Making cutting edge accelerated computing platforms widely available so that developers and solution providers anywhere in the world can leverage them is key to success in the modern data center. Over the past decade, cloud computing has utilized parallel computing to accelerate performance, which requires fracturing a solution into as many parallel tasks as possible so that all the computational units can be leveraged. An example of a parallel computing accelerator is the GPU, which is an enormous collection upwards of 2,000, computational units. Think of them as a minor league stadium full of yellow minions with each minion representing a million little logic gates. However, parallel computing accelerators like GPUs have critical performance limitations if your problem is such that you can’t get all your minions working in parallel, at the same time, to solve it. Some classes of problems fit the parallel computing “minion” model very well, but many do not
Tasks like compression, video encoding, genetics, and stock trading don’t fit this parallel model. Enter the logic gate approach and programmable silicon, the Field Programmable Gate Arrays (FPGA). This is why Nimbix’s announcement to be the first to make the recently launched FPGA-powered Alveo U50 accelerator available to developers in the Nimbix Cloud is important.
The fundamental unit of computing is the logic gate. From “Hello World” to “Half-Life,” every program, when executed, resolves into an enormous collection of logic gates opening and closing at blinding speeds. How fast a program runs depends almost entirely on the number of layers of abstraction between the intent of the programmer, captured in their code, and the number of gates eventually required to execute that intent. Each layer of abstraction brings with it the requirement for more gates.
Suppose we were to take a high-level program for video encoding and we translated it directly into gates, then loaded that gate definition into an FPGA. It turns out that that gate rendered version of the program may encode the video in 1/20th the time and use 1/10th the power of a conventional CPU approach. Furthermore, it may be possible to load 30 or more parallel instances of that encoder into the FPGA. This could produce 600 encoded streams in the time it might typically take the CPU to encode a single stream. That’s the power of a gate-based, FPGA approach to a solution. FPGAs also provide the added benefits of much greater adaptability than fixed-function accelerators such as GPUs. As workload algorithms and requirements evolve, FPGAs provide reconfigurable hardware that can adapt much faster than GPU product cycles. Now imagine finding over 100 solutions like the one above already optimized to use acceleration technology!
Once considered special-purpose chips found in embedded computing environments like automotive, industrial IoT, and high-end consumer products like drones and cameras, FPGAs are becoming commonplace in the data center with a host of new and emerging workloads ripe for FPGA-based acceleration. To address this growing demand, Xilinx has produced a line of accelerator cards called Alveo targeted explicitly at the data center.
We’ve partnered with Nimbix to deliver a cloud-based FPGA acceleration platform for the masses. One that developers can access and use today with over 100 applications from computational fluid dynamics to video transcoding. Deploying new hardware is often challenging, but dynamically deploying cloud-based FPGA acceleration hardware through Nimbix has made it one-click simple. Nimbix utilizes the full Alveo product line; the U50, U200, U250, U280 accelerator cards.
Not only that, but we’re also working with Nimbix and Samsung’s SmartSSD storage group to deliver intelligent storage using FPGA-based computational accelerators. The Samsung SmartSSD is an SSD with an FPGA accelerator installed in front of it and preloaded with programmable logic that dramatically improves the drive’s overall performance to deliver a scalable speedup of big data processing and analytics. Data is processed directly on the SmartSSD, before it reaches the host CPU, reducing data movement and speeding up time to insight. The Nimbix Cloud offers data analysts, data scientists, and acceleration IP developers easy access to accelerated in-storage computing for next-generation cloud and data center applications.
It’s never been easier to take advantage of FPGA accelerator development and deployment in the cloud. Xilinx works closely with many major cloud providers to make new platforms available as quickly as possible. Nimbix is the first accelerated cloud provider to go live with the Alveo U50 card and Samsung’s SmartSSD. Customers can sign up today for a free trial.
