Peralex and SKARAB
On a recent trip to Cape Town to scout out new products for the Cyntony portfolio, the good folks at Peralex Electronics let me peek at the goings-on in their development lab. A certain 1U box with a huge FPGA inside grabbed my attention immediately...and required a double take to remind myself that I was a long way from Rt. 128 or Silicon Valley, where sighting something like the SKARAB would, perhaps, be more expected. Turns out that this nifty device is a terabit per second I/O FPGA supercomputer built for extreme computing at 40 gigabit Ethernet rates! And, oh yeah, it is designed for state-of-the-art Hybrid Memory Cube RAM and can be reconfigured in less than a second.
Peralex Authorized Distributor
Happily, Peralex was looking for a distribution partner for their innovations, so we're now the global authorized distributor of their high performance computer and passive radar technology. Read on for an introduction to the SKARAB's :
- Architecture and componentry
- Capabilities and specifications
- Callout for applications
SKARAB Specs
SKARAB is designed to tackle high throughput data processing and big computing challenges by combining low-latency 40 GbE communications with one of the largest Xilinx Virtex 7 FPGAs on the market plus over-network configurabilityand device management efficiency so that many can be used cooperatively in a computing cluster. The first application for SKARAB is to perform the FX correlations for the MeerKAT radio telescope being built in South Africa. Hundreds of SKARABs will process petabits of data in real-time, reducing the data deluge so that it can be further processed by SKA scientists.
Four mezzanine sites allow for up to 8 x 40 GbE QSFP+ interfaces and two Hybrid Memory Cube modules. Low latency is achieved by implementing the PHY and MAC in the FPGA, much faster than using chipsets. Amazingly, this system is also very power efficient, using less than 50W when running base four 40 GbE networking code.
What can you do with this much processing power and throughput in an FPGA supercomputer?
Let your imagination fly...then give me a call!
