AMD’s new server chief has laid out a plan to help the company compete once again in the server space.
Suresh Gopalakrishnan, corporate vice president and general manager of servers for AMD, said he plans to use this week’s VMworld conference to meet with customers and talk over that strategy. His ammunition? Tests that show AMD can save customers 30 percent per virtual machine or $130,000 per rack—at least in a head-to-head comparison between the AMD Opteron 6248SE and SE2 and the Intel Xeon Model 2690.
To put it another way, AMD argues that Dell and Hewlett-Packard servers using its Opteron chips cost about $500 or so per virtual machine for an 800-VM implementation, versus just under $800 for the Intel Xeon systems.
Ironically, AMD helped standardize the server market onto the X86 architecture with the 2003 launch of the Opteron, which ushered 64-bit computing into a space traditionally dominated by proprietary RISC machines. By 2007, however, things had started going south. AMD focused less on server innovation, Gopalakrishnan said, and the company’s message became more about the price/performance mix rather than leading at the high end of the performance index.
Just over a year ago, AMD ousted Dirk Meyer as chief executive and replaced him with Rory Read, who restocked AMD’s upper management. Gopalakrishnan has been on the job just ten weeks, having served most recently as vice president of engineering at Extreme Networks, a high-performance switching company for the data center.
“We’ve been selling into the data center for a long time, and talking to customers, quite often I could see their challenges to come and transform the data center,” Gopalakrishnan said in an interview. “And at that point I’m selling switches. But looking at the whole thing is not just about switches, or servers, or storage. We had to look at them as a whole to solve this.”
AMD needs to do three things, Gopalakrishnan added: “Execution, both on products and go to market. And then what kind of innovation do you want to bring out in the market. And then, part three, what kind of investment do you need to kind of help fulfill that mission. The good thing is that the investment was already done by the time I came in here.”
That investment was in SeaMicro, a startup designed around the Xeon processor. SeaMicro has optimized its architecture for dense computing, power-managing as many components as possible and eliminating all but three components from its motherboard—CPU, DRAM, and I/O—helping to reduce power, cost and space.
Gopalakrishnan claims that AMD is in the midst of designing for a future where low-power options and co-processors assist the traditional computing node.
“The servers in the future are going to be heterogenous compute clusters,” he said. “Which means that these servers will not be one size fits all; there will be different kinds of processors going in there these clusters. There will be some workloads where an energy-efficient CPU can do it, but that CPU cannot do your heaviest workloads.”
That results in CPUs with different performance levels, he continued, and different energy levels for these clusters. “And then you have GPU assisted computing, for the highest performance or something where you have GPU-assisted computing. So server becomes this heterogenous compute cluster.”
The question, of course, is where AMD will find those low-power options. The company disappointed analysts when it failed to sign an ARM license, instead agreeing to a partnership in the HSA Foundation. But Gopalakrishnan points out that AMD sold its efficient “Trinity” APUs (accelerated processing units) into small business servers from HP and other vendors; on top of that, AMD has a long-standing presence in graphics chips despite competition from Nvidia and Intel. Expect to see Opteron-branded APUs, Gopalakrishnan said, while declining to comment further on the roadmap.