AMD today launches a new series of server CPUs. These CPUs are fabricated on a 32 nanometer process, and can be used for servers and supercomputers. Data centers now consume approximately 3% of all electricity produced, and that number will inevitably increase during the coming decade. As a result, more efficient CPUs have the potential to substantially ameliorate the problem. In an interview with Sander Olson for Next Big Future, AMD’s Product Marketing Manager Michael Detwiler describes the features and capabilities of the “piledriver” Opteron 6300 CPU series. AMD is hoping that the 6300 CPU series, which can be used in everything from servers to supercomputers, will change the server and high-performance computing landscape and return AMD to profitability again.
Question: The Opteron CPU has been out since 2003. How has the architecture evolved since then?
The first AMD Opteron had a revolutionary architecture, with an integrated memory controller and eliminated the front-side bus. It was a single-core processor that initially did very well in the High Performance Computing market. Over the past 10 years, we have seen a huge rise in multi-threaded applications, virtualization adoption, and cloud computing in which highly scalable energy efficiency processors are needed to deliver the required level of performance. Those requirements led us to our current architecture, which is a modular design with up to 16-cores per processor.
Compared to the AMD Opteron 6200 Series processors, the 6300 series has higher IPC (Instructions per clock), higher frequencies, and also a mature software ecosystem in which code has been written to specifically take advantage of our architecture. These improvements add up to a 24% increase in java performance that is important for many applications such as Hadoop and 40% improved performance per watt. Both of these are important metrics for cloud customers, which is one of the fastest growing markets and one of our targets.
Question: What are the specs of the 6300 series?
The AMD Opteron 6300 processors is our flagship server product. It is supported in both 2 and 4-socket configurations with 4, 8, 12 and 16 core options. It operates from 1.8-3.5 GHz depending on sku and is available in 3 power bands ranging from 85W to 140W. There are 4 channels of memory, and in 4-socket configurations, a memory capacity of 1.5TB is supported. The G34 socket, power bands, and chipsets are consistent with the AMD Opteron 6200 series, so existing AMD customers don’t have to redesign their power and cooling infrastructures, and OEM partners can easily carry past designs forward if they choose to do so.
Question: Data center power consumption is now nearly 3% of all energy use in the U.S., and is rapidly rising. Is this trend reversible?
Data creation is expanding exponentially so it’s critical to focus on energy efficiency. Smartphone and tablet adoption drives a need for people to access data at a second’s notice. The social media boom isn’t slowing down. All of these factors mean datacenters are going to continue to grow, so greatly improving their energy efficiency is an important goal. The AMD Opteron 6300 series processor has 40% better performance per watt than the 6200 series. This means it can do 40% more work at a given power draw. Our goal is to continue to drive that number in the right direction so that datacenters can continue to meet business demands while minimizing any additional power consumption.
Question: How much will the 6380 cost?
The 6380 is our top performing sku in the mainstream power band, and its 1Ku price is $1,088. As far as the 6300 Series as a whole, the 1Ku prices range from $415 to $1,392. We believe this gives us a very attractive price-performance value proposition. For instance, the AMD Opteron 6380 is performance competitive in many HPC applications against the Intel Xeon E5-2690, yet the 6380 1Ku price is $1088 vs. $2057 for the Intel processor.
Question: Is this CPU designed for High-performance computing?
HPC is certainly one of the target markets for the AMD Opteron 6300 Series and is a market where AMD has a proven history of success, but not the only one. It is also suited very well for public and private cloud as well as Big Data and virtualization. For HPC, our well balanced architecture of I/O, memory bandwidth and compute capabilities combined with our attractive price points have led to our success in this space. HPC is also a market that likes to be on the edge of technology, so that is where we typically see early adoption of our products. For example, Cray and Indiana University recently announced the intention to build the fastest university owned supercomputer based on the latest AMD Opteron processors. It’s called Big Red II, will have over 21,000 cores and will be used for a variety of research opportunities.
Question: Is there a desktop CPU equivalent to the 6380?
The desktop and server markets are completely different and are asked to perform completely different tasks. The AMD Opteron 6380 is a 2 and 4-socket computing powerhouse that is asked to solve complex scientific problems, perform real-time data analysis, consolidate datacenters through virtualization, and power cloud datacenters. There isn’t a desktop equivalent to that because you would never ask a desktop processor to do those things, nor would you attempt to measure it.
Question: Are there any plans to create a GPGPU card with the 6380?
We’re not going to create a GPGPU card that only works with a specific type of processor, so there won’t be a card that is created just for the AMD Opteron 6380. But we do have GPGPU solutions that are in the market and can be run alongside AMD Opterons as well as our competitor’s CPUs.
Question: How long has the 6380 been in production?
We have been in production for about a month on the AMD Opteron 6300 series. We do this so that there will be immediate availability once we officially launch the product.
Question: Does the 6380 support next-generation DDR4 memory?
The current standard in the server market is DDR3 memory, and that is what the AMD Opteron 6300 Series supports. We support up to 1866Mhz memory in the right configurations, but we expect the mainstream server market to use 1333Mhz or 1600Mhz speeds, depending on their needs and budget. We will evaluate the memory technology roadmap when making any memory support decisions for future products.
Question: What changes do you anticipate for the server market in the next five years?
I think you will continue to see growth in the large cloud datacenters as the smartphone and tablet markets grow and more people become connected to the internet across the globe. That growth will lead to a strong push for more efficient datacenter design as well as more efficient servers, which has clearly already started. AMD is on the right path to meeting those demands through our investments in low power technologies and focus towards maximizing the efficiency of our processors.