Micron, Altera, And Servergy Join OpenPower Alliance
March 31, 2014 Timothy Prickett Morgan
The OpenPower Foundation, formerly known as the OpenPower Consortium and dedicated to the advancement and opening up of the technology related to the Power8 and future Power processors, has three new and publicly announced members. They include Micron Technology, Altera, and Servergy.
As far as I know, the OpenPower Foundation had 14 paid members as of last week, when I talked to IBM about the effort, and presumably this included those three companies mentioned above. (I talked to Big Blue before these announcements, but their paperwork was done). I have been told that there are over 100 companies who have expressed interest.
The four original members beside IBM in the foundation include search engine giant Google, which is playing around with some homemade machines, as I reported over at EnterpriseTech back in February. Motherboard maker Tyan was an original member, as was Ethernet and InfiniBand switch and server adapter maker Mellanox Technologies and GPU coprocessor maker Nvidia, which is working with IBM on a shared virtual memory architecture that Nvidia announced last week at the GPU Technical Conference in San Jose. (You can see my coverage of this feature, called NVLink, at this link.) Back in January, a new Chinese startup called Suzhou PowerCore, a spinoff of an existing maker of PowerPC chips for embedded markets, became a member and licensed the Power8 chip from IBM to eventually create its own variants of the Power chip aimed at servers, storage, and networking gear in datacenters. Samsung Electronics joined in February, and has said very little about its plans, but it stands to reason it has to do with its energy-efficient main memory and quite possibly its flash memory devices and future ARM server processors.
Now, with Micron joining, there are two major server memory makers on the OpenPower Foundation. Micron has a slew of interesting main and flash memory products, and integrating them with the Power processor is all interesting and good. The non-volatile DIMM memory, which mixes DRAM and flash on a single memory module and has an ultracapacitor power source to make sure data in memory makes it to flash when a system loses power, is interesting. And so is the Hybrid Memory Cube (HMC) 3D stacked memory that the company is gradually bringing to market. Micron may be trying to position itself as the preferred 3D memory maker, an alternative to the High Bandwidth Memory (HBM) variant that is being pushed by Advanced Micro Devices and SK Hynix.
Like Nvidia, Altera is a peddler of coprocessors, and in this case the company makes field programmable gate arrays (FPGAs) that have had this role long before Nvidia and AMD tried to position GPUs as massively parallel compute engines. FPGAs are a kind of malleable chip, allowing for them to emulate just about any kind of hardware-software combination. They are tough to program, but a lot easier to change than a physical chip. They are commonly used in the oil and gas and financial services industries to accelerate certain kinds of algorithms. Altera is using Intel as a fab partner these days for at least some of its devices and, along with Xilinx is one of two makers of FPGAs.
Servergy is an upstart, low-power server maker that came out of stealth last year pushing systems based on Freescale PowerPC chips, and it is looking for engines with a little more torque and horsepower. Ironically, Servergy says it will be creating its own Linux-based servers using IBM’s Power chips, so IBM is setting itself up for competition with one of its partners. It looks like Big Blue has figured out that a market always wants to have two suppliers.