More environmentally friendly and energy-saving: Next-generation systems powered by NVIDIA point to the future development direction of data centers

More environmentally friendly and energy-saving: Next-generation systems powered by NVIDIA point to the future development direction of data centers

Although Moore's Law has ended, the need to improve computing performance never ends. For this reason, the cost and power consumption of traditional methods will also increase exponentially.

At the same time, to mitigate the effects of climate change, we need more efficient data centers. The annual energy consumption of global data centers has exceeded 200 megawatt hours, accounting for 2% of the world's total energy consumption.

The latest Green500 list shows the energy efficiency of accelerated computing. Those on the list are all the most efficient supercomputers in the world, and the top 30 systems all use accelerated computing. The impact of accelerated computing on energy efficiency is staggering.

NVIDIA estimates that the systems on the TOP500 list consume more than 5 terawatt hours or $750 million worth of energy annually.

But if the systems on the TOP500 list were as efficient as the top 30 systems on the Green500 list, then this number could be reduced by more than 80%, which would consume only $150 million in energy and save a full 4 terawatts. hour.

On the contrary, with the power budget of today's TOP500 systems, if these supercomputers can reach the efficiency of the top 30 systems on the Green500 list, they can provide 5 times the performance of today.

The efficiency improvements highlighted by the latest Green500 system are just the tip of the iceberg. NVIDIA is continuously improving the energy consumption of its portfolio of CPUs, GPUs, software and systems.

Hopper's debut on the Green500 list

Of the top 30 systems on the latest Green500 list, 23 use NVIDIA technology.

Most notably, New York City's Flatiron Institute's air-cooled ThinkSystem topped the Green500 list, manufactured by Lenovo and powered by the NVIDIA Hopper H100 GPU.

This supercomputer is also known as Henri. According to Green500 data, Henri can perform 65 billion double-precision floating-point operations per watt and will be used to solve problems in the fields of computational astrophysics, biology, mathematics, neuroscience and quantum physics.

Among the new systems on the TOP500 list, the fastest new computer is the Leonardo managed by the non-profit consortium Cineca. The system uses nearly 14,000 NVIDIA A100 GPUs and ranks 4th on the TOP500 list and 13th on the Green500 list.

A total of 361 systems on the TOP500 list use NVIDIA technology, and 90% of new systems use NVIDIA technology (see chart).

The number of systems using NVIDIA technology on the latest TOP500 list has reached a record high.

Next-generation accelerated data center

NVIDIA is also developing new computing architectures that can provide higher energy efficiency and performance for accelerated data centers.

The Grace CPU and Grace Hopper super chips released earlier this year will further significantly improve the energy efficiency of NVIDIA's accelerated computing platform. With the amazing efficiency of Grace CPU and low-power LPDDR5X memory, the performance per watt of the Grace CPU super chip can be up to twice that of traditional CPUs.

Assuming that a 1 million watt high-performance computing data center allocates 20% of the power to the CPU partition and 80% of the power to the acceleration partition using Grace and Grace Hopper, it is comparable to an x86-based data center using a similar partitioning method. Compared to 2018, the data center can complete 1.8 times the work on the same power budget.

DPU further improves efficiency

In addition to Grace and Grace Hopper, NVIDIA network technology has also greatly accelerated the speed of cloud-native supercomputing. At the same time, increased usage of simulation is also driving demand for supercomputing services.

The NVIDIA Quantum-2 InfiniBand platform, which includes NVIDIA BlueField-3 DPUs, delivers the extreme performance, broad availability and robust security required by cloud computing service providers and supercomputing centers.

A recent white paper describes this, showing how DPUs can be used to offload and accelerate network, security, storage or other infrastructure functions and control plane applications, thereby reducing server power consumption by 30%.

The power saved is proportional to the server load. A large data center with 10,000 servers can easily save $5 million in electricity bills over the three-year life of the servers, plus savings on cooling, power transmission, rack space and server investment.

Accelerated computing combined with DPU's capabilities in network, security and storage has become an important step in improving data center energy efficiency in the future.

Get twice the result with half the effort

The methods of scientific research are changing rapidly. Driven by methods such as data analysis, artificial intelligence and physics simulation, more efficient computers will be the key to the next generation of scientific breakthroughs.

NVIDIA is committed to providing researchers with a multidisciplinary high-performance computing platform optimized for this new scientific research method, while taking into account performance and efficiency, allowing scientists to carry out major research that benefits all mankind.