• What technology does Progeniq's accelerators use?
• When Progeniq launches new accelerated engines, do I need to purchase new hardware?
• Wouldn't the accelerator hardware be outdated after a couple of years, and will the CPU core performance catch up?
• What is the advantage of using an accelerator vs more CPU cores?
• Wouldn't I be able to match performance of an accelerator by just throwing in more CPU cores into the cluster?
• How does using Progeniq's accelerators compare to using GPUs or other fixed function accelerator chips (ASICs)?

Progeniq's accelerators use Field Programmable Gate Arrays (FPGAs). These are reconfigurable hardware processors, which will 'rewire' themselves on the fly when you run the application you wish to accelerate.

No, the accelerator hardware is reconfigurable, so it's as simple as downloading the new software and installing it in your system. What's more, if you're on the annual license, you'll get all the new accelerated engines free of charge when they're released.

Both the CPU and FPGA processor advance with Moore's Law. As the ability to add more transistors into a chip doubles every 18 months, the number of CPU cores per processor increases, and so does the logic space available in an FPGA processor. Both accelerators and CPUs have similar hardware refresh cycles; just as you would purchase a new CPU cluster every 3 years, you would also purchase new generation accelerators with that cluster.

Beyond giving you the speedups over CPUs, accelerators excel very well in space and power consumption. An accelerator requires just 2.5W, where a server usually requires 700W of power. Using accelerators gives you the best performance per dollar in acquisition and ongoing running costs, and allows you to scale up your cluster without hitting server room electricity or space constraints.

Very often, the answer is no. The clockspeed of a CPU core has remained constant for the past few years at 2.4GHz, and the CPU companies are focused on adding in more cores into a single processor. Many algorithms aren't able to take advantage of multiple cores, and these CPU cores are often leveraged on by splitting up the data for processing into multiple sets, one set targeted at each CPU core. However, in an accelerator, the algorithm itself is accelerated, giving you a speedup within a job itself.

GPUs have their strengths in certain algorithms, but they are extremely power-hungry. A high end GPU board can take up to 200W of power, which makes it infeasible for any large scale deployment in a server cluster. ASICs have their strengths in being able to offer higher efficiency, however they key drawback is that they are fixed in function. An ASIC can accelerate only the single algorithm it was designed for, and there can't be any future enhancements or new algorithms accelerated. This inherently makes it too high cost an investment unless an end user runs only a single algorithm 24/7 and nothing else.