GTC 2016: NVIDIA DGX-1, World's First Deep Learning Supercomputer (part 7)

TL;DR

Nvidia has created the Nvidia DJX1, the world's first deep learning supercomputer, with a 12x speed up year-over-year, enabling researchers and data scientists to train deep neural networks in a matter of hours instead of days or weeks.

Transcript

our strategy is to accelerate deep learning everywhere for every platform for every market for every single industry first accelerating the cloud the cloud is the fastest the best way the early adopters of deep learning and bring this technology to users all over the world however there are many companies that need to have resident on premise Compu... Read More

Key Insights

• 😶‍🌫️ Nvidia's strategy is to accelerate deep learning for every market and platform, with a focus on the cloud and on-premise computing.
• 🐎 The Nvidia DJX1 is the world's first deep learning supercomputer, offering a 12x speed up year-over-year.
• 🎨 The DJX1 has 170 Teraflops performance, eight Tesla P100 GPUs, and a dense computing node design, making it the most advanced computer for deep learning.
• ✋ Traditional high-performance computing architectures struggle to scale out for deep learning, while the Nvidia DJX1 can outperform over 250 traditional servers.

Questions & Answers

Q: What is Nvidia's strategy for deep learning?

Nvidia's strategy is to accelerate deep learning for all platforms and markets, focusing on bringing the technology to users worldwide through the cloud and OEM partners for on-premise computing.

Q: What is the Nvidia DJX1 and what makes it special?

The Nvidia DJX1 is the world's first deep learning supercomputer, engineered specifically for deep learning tasks. It has a performance of 170 Teraflops, with eight Tesla P100 GPUs, making it the most advanced computer ever built for deep learning.

Q: How does the Nvidia DJX1 compare to a dual Xeon processor?

The performance of a dual Xeon processor and the Nvidia DJX1 combined is around three Teraflops, with limited bandwidth. In contrast, the Nvidia DJX1 offers 170 Teraflops and 768 GB per second of aggregate bandwidth, making it vastly superior for deep learning tasks.

Q: Why is the Nvidia DJX1 important for researchers and data scientists?

Researchers and data scientists who work with deep neural networks require fast training times to make progress in their work. The Nvidia DJX1 provides the fastest supercomputer available for deep learning, enabling them to train networks in just two hours, compared to weeks or months.

Summary & Key Takeaways

• Nvidia's strategy is to accelerate deep learning for every platform and market.

• The cloud is the fastest way to adopt deep learning, but some companies require on-premise computing, which Nvidia provides through OEM partners.

• To meet the needs of users who require the fastest and most advanced deep learning technology, Nvidia has built the Nvidia DJX1, a deep learning supercomputer with 170 Teraflops performance and eight GPUs.