The computer has 160 T-bytes of reminiscence, and HPE expects the architecture to permit reminiscence to scale as much as 4.096 yottabytes.
The reminiscence is unfold throughout 40 bodily nodes which can be interconnected using a excessive-performance cloth protocol.
The pc runs on an optimized Linux-based working device walking ThunderX2, Cavium's flagship 2nd-technology twin socket-succesful ARMv8-A workload optimized gadget on a Chip.
It makes use of photonics/optical communique links, which includes the brand new HPE X1 photonics module.
The pc has software programming tools designed to take gain of plentiful persistent reminiscence.
The generation is constructed for the big information era, HPE stated.
One length fits All
"We think we've got were given an structure that scales all of the manner from part devices -- the sensible facet -- through a few bigger structures which are probable more in step with what we have constructed out as a prototype," said HPE Fellow Andrew Wheeler, deputy director of Hewlett Packard Labs.
"if you think competitively against that scale of things, you're going up in opposition to conventional supercomputers, that have their scale limits and could have electricity intake tiers 10 to twenty instances what we've here," he advised TechNewsWorld. The 400 nodes and one hundred sixty T-bytes of memory "all match very effectively inside a unmarried [server] rack."
HPE has discussed the generation with masses of customers, it said, such as industry verticals, high-performance computing companies, analytics companies, economic institutions and others, Wheeler said, noting that "all and sundry we talked to absolutely resonates."
The memory-driven Computing difference
HPE's memory-driven laptop gives "notable speedup," Wheeler said, due to the fact everything resides on the reminiscence fabric.
HPE is working on that cloth as a part of the Gen-Z Consortium, which includes ARM, AMD, Cavium, Broadcom, Huawei, IBM, Lenovo, Micron, Dell EMC, Seagate and Western digital amongst its participants.
The modern von Neumann architecture has computer systems moving records "all around the region," referred to Paul Teich, a primary analyst at Tirias studies. "Even if you're running with one of the huge [Software as a Service] suites, they spend loads of power and time shifting records around through the processors."
Having one hundred sixty T-bytes in a single area we could customers "go away all of the information in reminiscence, point to an address, and everything happens automagically," Teich told TechNewsWorld.
"the whole lot will become loads quicker and more fluid," he defined, "and given that your number one big data dataset does not trade, the power you store by way of not shifting that statistics out of garage into dozens or masses of machines is huge. Instead of having to transport the facts in the direction of a processor, you have got the processor towards the facts."
any other aspect of the story
The memory-driven pc "is an atypical blend of generation, warmth use of optical, and flash, so reminiscence speeds need to be first rate, however there's no GPU, and [it uses] ARM CPUs, so processing will be surprisingly gradual," stated Rob Enderle, fundamental analyst at the Enderle institution.
"This factor should deal with an outstanding quantity of information right away so long as you are not doing that a great deal with it," he told TechNewsWorld.
"I don't see the sector placing all fitness statistics in a unmarried gadget -- or facebook all its information," remarked Holger Mueller, a important analyst at Constellation research.
"maximum of the big information use cases we realize these days are quality with HDD, or HDD with a few reminiscence powered with the aid of Spark," he informed TechNewsWorld. That leaves "constrained room for the new presenting for most effective very excessive-value, high price-justifying use instances."
cost can be the principle thing using the marketplace, Mueller recommended.
Intel has launched 3D point memory, branded as "Optane," which also is continual, but it plays at near DRAM speed, Enderle pointed out.
That "probable makes the HPE effort obsolete earlier than it ships," he said. "Had [HPE] added this out in 2015 whilst it became anticipated, it might have been far extra exciting."
EmoticonEmoticon