The past few years have been a rocky ride for AMD, which holds second place after Intel in the processor market. AMD’s 2011 revenue of US$6.57bn was flat in comparison to 2010 and this year’s first quarter was not much better, with a 2% revenue decrease year over year recorded. This was followed by a 10% year-over-year drop in Q2 and a 25% drop in the next quarter.

As AMD’s quarterly revenue has shrunk, its losses have also increased. The company went from a $590m net loss in Q1 to a $37m net income in Q2 and then a net loss of $157m in the year’s third quarter. As it announced its third-quarter results, it also announced it would be laying off 15% of its workforce as part of a restructuring effort expected to save about $20m in the fourth quarter and $190m over the course of next year.

Commenting on third-quarter results, AMD president and CEO Rory Read attributed the downward trend to changes in the PC market. The changes, he said, happened faster than the company had expected, making the restructuring effort necessary.

Against this backdrop, AMD’s server business has gone through a lot of changes. In March, AMD bought SeaMicro (a microserver vendor). It was a $334m acquisition which first manifested with the announcement by AMD of the SeaMicro SM15000 server in September.

The first SM15000 available on the market has arch-nemesis Intel’s Sandy Bridge processors inside, and the AMD-based version is coming in the fourth quarter.

Late in October of this year AMD announced a plan to bring to market a 64-bit processor for servers based on ARM architecture in 2014. The architecture will be part of a System-on-a-Chip (SoC) card and integrated with SeaMicro’s Freedom fabric technology, which the chipmaker said was its main reason for buying the microserver firm. Aimed at data center customers whose applications require highly parallelized processing.

In July, AMD brought in a new man to run the server business. The company appointed Dr Suresh Gopalakrishnan as the unit’s general manager and corporate VP. He was previously in charge of marketing and product management at the Ethernet-switch vendor Extreme Networks.

FOCUS recently caught up with Gopalakrishnan to discuss AMD’s strategy in the server market going forward.

DatacenterDynamics FOCUS: Why did you make the switch from Extreme Networks to AMD?

Suresh Gopalakrishnan: While at Extreme Networks, I defined and developed networking products for data centers and carriers. I spent many hours with customers to understand the problems that they are trying to solve within their data centers.

I came to realize that servers, switches and storage as separate islands will not work at data center scale. AMD’s vision for the data center was matched with my thoughts. I felt that joining AMD was an exciting opportunity to bring significant innovation to the data center market.

DCDF: You’ve previously worked at Sun and HP?

SG: I worked at HP and then Sun in the early 1990s. At HP I was part of the integrated circuits business and worked on a wide range of chips from storage controllers to server chipsets for HP processors.

From HP I joined Sun as one of the managers on the UltraSPARC II processor. I was responsible for the design of internal and external memory systems and the verification of the whole processor.

DCDF: Do you have a strategy for reversing the downward revenue trend on the server-business side of AMD?

SG: As our CEO Rory Reed mentioned in the earnings call recently, [the] server business is one of the strongest opportunities for AMD.

There are two key parts to our server strategy – leverage our full portfolio of IP to offer industry-leading performance-per-watt for targeted workloads and focused go-to-market execution with value propositions aligned to target markets.

DCDF: What are the top three emerging markets for AMD’s server business?

SG: Virtualized workloads in public and private clouds, as well as “big data” workloads utilizing Hadoop are focus areas for us.

DCDF: AMD announced its latest Opteron 6300 series processors in early November. How do the latest Opteron processors differ from their predecessors and what do they mean for the operator?

SG: There are several architectural enhancements leading to improved performance-per-watt and instructions per cycle in the new Opteron 6300 processors. These processors offer 40% improvement in performance-per-watt compared to the previous generation. This reduces the power consumption and dissipation in a data center.

Opteron 6300 processors also offer 24% improvement in Java performance over the previous generation. This is a significant benefit for customers with Java-based workloads in public and private clouds.

DCDF: What are the cornerstones of your strategy for gaining market share in the enterprise data center?

SG: We will focus on virtualized workloads in the enterprise data center to leverage our strengths in virtualization and work closely with our go-to-market partners to get adequate enterprise-customer coverage.

DCDF: Do you think enterprise data center consolidation is a big opportunity for AMD?

SG: Data center consolidation is based on virtualization technology, which is a strong area for AMD’s Opteron processors. This was clearly demonstrated recently, when a cluster of HP ProLiant BL465c Gen8 servers using AMD Opteron 6200 Series processors achieved the highest VMmark 2.1 score ever posted, representing a 40% increase over the next best score.

I believe that data center consolidation inside the firewall or in public and private clouds is a big opportunity for AMD.

DCDF: How are you planning to tackle the IaaS provider market in the near future?

SG: Most Infrastructure-as-a-Service (IaaS) providers offer virtualized server instances. We allow lower infrastructure acquisition costs, which translates to lower cost per virtual machine.

Our goal is to improve the profitability of IaaS providers by offering them lower TCO (total cost of ownership) when using AMD.

DCDF: What are the key ingredients of AMD’s strategy in the high-performance computing (HPC) market?

SG: We believe that heterogeneous system architecture (HSA) brings the right balance of CPU (central processing unit) and GPU (graphics processing unit)-based computing to the HPC market. Therefore, HSA will be the key ingredient of our HPC strategy.

DCDF: Following the SeaMicro acquisition, AMD was saying it would not become a hardware vendor and that it made the acquisition strictly to gain access to the Freedom Fabric technology and make it available to its OEMs (original equipment manufacturers). Yet, in September AMD launched the latest SeaMicro server. Can you clarify the company’s strategy in relation to the acquisition?

SG: The SeaMicro acquisition gave us access to the Freedom Fabric technology, and we are committed to making this technology available to our OEM partners.

There are two reasons why you see system-level announcements from AMD: first, we plan to honor roadmap commitments to SeaMicro’s customers and second, bringing out servers in the roadmap is important to demonstrate the capabilities of the fabric, while we work with OEMs to bring the fabric technology to end customers.

DCDF: Why sell SeaMicro SM15000 servers with Intel chips?

SG: SeaMicro was already selling servers based on Intel chips and had a pipeline of Intel-based products at the time of the acquisition.

In response to customer demand, we continue to execute on that roadmap. At the same time, we will ship SeaMicro servers with AMD processors before the end of this year.

DCDF: What is your philosophy on “wimpy-core” processors in the data center?

SG: Companies build data centers with certain types of compute elements to meet specific business needs, all within the best economics.

I believe that in the future, data centers will be built out of different types of compute clusters that are tuned to offer the best economics for specific sets of workloads. In this context, it is possible to visualize “wimpy-core” processors to be used to address workloads that are not latency-sensitive in an energy efficient way.

[AMD announced plans for wimpy-core ARM-based processors after this interview]

DCDF: How big is the original design manufacturer business for AMD in comparison with original equipment manufacturer? Do you sell processors directly to large end users like Google, Facebook or Amazon?

SG: We work with a variety of customers and support their favorite ways for acquiring servers, but I cannot comment on how we work with specific customers.

Our work on Open Compute platforms is a clear example of our ability to work flexibly with customers. (The Open Compute Project is a Facebook-led open-source community for hardware and data center design.)

DCDF: How is the concept of software-defined data center (SDDC) going to change the server market?

SG: SDDC is an extension of the growing trend to virtualize assets within a data center and manage it as a whole, instead of managing servers, switches and storage as standalone entities. It takes advantage of policy-based automation of data center operations and therefore has the potential to simplify how data centers are operated. This increases business agility and lowers data center op-ex (operating expenses).

Server virtualization is very advanced, compared to virtualization of other components of the data center. Therefore, I see SDDC changing the non-server components of the data center more significantly than servers.

Server processors will continue to get better at supporting virtualization. And as SDDC advances, I anticipate that servers will have to support a richer array of management, provisioning and monitoring APIs (application programming interfaces).

DCDF: Does software-defined data center have a place in AMD’s strategy?

SG: Servers have to play well in the context of SDDC, so we will continue to work on improving how our processors support virtualization.

And as we drive towards denser and denser server form-factors, we will work closely with leading ISVs (independent software vendors) in the SDDC market to adequately support the management, provisioning and monitoring requirements.

A version of this article first appeared in the 24^th issue of the DatacenterDynamics FOCUS magazine. Visit the FOCUS registration page for a free subscription.