At the Data Center World 2023, Omdia’s Vlad Galabov and AMD’s Kumaran Siva discussed the elements that contribute to an optimal computing experience, which include performance, security, and efficiency. They explore the growing importance of efficiency in data centers due to sustainability concerns, which has intensified efforts to achieve greater performance per watt.
This is Part 1 of their conversation.
The transcript follows below. Minor edits have been made for clarity.
Transcript:
Vlad Galabov: Hi, I'm Vlad Galabov, head of Omdia’s cloud and data center research practice, and I'm currently at Omdia’s Analyst Summit at Data Center World. I'm joined by Kumaran Siva, who runs strategic business development for the CPU product line at AMD and is one of our key partners at the summit. Hi, Kumaran.
Kumaran Siva: Hi, Vlad. Thank you so much. it's an honor to be here.
Vlad: Yes, it's great to have you. I think that the topic of computing is so key for our conversation today. We're talking about sustainability and how to turn the effort to be more sustainable into also a very practical and business-driven initiative. So, as we go into that, I wanted to talk to you about computing. Computing is the essence of the data center. So, what do you think, and what do people tell you as you speak to end users? What makes an optimal computing experience?
Kumaran: It's a good question. So, in terms of an optimal computing experience, I feel like you have certain baseline pieces that need to be there. You always have to have performance. A user expects a job to be done overnight or a job to be done with a simulation to be completed within a certain period of time. So, there's always an element of performance, an element of time to completion that's there. There's also an implicit requirement for security, especially in this world where we have lots of ransomware attacks. You hear about these things all the time.
Every ingredient in the data center has to have some element of security and understanding of a piece of a security strategy that the data center operator, whether it’s an enterprise or a hyperscaler, implements. On top of that, though, increasingly now, it’s becoming about efficiency. I think data centers have always been about efficiency. They've been about efficiency because of physics. Like, you have to provide a certain amount of power and be able to cool it.
But now, there's a renewed effort because of sustainability and because of our more global awareness of the impact of climate change. There is a push to be more and more efficient, and that changes the scale a little bit. We were heading towards as much efficient performance per watt as you could get, but sustainability has redoubled that effort and maybe shifted that curve even more towards greater efficiency.
Vlad: One of the things that impressed me when the Epic product line first came out was a story about consolidation – the opportunity to change the components within a server to reduce the number of CPUs or to use fewer servers. Could you talk about what role consolidation has to play in this pursuit of data center sustainability?
Kumaran: Yeah, absolutely. One thing that AMD has done with our architecture and coming into the market, and obviously we provided an alternative as a baseline, but really if you look at our product line – it ranges from very low core count up to very, very high core counts. And this is because of our chip-led architecture. And what this allows customers to do is pick where they're comfortable with, in terms of getting the best energy efficiency combined with the performance that they need. And then also combined with cost savings that they can get. So, for example, our top of the stack for Genoa, our densest CPUs, are 96 cores.
So, you can take up to 15 legacy Intel servers and consolidate them down into five of these Genoa servers. But we have customers that do 15 to 10 and still get a lot of benefit from a TCO point of view, but also get an amazing performance uplift on those workloads and when they have customers that need to have jobs done in a particular timeframe. Or they're not comfortable with having so many workloads all over the same server.
There are concerns around blast radius. I think some things are being done to mitigate those concerns. But still, I think just overall, people are cautious, but incrementally you get all three benefits: power efficiency, better sustainability, and better performance and TCO reduction. So, it is the best of all worlds.
Vlad: I think very often we talk just about compute-intensive workloads, but some workloads require a different set of an equation, in terms of capacity within a server. Could you give us an example of a memory-intensive workload? For example, a client ended up choosing a different server configuration because they were not optimizing for computing.
Kumaran: Yeah, we have customers that will design in two sockets of 16-core Genoa, with two terabytes of memory space in it. So, these applications include things like a mission-critical database where you need a large memory set. But you need that single type of performance because the coordination between cores is too costly, so they'll do that. We have examples of high-performance computing commercial HPC applications. So, computer-aided design and electronic design automation – those fields definitely benefit from larger amounts of memory, memory bandwidth, and then higher-performance CPUs. So, our architecture fits those customers perfectly and still produces a power efficiency benefit. So, it is neat and has the benefit of being on the latest process node from the foundries.
Watch Part 2 of Kumaran’s interview
