While GPUs enable faster data processing than USCs, few companies currently have enough data to process large enough to feel the need to make the transition. As a result, cloud providers will decide when this transition will take place.
In recent years, advances in Central Processing Units have been restricted to increasing the number of hearts, the performance per watt, and the amount of RAM that can be supported. With the recent announcement of its Xeon E7 V4, up to 24 cores and 24TB of RAM , Intel continues to lead the dance to businesses to meet the analytical needs.
GPUs are the future of Big Data
However, innovation in the field of GPUs is currently much more significant. For example, Nvidia’s recent GeForce GTX 1080 , offering twice the power of the previous Titan X for half the price, is a testament to the giant leaps in the industry.
Moreover, in theory, GPUs offer a computing power far superior to CPUs. Nvidia also claims that GPUs represent the future of Big Data analysis, especially for real-time analysis of unstructured data such as video, images or audio tracks . The company recently launched its DGX-1 server, specifically dedicated to Big Data, powered by the Tesla P100 GPU.
Needs still met by CPUs
Despite these qualities touted by Nvidia, very few companies currently use Big Data systems powered by GPUs. According to Rod Fontecilla, vice president of Advanced Data Analytics at Unisys Federal, the use of CPUs for computations is for the moment quite sufficient .
According to him, Unisys Federal has not yet felt the need to use GPUs. The computing power of Spark and the Spark cluster is powerful enough to not need to change the architecture. According to Fontecilla, it is not the raw computing power that counts but the analytical model used .
The company places more emphasis on predictive models and the ability to solve problems accurately. It does not matter if the tools used take a few more seconds to give a solution, because the analyzes do not have to be in real time to the nearest second .
A decision in the hands of cloud providers
For the time being, most cloud providers use Intel Xeon. According to Intel, more than 9 0% of the servers used by cloud service providers are two-socket Xeons . As a result, the transition to GPU computing largely depends on cloud providers. Intel has the power to migrate the industry to this technology.
According to Fontecila, Unisys is closely monitoring market developments. The company looks at GPUs’ advancements and AWS’s new GPU-based offerings . The company is open to this change, but does not feel the need to change today.
In conclusion, companies that need to manage exabytes of data have a good reason to think about using GPUs. For all others, it’s better to wait until cloud providers make this technology affordable and simple enough to use.