The machine is to ahve specialized cores, a purpose-built open source operating system optimized for non-volatile systems, and was to have: memristor non-volatile memory, a special kind of resistor circuit that functions as both storage and memory.
HP will use DRAM memory for its prototype, and will convert the shared memory pool to non-volatile memory, for example phase change memory, in future versions.
HP had said the Machine would rely on a memristor, which is a kind of digital memory that has been talked about for four decades, but has been difficult to bring into commercial use. The company expects to spend close to $500 million on the Machine as it develops the many-years-off computer, putting it on a par with its move from selling computer servers to offering cloud computing systems.
Martin Fink, HP’s chief technology officer, repositioned the Machine as a “memory-driven computer architecture,” which focuses on the large amounts of data stored, rather than the processing power.
Memristors were barely in sight. A prototype of the new computer could be out next year, Mr. Fink said, based on more conventional DRAM memory. Instead of a special-purpose computer operating system, he said, the Machine will initially have a version of the popular Linux system.
Memristors are still on the table and HP is aiming to have them inside the system when it makes its market debut five years from now.
Next year’s promised Machine could still be impressive. Mr. Fink said it would have 320 terabytes of memory, compared with 12 terabytes in the most memory-rich computers HP now offers.
If HP can build a market for the tech, it wants to shrink down the Machine, Mr. Fink said. Devices that now rely on cloud computing for much of their functionality, like smartphones, could become self-contained objects, capable of memory-intensive things like voice recognition and language translation without calling on external computers.
Specialized processing was one of the hallmarks of the original announcement. The right compute for the right workload would make it possible to achieve a factor of six times performance increase using 80 times less energy, HP said a year ago. Since repositioning the Machine as a “memory-driven computer architecture” last week, the messaging has focused more on the democratization of fast memory and less on processing power. While power-efficient memory is crucial for reaching computing milestones, such as exascale, it was the combining of component technologies into a single project that made the Machine such a radical departure from the status quo.
SOURCES – NY Times, HPCwire, Youtube, HP