You can hardly find a technology-related article today that doesn’t mention artificial intelligence (AI), deep analytics or the digitally transformed business. With good reason, I should add. These workloads are driving a lot of the intelligent business processes that enable all those digitally transformed businesses.
The hype around the platforms and accelerators that power AI is related to this dynamic. Again, with good reason. CPUs combined with GPUs deliver performance acceleration, making machine learning models and AI inferencing happen faster.
But there’s an integration point for all of this silicon, software, and data—the server. And in this age of software-defined everything, we sometimes overlook the server’s critical role as the delivery vehicle for transformation. In this post, I will discuss Dell’s updated four-socket server portfolio and how the company has partnered with Intel to deliver platforms to drive the most demanding workloads.
To look deeper into Dell’s four-socket portfolio, you can also read my research report here.
The distributed data-driven enterprise requires a diversity of computing
Data drives the modern enterprise. Data is generated, transformed and used everywhere—at the edge, on-premises and in the cloud. The distributed nature of modern IT environments requires compute solutions tailored for the workloads they run and the infrastructures where they reside. Edge environments can be anything from retail outlets to manufacturing floors or oil rigs. And the computational needs in these environments are wholly dependent on applications and architectures.
The on-premises datacenter is similarly diverse in the computational needs of workloads. The requirements of traditional line of business (LoB) applications and ERP systems differ from those of cloud-native applications and even AI-driven workloads. Cloud-native workloads deployed in containers benefit from scale-out infrastructure, but ERP systems, database and analytics platforms and virtual desktop infrastructure (VDI) can benefit from richer server configurations, scaling up instead of out.
Forbes Daily: Get our best stories, exclusive reporting and essential analysis of the day’s news in your inbox every weekday.
To be clear, the diversity of infrastructure is not limited to edge servers versus on-premises servers versus cloud servers. This infrastructure diversity is most important within the on-premises datacenter. As I talk to IT professionals, I hear regularly that talking about computing as a commodity in the abstract works fine, but that when work needs to get done, having the right platform for the right workload is critical in terms of performance, cost of ownership and perhaps most importantly time to value.
Further, I continue to hear that computational power across the range of servers—single-socket, two-socket, and four-socket—is critical for supporting the business operations of the digitally transformed organizations.
Four-socket servers: heavy-duty computing for heavy-duty workloads
Four-socket servers are the foundational element of the modern datacenter—kind of the bedrock. It is on four-socket servers that the most critical workloads, and the most computationally demanding workloads, exist. ERP, LoB, data warehouses, analytics platforms—these workloads require the computational support, the memory footprint and the richness of I/O that only comes in a four-socket server. And while the industry has seen “bigger iron” servers shrink down (remember 16-way and 8-way servers?), the four-server socket is here to stay for the foreseeable future.
I see a few distinct deployment scenarios for four-socket servers:
- ERP and management. The business’s applications are built on database management platforms like Oracle, Microsoft and SAP.
- Data analytics. Though tightly coupled with ERP, analytics has expanded into all data sources, both structured and unstructured. One of the more interesting emerging analytics trends I’ve seen is in hybrid transactional/analytical processing (HTAP)—performing analytics on transactional data in-memory.
- High-density virtualization. The case for high-density virtualization comes down to TCO. More virtual machines on fewer servers equals fewer servers. Which means lower CapEx and OpEx.
- Enterprise virtual desktop infrastructure (VDI). As with high-density virtualization, four-socket servers can deliver more VDI sessions on fewer servers without sacrificing performance.
- AI and high-performance computing (HPC). Both AI and HPC perform better with more computational power, a bigger memory footprint and the use of accelerators. Four-socket servers deliver on all three fronts.
Dell’s four-socket lineup
Dell updated its four-socket portfolio when it released its 16G lineup in February of this year. This server lineup, powered by Intel, is worth looking at.
I look at the workload characterizations discussed above and further categorize them into three bigger buckets. Business-critical, data analytics and accelerated computing. I do this to demonstrate further how the Dell portfolio lines up nicely in support of the most important workloads.
In the 16G generation, Dell trimmed its four-socket portfolio to two platforms. The R860 is about balanced performance, with lots of compute and local storage to support workloads with low latency requirements. The R960 is about acceleration; it’s where customers can achieve 1:1 pairing of CPU to GPU. I like the approach that Dell has taken, because customers can configure the best platform that fits their needs without much complexity.
Intel Sapphire Rapids brings acceleration
Dell’s four-socket portfolio is built entirely on Intel’s Sapphire Rapids—the 4th Generation Xeon Processor. (AMD does not support four sockets in its EPYC processors.) I think that the workload acceleration built into Sapphire Rapids gives it a strong affinity for the areas where four-socket servers play.
Intel took a novel approach when designing its Sapphire Rapids processor. While it delivered more power through the cores and compute complex, it also added specific acceleration engines to help offload some tasks associated with modern workloads. While already interesting in concept, these dedicated offload engines have meant a lot in real-world performance.
I like Intel’s approach with Sapphire Rapids. While the CPU doesn’t match the extreme core counts of Intel’s competition, it delivers additional performance with this acceleration engine approach. And as it hits parity from a core count perspective, these acceleration engines will give the company an even stronger competitive position.
Finally, I like the amount of time and energy the company has put into enabling the ISV ecosystem. These engines mean something only when an application can realize the benefits. And that, in turn, happens only with a lot of enablement work from Intel.
What this all means — an analyst’s take
In this cloud-native-dominated world, a lot of attention is paid to the new and emerging technologies that support digitally transformed business. And that is understandable.
However, any enterprise organization driven by data understands that the historical data it has generated and stored over the years is critical to creating differentiation in the marketplace. That historical data feeds customer service, product planning and the business’s understanding of market trends. Because of this criticality, the historical data has to keep feeding machine learning algorithms and analytics platforms—and do it faster than ever. That’s why four-socket servers are critical to the success of digital transformation projects. And why investing in the right server infrastructure is essential.
I like Dell’s revamped four-socket strategy. I like the simplification and how the company is leveraging Sapphire Rapids to deliver the right amount of computing for these demanding workloads.
I’ll be tracking how the market responds to the R860 and R960 over the next few quarters, and I’ll be sure to share what I see.