz, overclocking all cores at once to a frequency of 2.8 GHz (base frequency 2 GHz).

EPYC models for single-processor systems are marked with the suffix "P": 7551P, 7401P and 7351P with 32, 24 and 16 cores, respectively. EPYC P processors operate at slightly higher clock speeds than models for dual-processor systems, although according to general specifications they do not differ. We add that for each processor connector in the system there can be up to 2 TB of RAM with eight-channel access. In addition to 128 PCI Express lines per processor, the system offers SCH (server controller hub). It provides, for example, four USB 3.0 ports, numerous I2C and SMBus interfaces. The chipset also contains several clock generators.

Deliveries of systems based on EPYC processors are expected at the end of July 2017.

AMD opts for InfiniBand and Mellanox Ethernet solutions for EPYC Datacenter platform

Ethernet and InfiniBand switching solutions from Mellanox are selected AMD as the preferred solutions for accelerating new EPYC data center platforms. Mellanox 25, 50 and 100Gbit/s Ethernet and EDR (100Gbit/s) InfiniBand solutions are networking solutions for connecting AMD EPYC processors, providing the maximum return on investments in cloud computing, Web2.0,,,, and big data machine learning storage systems high-performance computing infrastructures.

Ethernet and InfiniBand solutions from Mellanox are selected by AMD as the most preferred for accelerating new EPYC data center platforms. AMD EPYC processors for data centers provide unparalleled capabilities in all respects, and when connected to intelligent and high-speed Mellanox network connections, the combined platform provides the highest performance for a wide range of applications. As AMD demonstrated at the EPYC announcement event, this fully integrated solution brings together the world's best technologies and is designed for the industry's most reliable applications.

The combination of Mellanox's intelligent networking solutions and AMD's innovative EPYC processors will enable our joint customers to improve data center performance and optimize their total cost of ownership, "said Mikhail Kagan, CTO, Mellanox Technologies. "Exponential data growth and the need to analyze it faster and in real time require higher data throughput and intelligent data processing, which are the main advantages of EPYC platforms that include Mellanox networking technology. We look forward to continuing to work with AMD and OEM partners to offer industry-leading computing and DSS of infrastructure for data centers

AMD EPYC

EPYC is a processor for data centers.

On May 18, 2017, AMD announced the release of high-performance server solutions for data centers based on the EPYC processor platform (codename Naples).

EPYC focuses on dual-processor servers for cloud and traditional data centers. The processor scales to 32 physical cores. EPYC-based servers will be released in June 2017^[3]

AMD EPYC, (2017)

Together with the EPYC processor, AMD is taking another step on its way to the world of high-performance computing. As we showed today, we see the possibilities along with a solution based on a single-socket processor. We are confident that this product line has the potential to transform a significant portion of the data center market by offering a combination of performance, design flexibility, and cost of ownership. Forrest Norrod, Senior Vice President and Head of Enterprise, Embedded & Semi-Custom

EPYC features

• a scalable 32-core system-on-a-chip (SoC) supporting two high-performance threads per core.
• Memory performance with 8 memory channels per EPYC device. Dual-processor servers supporting 32 16-channel DDR4 DIMMs can achieve 4 terabytes of RAM.
• the ready-made SoC with integrated high-speed I/O supports up to 128 PCIe 3 lines, eliminating the need to install a separate chip.
• Optimized cache design for performance and energy efficiency.
• Infinity Fabric bus is used to connect two EPYC processors that are installed on the same system.
• special security hardware

AMD EPYC Processors Presentation, (2017)

2011: AMD Opteron 6200

After a short delay and a year of discussion on November 14, 2011, AMD announced the release of its first 16-core Opteron server chip, which contains the largest number of available cores on x86 processors at present.

The Opteron 6200 processors, codenamed Interlagos, are 25% - 30% faster than their predecessor, the 12-core Opteron 6100 chip, said John Fruehe, AMD's chief product marketing officer.

The delivery of microchips has already begun. According to Frue, in the coming weeks, servers with these processors will begin to be released by Hewlett-Packard, Dell, Acer and Cray. This microprocessor model is based on a technology called Bulldozer (Bulldozer), which provides throughput and performance with economical energy consumption.

According to Dean McCarron, chief analyst at Mercury Research, AMD's supply delay is trying to change the state of the server market as a result of the previous year's losses in the fight against Intel. AMD's global share of the server market in the second quarter of this year, according to IDC, was only 5.5%, while Intel was 94.5%. AMD also negotiated with the company's customers, who complained about the failure to meet product delivery plans.

The company has prepared a line of five models of processors of the 6200 series: 6262 HE, 6272, 6274, 6276 and 6282 SE, they operate at clock frequencies from 1.6 GHz and 2.6 GHz. The cost of products is from $523 to $1019. The chips consume between 85W and 140W and plug into an existing server socket. It is possible to replace the old 12-core processors.

According to Frue, the 6200 lineup is designed for really scalable applications, multithreaded tasks such as databases, cloud and high-performance computing. They are designed for heavy-duty computing threads and virtual server environments. Linux operating systems, Microsoft Windows Server 2008 are configured to take advantage of 16-core systems, while more instances of virtual machines can be created to process transactions, Frue emphasized.

The Bulldozer architecture combines an integer processing device and a common floating point processing module in the processor to increase the number of operations performed in one clock cycle while consuming less power. Other processor improvements include Turbo Core technology, which can increase the clock speed to 500 MHz of all cores and up to 1 GHz on some cores, depending on the required performance.

AMD has been involved in management restructuring in recent months, and new CEO Rory Read said the priority is to improve profitability in the business process. Reed acknowledged that customers complained about the timing of AMD's product shipments, and the top priorities were chip manufacturing, shipping and delivery.

The Opteron 6200 is very important for AMD in the server market plans, as it is an anchor product, noted Mac Carron. According to the analyst, AMD has made several important victories with the release of this chip. One of them is a project power engineering specialists USA at the Oak Ridge National Laboratory, where Nvidia it is built on the basis of 16-core AMD processors and GPUs supercomputer under the name Titan. Its peak performance will reach 20 petaflops and surpass Japanese the K supercomputer, which produces about 8 petaflops and is mentioned Top500.org in June as the fastest computer in the world.

Opteron launches ahead of Intel's Xeon E5 server processors, which are already in production. The first servers with these processors will be released in the first half of next year. According to Kirk Skaugen, vice president and general manager of Intel's Data Center Group, the company has implemented about 400 technical innovations in the Xeon E5.

About the product

Opteron is a processor for servers and workstations on the x86 platform. This is the company's first processor to support the AMD64 architecture and a set of instructions known as x86-64. The product was released on April 22, 2003. The development, called SledgeHammer (K8), aimed to compete in the server and workstation market in the same segment as Intel's Xeon processors.

In March 2006, a dual-core release of the processor was released under the code name Denmark.

Product development continued based on the AMD K10 microarchitecture (codenamed Barcelona), which was released on September 10, 2007 as a quad-core configuration.

In March 2010, the company released Opteron models with 8 and 12 cores, codenamed Magny-Cours.

Notes