AI, Robots, New Chips, and New Servers
Another exciting Oracle OpenWorld is now behind us and we want to thank everyone who stopped by to say hello and discuss our technology solutions. Fujitsu generated a lot of buzz with our demonstrations, presentations, and show presence. Tens of thousands of attendees took advantage of more than 2,500 educational sessions and engaged with the more than 400 exhibitors.
In his talk, Goro covered the latest advances in artificial intelligence, deep learning, and machine learning. This talk was very well attended, with nearly every seat in the room occupied and an audience outside watching on screens. You can watch Goro’s Executive Solution Session here: https://www.oracle.com/openworld/on-demand.html#executivesolutionsessions
During the talk, Goro discussed an AI processor, the DLU (Deep Learning Unit), that will deliver 10x the performance per watt vs. competitors by 2018. The processor has an architecture that is designed from the ground up for Deep Learning, with the ability to handle large-scale neural networks.
The DLU will also be able to handle varying precision, including FP32, FP16, INT16, and INT8, which will allow it to provide very high performance on deep learning jobs that don’t require high precision. This design will result in a chip that doesn’t need nearly as much power as current 64bit processors.
Each DLU includes 16 deep learning processing elements (DPE) which contains eight single-instruction, multiple data (SIMD) execution units that are connected via Fujitsu’s Tofu interconnect. The DLU will be introduced in 2018.
Visitors to the Fujitsu booth had the opportunity to get up close and personal with a DLU driven robot, Unibo. Unibo was a real crowd pleaser, showing off its ability to understand and respond in natural language.
Unibo’s ability to process natural language was enabled by Fujitsu’s Human Centric Al Zinrai software, along with a camera to recognize emotions and human faces. Its cloud-driven intelligence stores past conversations, allowing it to learn even better conversation skills as it gets more experience with humans.
Fujitsu Limited and 1QB Information Technologies Inc. also showed how they’ve been collaborating on a new quantum-inspired technology to the field of artificial intelligence (AI), focusing on the areas of combinatorial optimization and machine learning.
This collaboration will enable software developed by 1QBit for quantum computers to run on a “digital annealer,” jointly developed by Fujitsu Laboratories Ltd. and the University of Toronto. A digital annealer is a computing architecture that can rapidly solve combinatorial optimization problems using existing semiconductor technology.
One of the stars of the Fujitsu Oracle OpenWorld booth was the Fujitsu SPARC M12 server. This system can scale from two to 32 processors and can host up to 32 TB of RAM.
Fujitsu engineers put together a side-by-side performance comparison featuring the Fujitsu M10 and the new Fujitsu SPARC M12.
The photo on the left is a core-to-core comparison, showing that eight cores of the new SPARC64 XII running at 4.25 GHz more than doubles the performance of the Fujitsu M10 running the SPARC64 X+ processor.
The picture on the right shows that four cores of SPARC64 XII offer significantly more performance, almost 32% more, than eight cores of the Fujitsu M10 SPARC64 +. In fact, the SPARC64 XII offers the most per-core performance in the industry, as measured by SPEC_int and SPEC_fp benchmarks.
We again thank all of the visitors who took time to stop by our booth, we really enjoyed meeting you and hearing about your IT challenges. We will see you all next year at Oracle OpenWorld.