Comment HPE will use a research grant awarded today by the US Department of Energy to develop blueprints for a Machine-based exascale supercomputer.
An exascale system is a beast that can hit at least one exaflops – a billion billion floating-point math operations per second. Uncle Sam is bent on getting at least one of these mega-machines up and running on US soil by 2021 and in production the following year.
These computers will be so powerful, they could each catalogue an iTunes music folder in less than ten minutes.
HPE is, essentially, getting a fistful of coins from a $258m pot provided by the Department of Energy to research and develop technology needed to build screaming-fast computers for scientific and national security applications.
We might suppose the US would quite like to build an exascale system before China, possibly regarding it as being as important as the early space race, which saw the USSR launch its Sputnik satellite in 1957, before the US launched its first space satellite.
The Register wrote about the first round of grants in September.
Using this latest cash injection, HPE will dust off its so-called memory-driven computing architecture and try to turn it into something that could be possibly used within a next-gen supercomputer. This architecture is a portfolio of technologies, including a memory fabric and low-energy photonics interconnects, that Hewlett Packard Labs developed under the codename The Machine. An ARM-powered prototype of The Machine, said to be the world’s largest single memory computer, was demonstrated in May.
This beast was a far cry from the memristor-based wonder system we were all promised years ago. The Machine has been, effectively, relegated to an internal lab experiment, with some of its techniques and designs filtering into future shipping gear and things like HPE's supercomputer efforts. The Machine as a standalone product isn't going to happen anytime soon, but that won't stop HPE touting parts of the radical tech as a solution for stuff like exascale machines.
HPE says Gen-Z – which provides a memory-semantic chip-to-chip communications protocol that allows for the tight coupling of many devices including CPUs, GPUs, FPGAs, DRAM, NVM, system interconnects and a host of other devices, all sharing a common address space – will also play a role in this. ®