CloudGreen-ITInnovationServerWorkspace

First AMD ‘Bulldozer’ Opteron Server Chips Ship

AMD logo top
0 0 1 Comment

The Opteron 6200 Interlagos chips offer 16 cores with improved energy efficiency over current 12-core chips

Advanced Micro Devices (AMD) has begun shipping the 16-core “Interlagos” Opteron server chip, the first of its processors based on the new “Bulldozer” architecture. AMD said that initial production of Interlagos – known now as the Opteron 6200 Series – began in August, and will start appearing in servers in the fourth quarter.

“This is a monumental moment for the industry as this first Bulldozer core represents the beginning of unprecedented performance scaling for x86 CPUs,” Rick Bergman, senior vice president and general manager of AMD Products Group, said in a statement. “The flexible new Bulldozer architecture will give Web and data centre customers the scalability they need to handle emerging cloud and virtualisation workloads.”

Valencia And Zambezi For Mainstream

The Interlagos chip family will offer from 12 to 16 cores and will target servers with two to four sockets which will be used primarily in large supercomputer installations that are already being built, according to AMD.

Upcoming Bulldozer-based “Valencia” Opteron chips will hold six to eight cores and will be aimed at one- to two-socket servers. In addition, AMD officials are also prepping “Zambezi”, a quad-core Bulldozer-based chip for high-end desktop systems.

The shipping of Interlagos continues what has been a busy year for AMD, which in January launched its first Fusion accelerated processing units (APUs), which offer integrated graphics and CPUs on a single piece of silicon. The new Bulldozer chips continue AMD’s drumbeat of greater performance and energy efficiency.

AMD officials have touted new capabilities in the Bulldozer-based Opterons that will enable them to offer as much as 50 percent more throughput than the current 12-core Opterons while remaining within the same power envelope. In addition, a redesigned memory controller will offer 30 percent more memory performance and a flexible 256-bit floating-point unit, according to the company.

AMD engineers also have created a number of capabilities designed to increase the energy efficiency of the chips. Among those is a feature called TDP Power Cap, which will enable users to set the TDP (thermal design power) of their chips. That capability will allow enterprises to customise their processors to meet power and workload demands.

Another capability, called Application Power Management, will manage the power budget and resources of the chip, including pushing power to specific cores when needed to keep them below the TDP setting, according to company officials.

AMD officials, touting the performance and energy-efficiency gains, have said that the Bulldozer chips will be the most significant advancement in server technology in more than a decade. They also have said they expect the technology to help keep the momentum AMD has generated this year since the release of the first Fusion chips.

In an interview with eWEEK in June, Nathan Brookwood, principal analyst at Insight 64, said businesses are continuing to look for ways to squeeze more performance out of their systems without driving up power consumption and costs. Given that, an enterprise’s decision between AMD’s Opterons and rival Intel’s Xeon chips will come down to a key metric.

“Ultimately it will depend on the performance-per-watt in the Bulldozer versus the performance-per-watt in [Intel’s] ‘Westmere’ and ‘SandyBridge’ [Xeons],” Brookwood said. “Performance-per-watt is what people are always going to look at.”

AMD’s Opteron road map is calling for “Sepang”, a successor toValenciathat will offer up to 10 Bulldozer cores that is due out in 2012. Interlagos will be followed next year by “Terramar”, which AMD officials have said will have as many as 20 Bulldozer cores.

  1. There is more to system design than playing ‘look at my Ghz…”

    Especially in virtualisation a huge role is played by L2/L3 cache sizes, true core to workload ratios, Hyper-Threading and removing the contention between components both at board and system level. At the moment AMD are getting it right in this area – you only have to look at the RAM Speed ‘step downs’ that XEON forces in dense RAM deployments as much as halving the Speed of the RAM in some instances.

    I work on Hyper-dense virtualisation platforms and and run a HPC project lab, we have standardised on the Opteron processors from AMD at the moment as in our experience they have the better performance characteristics. Check out http://tinyurl.com/bx5o73 and http://tinyurl.com/4y85uz9

    Certainly not religious about it as have done plenty of Xeon based development for post-production and other intense graphics requirements, but for the moment especially for virtualisation multi-core AMD definitely the right choice. Looking forward to getting hands on in the next few weeks.