Much of today’s server performance and consolidation initiatives hinge on the features and capabilities of emerging server processors. Fortunately, we live in a time when two vibrant chip manufacturers, Intel Corp. and AMD Inc., are competing to add more value and functionality to their processor products. Upstarts such as ARM Ltd. and other “boutique” chip and architecture developers continue to push the envelope in terms of form factor, size, power consumption and efficiency.
Let’s take a look at some trends in server-oriented processor design for the next year or two, and then examine some of the upcoming processor developments from Intel and AMD to see how those trends may influence the number and capabilities of servers in your next technology refresh cycle.
Server processors: Trends for 2011
This year will likely prove to be an incremental one for server processors. The world is focusing on tablets and mobile devices, and most of the heavy research and development expenditures are on that side of the field. I expect that server processors will see enhancements in four main areas this year and next:
- Improving the hypervisor -- Virtualization is huge. The biggest phenomenon in a long time, and enhancing the way that the hypervisor interacts with the operating system (OS) and other system components can improve virtualized server performance and capacity. You’ll start to see other system components, such as network cards, become virtualization-oriented as well. This trend will make it easier to set up vast swaths of machines with complicated networking and configuration requirements, such as those for a cloud.
- Demand-driven CPUs -- The idea here is to combine capabilities like Hyper-threading, where an OS sees two virtual processors for every physical CPU (unlike dual cores), with the dynamic power-driven capabilities of dual-core processors, as well as an OS similar to Windows Server 2008 R2. Loads will become dynamically managed among demand-driven CPU sets, causing power consumption to decrease while maintaining a balanced load and the speed expected from applications.
- HyperTransport -- HyperTransport is an interconnect bus that connects various components of a computer or a processor to other components. HyperTransport most directly affects bandwidth and processing among cores on a chip, and from the cores themselves to memory. Obviously, higher speeds and better links between those components make for faster computations and more efficient workload distribution among components. AMD is a primary user of this technology and is integrating fast HyperTransport speeds into their new server processors.
- Graphics additions and supplemental instructions -- Intel is leading the way in these aspects. You’ll read more about the company’s new chip line in the next section.
Intel’s 'Sandy Bridge'
Intel took CES by storm when it announced the second-generation core processor chips on Jan. 3, 2011. These chips are meant for a wide variety of systems, including laptops and desktops, but have server interest as well.
The Sandy Bridge processors will include a new, 256-bit feature called Advanced Vector Extensions (AVX) and OS support, which, at this time, is limited to Windows 7 and Windows Server 2008. The AVX is a new extension to the SSE instruction set that is specifically designed for floating-point intensive applications. The AVX set will allow for faster and more efficient processing of image data and help in audio and video processing, simulations, modeling and financial analytics.
What’s interesting about these Sandy Bridge processors is the media focus. Typically, that’s a minor point in server CPU circles, but the recent trend of using dumb or thin clients with RemoteFX-enabled Windows Server 2008 R2 Terminal Servers has put a new emphasis on I/O. Bringing dumb or thin clients in the mix would require graphics hardware capabilities to render complex graphical sequences on the server and direct it back to the client over the wire. Therefore, putting a capable graphics engine directly on the processor becomes useful for desktop virtualization. Sandy Bridge processors have a built-in graphics engine that shares the L3 cache, which should increase efficiency and use less power. The processor handles HD content as well as 3-D and should be good for media streaming and visualization.
Time will tell whether Microsoft will support the graphics engine on the Sandy Bridge chips for RemoteFX, but the trend is certainly there. It offers intriguing potential for folks who crave a bit of virtual desktop infrastructure (VDI) capability in their shops but have users with animation and rendering needs that are too much for a thin client to handle.
AMD’s plans have centered on developing a new CPU architecture known as “Bulldozer,” which is to be introduced this year, while keeping its existing Opteron processors on the market for existing users and customers today. The Bulldozer architecture emphasizes more processing threads, lower power usage and better memory access and bandwidth. AMD is planning a suite of energy-efficient chips that will scale to 10 cores, use less power than before and are intended for scale-out based data centers serving cloud applications.
The AMD roadmap is split essentially into two categories: CPUs for two- and four-way (2P and 4P) servers for mainstream and enterprise platforms, and for one- to two-way (1P and 2P) servers designed for scale-out endeavors where cost and energy optimization is paramount. On the 2P and 4P side of the spectrum, the view is pretty clear:
- In 2010, The AMD Opteron 6100 series was commercially available. Chips can be found with eight or 12 cores, a 12 MB L3 cache and quad-channel DD3 memory bus support.
- This year, the “Interlagos” CPU emerges with the new Bulldozer-based CPU cores, which should be available in eight, 12 and 16-core-per-chip combinations.
- In 2012, AMD’s Terramar chip promises to add even more cores per processor chip. Estimates suggest 20 cores per chip by the end of the year.
If you look at the 1P and 2P side of the AMD spectrum, you see essentially the same ideas, just with halved capacities:
- In 2010, we see the AMD Opteron 4100 series processors with four and six-core options, 6 MB of L3 cache and dual-channel DDR3 memory bus support.
- This year, the “Valencia” CPU comes onto the scene with the new Bulldozer cores in 6-per and 8-per core options.
- Later into 2012, we’ll find the “Sepang” CPU brought to market with up to 10 next-generation Bulldozer CPU cores.
AMD appears to be embracing the capacity and scale side of the equation, while Intel is focusing on enhancing the end user experience with tremendous power, speed and media focus. No matter your vendor of choice, the relentless evolution of modern microprocessors should continue to accelerate server power and utility.