In April, when NVIDIA company presented the new graphics system architecture Pascal, it was stated that this is the best at the moment decision to supersystems market. On the base of eight core GPU P100 shelf computer the company has created DGX-1 in the body 3U height. One such unit, according to developers, can replace as many as 250 dual-processor servers based on x86-compatible Intel Xeon processors. Today, NVIDIA is the statement in some way confirmed by independent experts. According to the updated list of the top 500 supercomputers in the world, NVIDIA DGX SATURNV system is recognized as the most energy-efficient supercomputer.
System DGX SATURNV company collected for their needs. This required a total of 12 engineers and 124 NVIDIA block DGX-1 (total 992 GPU P100). This brought DGX SATURNV on the 28th place in the rank of the world's supersystems. But most importantly - a 40% increase in energy efficiency. The previous record, reflected in the June list of Top-500, the ratio was 6.67 gigaflops / watt. NVIDIA DGX SATURNV system demonstrates the effectiveness of 9.46 gigaflops / watt (one DGX-1 block in the peak consumption of 3.2 kW). If we compare the consumption of the system SATURNV Camphore 2 to Intel Xeon Phi generation of Knights Landing, the efficiency of NVIDIA solutions in more than 2.3 times better than that of the system on an Intel accelerators.
As mentioned above, SATURNV system is assembled to help the company's employees. In particular, it helps to design new graphics processor companies, with designers taking a lot of routine tasks to optimize electronic circuits. SATURNV also used for the software development for the car with the autopilot on NVIDIA processors DRIVE PX 2. Another application of the system - deep machine learning and development, as well as the training of neural networks. In other words - work towards creating artificial intelligence.
At the same time it should be noted that the company is still too early to celebrate the success of the new development. According to the official statement of NVIDIA, there are very few customers in the system based on the DGX-1. For the "leadership" of development - this is extremely low. the company's main clients - the largest software developer in the face of the company SAP. SAP is using the DGX-1 develops the topic of machine learning. The second group of customers on the DGX-1 steel Open AI and two universities - Stanford and New York. They all use NVIDIA System for some advanced work. Finally, the third client was the start BenevolentAI, who is working on the AI problems. Related Products :
|