March 13, 2012
HPC installations make use of the best chips available, but performance doesn’t rely solely on processors and memory. The network connecting each compute node serves as the backbone for a supercomputer. Last week in a press release, IBM announced a prototype chip to make that backbone a little stronger.
The Holey Optochip is a parallel optical transceiver chipset capable of transferring 1 terabit of data per second. Compared to copper wires using electrons as the data medium, optical networking provides much higher rates of transfer via pulses of light.
To get some sense of its performance, the IBM press release provides this context: “The raw speed of one transceiver is equivalent to the bandwidth consumed by 100,000 users at today’s typical 10 Mb/s high-speed internet access. Or, it would take just around an hour to transfer the entire U.S. Library of Congress web archive through the transceiver.”
The manufacturing process involves creating 48 holes, called optical-vias, on a single 90-nanometer IBM CMOS die. Optical access is gained through the holes to the back of the chip, where 24 transmit and 24 receive channels are located. Photodiode arrays and vertical cavity surface emitting lasers (VCEL) are then soldered to the chip itself. The resulting Holey Optochip is then ready to couple with a 48-channel multimode fiber array.
For all the positive innovations the company mentions regarding the Optochip, its underlying technology uses parallel optics, where bandwidth is maximized by providing full duplex connectivity over multiple fibers. While this technology can result in higher speeds of connectivity compared to traditional fiber technology, the maximum operating distance is usually less than 150 meters.
Along with transfer speed and manufacturing innovations, IBM also touts the Holey’s power efficiency. The device requires fewer than five watts to operate, which they say is necessary to reducing the power consumption of data communications. According to the researchers, the technology “represents the first practical demonstration of an optical interconnect that attains the efficiency levels that will be required for exascale computers circa 2020.”
IBM also makes a point to mention that the chip components are currently available today, which should result in cheaper production costs. There are no definitive plans to produce the Holey Optochip in the near-term, however the researchers estimate commercialization could take place over the next decade.
Crossing the 1 Tbps threshold will certainly open new doors for a multitude of HPC applications, and is especially relevant for exascale computing. The Holey Optochip’s impressive power consumption and data performance are the key attributes in this regard. However, there is plenty of time before exascale computing becomes a reality, and there are a number of competitive technologies cooking in research labs elsewhere that are offering similar capabilities.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?