NVIDIA's HGX-2 will transform trillion-dollar industries through a new cloud-server architecture
NVIDIA’s HGX-2 will transform trillion-dollar industries through a new cloud-server architecture
If 2017 was the coming of the AI revolution, 2018 is surely the era of delivering AI compute to propel the AI revolution. It’s no wonder NVIDIA can deliver up to a 10x performance increase in certain training workloads in a span of just six months when comparing the DGX-1V and the more recent DGX-2 supercomputer.
But if NVIDIA is going to bring this power of the DGX-2 to scale in a big way, it must come up with a cloud-server architecture that other OEM and ODM server systems can adapt and bring this technology to more diverse applications. The continuing explosive growth of AI model size and complexity means the appetite for more powerful compute solutions continues to accelerate rapidly.
Enter the second generation of the NVIDIA HGX reference cloud server platform architecture, HGX-2, which fuses both high-performance computing and AI computing needs in one platform. Like the DGX-2 supercomputer, the HGX-2 comes with multi-precision computing capabilities, allowing high-precision calculations using FP64 and FP32 for scientific computing and simulations, while also enabling FP16 and INT8 for AI training and inference.
The HGX-2 shares much of the key components of the DGX-2, such as the twin GPU baseboards that are a key building block to the server platform. Each baseboard takes in eight Tesla V100 NVLink GPU modules and six of those amazing NVSwitches for a total of 16 Tesla V100 GPUs that are fully connected and can connect with any other GPU simultaneously at the full NVLink speed of 300GB/s.
Ultimately, the HGX-2 is NVIDIA’s solution to deliver an optimized GPU sub-system while allowing their server partners to focus on the complete server design that relates to mechanical, power, cooling, topology and connectivity needs of their customers’ specific cloud data center requirements. Working together, NVIDIA hopes to bring their AI compute platform sooner to the market and a variety partners will bring the HGX-2 server platform to the cloud later this year.
For more technical details of the HGX-2 platform, NVIDIA has a dedicated blog.
Like the DGX-2, HGX-2 will enable developers with this higher performance platform to build larger and more complex AI and HPC models while training them in less time. These servers will also supply these trillion-dollar industries with compute power to revolutionize their verticals:-
- NVDIA’s new RTX technology for accelerated ray tracing – the standard for cinema quality visuals – coupled with real-time graphics technology and AI will accelerate traditional CPU-driven render farms and offer creative professionals visualizing options like never before to fast track production workflows.
- NVIDIA’s Project Clara can revolutionize the healthcare industry through redefining medical imaging, providing more fidelity out of current medical instruments or generate images with lower power that will make it safe for use with younger children.
- Autonomous driving through NVIDIA’s Drive platform will change our world with safer travel, increased consumptions and lowering costs at the same time thanks to NVIDIA’s ability to collect, process data, train models and simulate billions of driving miles.
- Safer and better city management through NVIDIA’s smart city vision and NVIDIA Metropolis city-scale video analytics platform.
Source: NVIDIA