Stay on Top of Enterprise Technology Trends
Get updates impacting your industry from our GigaOm Research Community
The other shoe has dropped. When Hewlett-Packard(s hpq) unveiled plans for ARM-based low-energy servers last November, it said to expect versions based on Intel’s (s intc) Atom processors. Well, they’re still coming.
The server giant today promised new low-energy servers — under the Project Gemini code name — that will use brand-new Intel 64-bit Atom Centerton chips, which are actually systems on a chip or SoC. There were big promises about energy efficiency. “A standard [comparable] server today would use 150W. The new servers will do the same load in a 12 to 14W envelope,” said Paul Santeler, VP and GM of HP’s Hyperscale Business Unit.
These new servers, due by year’s end, will allow the use of “pop-in” server cartridges to let users update or modify the system for different workloads within the same system. “This will be one server we can change to address workloads depending on need,” Santeler said. “We are doing this in processor neutral standpoint starting with Intel’s new Centerton chip,” he said.
According to HP’s press release it went for Centerton first:
due to the processor’s data-center-class features, such as 64-bit support, hardware virtualization (VTx), error correcting code (ECC) memory, lower power requirements, increased performance and broad software ecosystem. These features, coupled with the revolutionary Gemini infrastructure, make the new Centerton-based servers ideal for hyperscale workloads, where using many extreme low-energy servers densely packed into a small footprint can be much more efficient than fewer standalone servers.
IBM(s ibm) has promised workload configurable servers with its upcoming Pure Systems lineup announced in April..
Jason Waxman, GM of cloud infrastructure for Intel’s Data Center group, said there are many uses for this type of server.
“There are new CDNs driving a lot of I/O, there’s an opportunity to add SSD (solid-state drives) in combination with Centerton to solve that problem. And in analytics, there are Hadoop applications, some where I/O is the bottleneck and if they can only afford 1 gig Ethernet network, that’s a good use for SSD and a shared infrastructure chassis,” he said.