AMD Continues Its Datacenter Push With 3rd Generation Of EPYC


The issues facing enterprise IT in this new datacentric era are not new. They are issues that have plagued IT for a long time, such as data management, security and the ability to support business initiatives faster and cheaper. That said, new wrinkles have been introduced. Enterprises must know how to manage the data being generated in the wild, on the edge and in the cloud. They also must be able to deploy and support infrastructure that can support the unique requirements of the many new workloads creeping into the enterprise every day. What’s more, they must do all of this while hardening an environment to an increasingly hostile cyber world. 

In short, these challenges aren’t new to IT organizations, but they are bigger – much bigger. And they are more complex – much more complex. And these challenges crop up more often due to the open, software-defined world in which we live. Much more often.

AMD’s initial launch of EPYC back in 2017 had the theme “born in the cloud.” Catchy, but not a strong message for the enterprise. The 2nd Generation of EPYC carried more enterprise datacenter appeal, and the addition of “F” series processors gave it a stronger affinity for enterprise workloads.

Casual observers of the server market may look at the launch of AMD’s 3rd Generation EPYC CPU and note the topline specifications are largely unchanged. This assumption would be a mistake. With the launch of “Milan,” AMD has further cemented its footing in the enterprise datacenter. The architectural and security updates make this CPU worthy of a hard look by any IT organization that values the all-important “performance per dollar” metric. 

In the following few paragraphs, I will cover some of the areas where I believe this CPU will be of particular interest to enterprise IT organizations struggling in this datacentric era. 

Performance – it still matters

While the way we measure server CPU performance has shifted, many in the IT community still look at things like core count, memory capacity and clock speeds to gauge how well a CPU will support its workloads. Rough estimates around VM capacity or database performance are (mentally) sketched and biases are established. I’ve done it in my life as an IT Director and know very well that this thinking is still prevalent. And in this regard, EPYC – though still significantly in the lead – looks similar to its previous generation. 

However, the following generational performance improvements over the 2ndgeneration are anything but incremental: 

  • 19% instructions per clock (IPC) gain 
  • 17% gain in 2P Floating Point performance
  •  14% 2P integer performance gain
  • 20% improvement in Java virtual machine performance 

AMD’s ability to achieve this performance without making changes to “speeds and feeds” speaks to the Zen3 core’s architectural improvements, cache changes (32MB shared per core complex) and the overall improvements made to the core complex itself.

AMD’s Zen Gen 3 Core Complex

Within the Zen3 core, the 32MB shared cache represents a doubling of available addressable cache per core. It positively impacts memory latency, making the applications and workloads that power the enterprise run that much better. AMD seems to think that this, combined with other improvements, has enabled the many relevant enterprise workload performance claims in HCI, database and data analytics. 

One of the things I like about this 3rd Generation EPYC CPU is its real-world relevancy. While speeds, feeds and benchmark results are great, what matters to an IT person is how much faster and more efficient a line of business application will run, how many more VMs they can get per server and finally, how much they can save every year by reducing their server footprint and gaining utility from every server. 

Lastly, I like what I see in the EPYC performance profile and how well designed it is for pretty much any enterprise workload. The embarrassingly rich feature set, combined with continued improvements to core performance, makes this a strong building block for the workloads that power both today’s and tomorrow’s enterprise. While OEM partners may try and position servers based on EPYC for a specific workload, this CPU checks boxes across the performance spectrum. Integer performance? Check. Memory intensive? EPYC leads in capacity and bandwidth. Acceleration required? 128 lanes of PCIe Gen4. Indeed, this CPU design makes it not a jack of all trades but a master (of all). 

Security – it’s not just for CISOs

I can remember executing a series of focus groups some time back teasing out the topic of datacenter security. The response from IT admins was almost universal – they had a cybersecurity organization to worry about security. While this is largely accurate, the way hackers infiltrate, proliferate and steal data should make security a top-of-mind concern for every employee of every organization (or at least every IT professional). 

Security is another area where AMD has improved on an already solid foundation. There are three elements to EPYC’s security strategy – 

  1. Ensure a clean boot. Through what AMD calls “Hardware Validated Boot,” the server is booted into a pristine, authenticated state – no hacked firmware or  BIOS. 
  2. Physically secure the components. Through Secure Memory Encryption (SME), all system memory is cryptographically protected, ensuring memory can’t be scraped or otherwise physically removed and read. 
  3. Secure the apps and data. Through Secure Encrypted Virtualization (SEV), EPYC can deliver virtual machines cryptographically isolated from each other and the hypervisor.
New security features for 3rd Gen AMD EPYC.

In the 3rd Generation EPYC CPU, AMD added a couple of new security features, further solidifying its support for the enterprise. One in particular, SEV-SNP (Secure Nested Pages), locks down memory for virtualized environments. Again, this is not a feature that would necessarily stand out to the casual observer, but it is extremely important for securing the enterprise datacenter. 

It’s about balance (and that’s not just a cheesy play on Zen)

Perhaps the most impressive thing about EPYC is the balance of features and what they mean to an IT organization, whether that’s a server room, datacenter or the cloud. The balance of cores-to-memory-to-I/O makes EPYC a strong candidate to support the workloads that drive today’s business. It’s not the richness of cores, bandwidth of memory, or what a systems provider can do with all that I/O. It’s the balance between these subsystems and the richness of cache that enables workloads with a variety of “performance” needs. 

In closing

The world has fundamentally changed in the last year or so. Digital transformation projects have been rescoped to account for the changes brought about by Covid. Technology is more crucial to the business than ever before. And while we live in a software-defined world, the underlying infrastructure has to be able to support the diverse needs of this hybrid multi-cloud world – today and tomorrow. 

With the launch of this new EPYC CPU, AMD is showing that it understands the needs of enterprise IT. It must always leverage its partners (server OEMs, channel partners) to ensure the IT consumer sees the value of EPYC. Further, the company must ensure its partner ecosystem is positioning the EPYC portfolio for the projects consuming IT’s energy – digital transformation, hybrid multi-cloud, app modernization (i.e., cloud-native) and the like. If so, we could see significant gains for EPYC in the coming year. 

To read about how HPE has strengthened its portfolio with EPYC, visit here.