Intel's First Server Chips With 3D Transistors Coming This Quarter
Intel said on Wednesday it will release its first Xeon server chips with 3D transistors this quarter, in a move that analysts said would intensify the cloud hardware battle with rival Advanced Micro Devices.
The company's new Xeon chips will be based on the upcoming Ivy Bridge microarchitecture and initially will be aimed at the emerging category of microservers, Intel said. Microservers are low-power servers with shared components designed mainly for Web serving and cloud applications.
The new chips will replace existing Xeon E3 chips, which are based on the Sandy Bridge microarchitecture. Intel introduced the E3 chips in March last year to jump-start the microserver category, which is still in its infancy but expected to grow along with Web and cloud activity in data centers.
Intel currently dominates the data center, and the company's chips are used in a majority of servers that ship today. But Intel has been putting more resources behind microservers as a low-power alternative to traditional racks, blades and towers, which handle more intense workloads such as databases.
The new Xeon microserver chips will outperform their predecessors while drawing the same amount of power, the company said. The driving factors include 3D transistors, which are part of Intel's new 22-nanometer manufacturing process. Intel has claimed that 3D transistors will consume a little less than half the power and be 37 percent faster than Intel's existing 32-nm process chips, which have 2D transistors. The 3D transistor technology, called tri-gate transistors, replaces a flat, two-dimensional arrangement of transistors with a 3D structure that rises up from the silicon substrate.
Enthusiast website Anandtech has measured a 5 percent to 15 percent improvement in CPU performance with Ivy Bridge compared with Sandy Bridge.
Intel's attempts to expand its presence in microservers included a partnership with dense server maker SeaMicro, which was snapped up by AMD for US$334 million in late February. That acquisition was viewed as a setback for Intel, which hit back by saying it was developing its own I/O and chip technology to boost microserver performance.
Microservers will be more relevant as data centers try to operate within physical and economic constraints, said Dan Olds, principal analyst at Gabriel Consulting Group.
"When you start looking at high-performance computing, you come to the realization that the limitation is power. It's inescapable that that becomes the final constraint," Olds said.
Intel and AMD are chasing the microserver market to address those limitations. Intel has held the high ground on performance, while AMD largely competes on price, Olds said.
AMD recently launched the Opteron 3200 server chips for microservers, which the company pitched as a "low-cost-per-core" product. The AMD 3200 chips are priced between US$99 and $129, while Intel's E3 chips are priced starting at $189. Analysts have said that AMD will also replace Intel chips currently being used in SeaMicro servers with Opteron chips.
Intel's new chips bring more performance, but microserver adoption may ultimately depend on what the end application is, said Dean McCarron, principal analyst at Mercury Research.
In some cases, price may matter less for companies that need scalability, McCarron said. Different microserver designs are just reaching the market, and there is intense competition between Intel and AMD for those building mega data centers, in which thousands of servers are deployed to process fast-moving cloud applications, McCarron said.
Intel also said it was on track to release its low-power Atom chip for microservers in the second half of this year. The 64-bit chip will draw six watts of power and have all key server features, including virtualization and ECC memory. The chip will be made using a 32-nanometer process, so it will not include 3D transistors.