Sales leaders of DELL servers in Ukraine.

Quick search filter of DELL servers

AMD 128-ядерный Epyc 9754 Server Solutions

Dual-socket servers are not always better than single-socket servers.

In the world of server technology, dual socket was once the cutting edge technology. Dual sockets were the best way to scale the performance of individual servers by placing two processors in one server - theoretically up to double the performance in roughly the same server footprint. They became prominent during the transition from scale-out (more sockets) to scale-out (more servers) computing. This was the transition from the era of mainframes and multi-socket servers to the mainstream two-socket x86 servers.

Today, two-socket servers are not necessary for scaling, and in fact, the middle ground is shifting for several reasons. Modern "true single-socket servers" based on AMD EPYC™ processors can achieve a new optimal level of total cost of ownership (TCO). This is based on past and new variables that are contributing to the wider adoption of single-socket servers. Modern EPYC processors have enough cores, memory and I/O to meet most application performance needs. But be careful, not all single-processor servers are "true single-processor servers" because they skimp on cores, memory channels, I/O lanes, RAS, etc. To be a "true single-socket server", they must almost match the capabilities of their big brother - a two-socket server.

Myths and legends of two-socket servers.

The biggest myth of all time about dual-socket servers is redundancy. Let's call it an "urban legend" - if one of the server's sockets fails, then everything that works on it switches to another and continues to work. This is simply not true for common dual-socket servers with common operating systems, but I can't tell you how many times in my career I've heard about redundancy.

Reliability is another dual-socket vs. single-socket myth, because reliability is a decreasing probability function that depends on the failure rate. The failure rate is directly related to the complexity, in other words, the more components in the server, the higher the failure rate. Single-socket servers have fewer components in the specification, usually have better power and cooling characteristics, can be built around server-class processors, etc., all of which contribute to lower failure rates.

Another misconception about single-socket and dual-socket processors is core density. This myth states that servers need dual-socket processors to achieve high core density and maximum performance. This may be true for the competition, but not for AMD EPYC™. The 4th generation AMD EPYC™ processor supports up to 128 cores per socket. In comparison, the competitor's Intel® Xeon® Platinum 8490H processor supports only 60 cores. This means that customers of AMD's competitors using x86 servers have to take on the additional cost and complexity of a dual-socket server to get performance competitive with what AMD can achieve in a "true single-socket server".

Memory and I/O limitations are an additional myth between dual-socket and single-socket servers. Again, while this may be a competitive limitation, AMD offers a "true single-socket server" with 12 DDR5 channels (up to 24 DIMMS) along with 128 PCI Express® Gen5 lanes in a single-socket server. By comparison, the top-of-the-line Intel® Xeon® Platinum 8490H processor in a single-socket configuration would be limited to 8 memory channels (16 DIMMS) and 80 PCI Express Gen5 lanes, requiring customers to purchase more processors to achieve I/O parity and memory required for the server.

Until AMD released EPYC™ server-optimized single-socket processors, competitors offered very limited single-socket processors, namely the Intel® Xeon® E family, which was limited to 8 cores, 2 DDR channels (4 DIMMS), and 40 PCI Express Gen4 bus lanes. Such a non-competitive proposition likely led to the low adoption of single-socket servers in the past, and may account for many of the myths and legends between single-socket and dual-socket servers that arose out of a lack of choice for a reliable "true single-socket" enterprise-class server.

When a single-socket server is the right solution.

However, there are times when a single-socket server is often better than a dual-socket server. Considerthe Dell Poweredge R7615 server based on the AMD EPYC™ 9554P processor, which has 64 cores, compared to the Dell Poweredge R760 dual-socket server based on the 40-core Intel® Xeon® Platinum 8380 processors with a total of 80 cores.

At first glance, it may seem that an 80-core server with two sockets is superior to a 64-core system with one socket. However, you only have to look a little closer to see that this is not the case. Although an Intel processor-based machine has 25% more cores than an AMD processor-based server, the AMD processor-based machine is more powerful - about 5% higher performance. And this performance comes from 25% lower power consumption, a 62% lower CPU list price compared to a competitor's dual-socket server, lower NUMA complexity, and fewer parts to fail.

The single-connector configuration allows for significant efficiency gains without sacrificing performance. Moreover, this efficiency gain is scalable. For example, it takes just 12 servers with single-socket 96-core AMD EPYC™ 9654P processors to achieve 10,000 units of integer performance, an important measure of computer speed, compared to 17 servers with dual-socket, 40-core Intel Xeon 8380 processors to achieve same result. The AMD EPYC™ solution uses 29% fewer servers and consumes 47% less energy, saving about 278,148 kWh of electricity over three years, which lowers the total cost of ownership. Reducing energy costs and energy use is a particularly important and challenging task today, when energy prices are rising and companies are trying to maximize investment.

128-core AMD Epyc 9754

This is not the death of dual-socket servers.

Of course, this doesn't mean that two-socket servers will die; single socket servers are just another tool to consider. Dual-socket servers can be the right choice when companies need maximum performance and are willing to pay more for hardware, power, and data center real estate to meet those needs. Such intensive applications include customer relationship management systems, content delivery systems, and others. Single-socket servers are great for standard business applications: networking and security, most database applications, enterprise resource management, special business applications, supply chain management, HR applications, and more. Two-socket servers aren't going away; the more likely reality is that four-socket servers will converge to two-socket servers, and some percentage of two-socket servers will convert to single-socket servers. But the big picture is that the world has been migrating from many sockets to fewer since the dawn of computing, dual-socket servers have been the workhorse for years, going from many sockets to one is just part of the optimization process, fighting gravity never is a winning strategy.

What are 10 reasons to consider a "true single socket server" now?

  1. No compromises in performance - a large number of cores and performance cores are available in a single-socket server: AMD offers 96 cores in the EPYC™ 9654 and 128 cores in the EPYC™ 9754 for single-socket servers.
  2. Memory scaling without adding sockets - EPYC™ 4th generation single-socket servers can support 12 channels of DDR5 and up to 24 DIMMs with a capacity of up to 6 TB.
  3. I/O Scaling Without Compromise - AMD's unique EPYC™ processor architecture provides massive I/O capabilities in both single-socket and dual-socket designs, so customers don't need to buy more processors just to get I/O.
  4. Reduced complexity - Avoiding the complexity of NUMA with two connectors makes it easier to deploy and plan workloads across a fleet without having to worry about NUMA implications.
  5. Reduced Consolidation Costs - Reducing server complexity makes it easier to deploy, manage, maintain, and upgrade.
  6. Reliability and cost-effectiveness of failover clustering - Many cloud applications require a minimum number of nodes for failover clustering. A single-socket server can reduce the cost of fault tolerance to achieve a minimum number of server nodes.
  7. Avoid Data Center Hot Spots - Many data centers have power and cooling issues. Single-socket server deployments allow power and cooling to be distributed throughout the data center.
  8. Increased energy efficiency - A higher level of integration has been proven time and time again to provide higher efficiency and overall lower energy consumption. In the case of the EPYC 9754 compared to the Intel Xeon 8490H, 128 EPYC processor cores in one socket consume less power and have higher performance than 64 cores per socket in two Xeon sockets.³ In the past, dual sockets had the advantage of SPECpower® in overall energy efficiency due to the distribution of more resources between the two connectors. With AMD EPYC™, this is no longer the case.
  9. Better licensing cost model (for some applications) - Some applications are still licensed by sockets. Avoiding dual-socket servers can reduce licensing costs for some software. Even in per-core licensing models, it is cheaper to deploy servers with a large number of cores (single-core or dual-core) to amortize the cost of the infrastructure on more cores.
  10. Moving to a multi-host infrastructure - Modern infrastructure moves most of the operating system and hypervisor software stack to the DPU/SmartNIC. DPUs such as AMD Pensando™ enable sharing of DPUs between multiple servers. Single-socket servers are ideal for this new deployment model of connecting multiple single-socket servers to a single DPU.

Single socket server: Excellent choice.

"True single-socket servers" are a great alternative to two-socket configurations and are often the best alternative in many cases. They give IT professionals more options to optimize the modern data center with increased energy efficiency as well as exceptional system performance. Users can even achieve similar results using AMD's single-socket servers, but at significantly lower power consumption and cost than competing dual-socket servers.⁴ In other cases, dual-socket servers are the best choice. AMD offers a better choice.

How can we help?

For more detailed information about the DELL PowerEdge R760 server with DDR5 4800 or the DELL PowerEdge R750 server with DDR4 3200, you can visit our website SERVER SOLUTIONS , to find out the price of the server, go to the DELL Server Configurator link .

Leave a comment

Please note, comments need to be approved before they are published.