Enterprise, Event, Graphics, Hardware, NAB 2013

Fusion-io Fills in the Gap with New 1.6TB ioFX Card

5 GHz 16-core CPU? Check. NVIDIA workstation GPU with 5000 cores? Check. And tons of RAM to coordinate it all. But can your SSD storage keep up with this powerhouse? Fusion-io knows it can with their ioFX storage acceleration cards.

The ioFX is a PCI-E NAND flash memory expansion device that balances out the graphics processing workstation, making sure that each of the system’s cores is fed quick enough resources without any bottlenecking. This new model has a whopping 1.6 TB capacity, four times bigger than their 420 GB model introduced last year, complemented by a 1.4 GBps read and 1.1 GBps write speeds.

NVIDIA?s booth at this year?s NAB Show already demonstrates the raw power of this device, driving four 4K displays in real time off of a single ioFX card. Several partners have already joined Fusion-io in both systems integration and SDK development.  Most notably, HP will soon offer these ioFX cards in their Z820, Z620, and Z420 workstation systems.  

The ioFX cards are fully streamlined with NVIDIA?s CUDA-enabled GPU?s, using commands such through their SDK to stream data right off the device itself, bypassing the need to go between storage controller, and memory sources. The ioFX can be used as an extension of the OS, so that the NAND flash can be used as a memory tier.  The OS treats the ioFX as a memory allocation device, but is also seen by the OS as a storage device, emulating storage tasks directly to graphics processing. This means that any application that can already uses GPU resources can transfer high bandwidth data at a much quicker rate. 

As soon as the application makes a storage command, it?s converted to a memory command, sent to send to the ioFX, which sends a memory address back, and feed back to the software, all before touching storage. When dealing with SSD?s in RAID, these same commands need to go through storage controllers, RAID controllers, then each individual SSD chip before it?s converted from memory to storage, and back again. Because the ioFX can be used for its memory allocation or storage capabilities, users will benefit from ultra low latency.  The typical response time of a command sent to an SSD is 1 ms.  With these new ioFX cards we?re looking at 0.06 ms, which means around 20 files delivered in the time it takes for one file on an SSD.

What about all the SSD?s you?ve invested in?  Sure, combining SSD?s will increase bandwidth, but only up to a certain point. When there?s more controllers and chips to query, latency will also increase, which ultimately results in dropped frames at very high resolution content in real time. Dual SSD?s give you double speed, two more may give you 80% more, and another two will give you 60% more performance.  ioFX can break through this barrier because it?s a memory device with minimal storage infrastructure at its core, which means multiple cards give you exponential growth while still keeping latency at bay.  Four of these 1.6 TB models will feed up to 6 GBps of bandwidth between your GPU and your apps. 

NVIDIA tested the ioFX devices inside a system running three K5000 devices feeding 9 monitors in a grid, creating a 6K display. When running displays in a mosaic mode, all GPU?s need to be fed all the data at the same time. So these data copies are fed to system memory, and then to each GPU individually.  Using the ioFX device as storage, and pinning its resources at the same address as the GPU?s, they could send the same copy of data to all GPU?s simultaneously.

Having an intelligent way for data to travel from large capacities into memory is very beneficial, especially at Ultra HD resolutions and high frame rates.  The Hobbit?s stereoscopic 4K 48 FPS footage is 24 times more data than a standard 2K movie.  Avatar 2 is said to be delivered at 4K stereoscopic 60 FPS, which will increase this need even further.  Fusion-io believes that the ioFX is the answer to making sure no high resolution content is left behind, in the most demanding environments.