Last Updated on
Generative AI takes a tonne of compute power, and no one is benefitting more from that right now than NVIDIA. The GPU manufacturer has just unveiled its latest high-speed chip, the GH200. Designed with generative AI workloads in mind, the NVIDIA GH200 ‘Grace Hopper’ AI superchip will power the artificial intelligence servers of tomorrow – So what’s so special about it? NVIDIA CEO Jensen Huang explains.
What is the NVIDIA GH200 superchip?
Processor demand has skyrocketed with the advent of accessible AI, and NVIDIA are the technological front line. While thousands of entrepreneurial early-adopters are building software businesses from LLMs (Large Language Models) like Bing Chat, Bard, and of course OpenAI’s ChatGPT, someone has to design and manufacture the hardware it all runs on.
Essential AI Tools
Winston AI detector
Originality AI detector
“To meet surging demand for generative AI, data centers require accelerated computing platforms with specialized needs,” explains founder and CEO of NVIDIA, Jensen Huang. “The new GH200 Grace Hopper Superchip platform delivers this with exceptional memory technology and bandwidth to improve throughput, the ability to connect GPUs to aggregate performance without compromise, and a server design that can be easily deployed across the entire data center.”
Advancements of the NVIDIA GH200 Grace Hopper platform
To quote the press release, “The new platform uses the Grace Hopper Superchip, which can be connected with additional Superchips by NVIDIA NVLink™, allowing them to work together to deploy the giant models used for generative AI. This high-speed, coherent technology gives the GPU full access to the CPU memory, providing a combined 1.2TB of fast memory when in dual configuration.” Additional advances in AI infrastructure include “HBM3e memory, which is 50% faster than current HBM3, delivers a total of 10TB/sec of combined bandwidth, allowing the new platform to run models 3.5x larger than the previous version, while improving performance with 3x faster memory bandwidth.”
The Grace Hopper superchip is the worlds first HBM3e (High Bandwidth Memory 3 extended) is capable of 5TB/s speeds – for those unfamiliar with data speeds, this is hilariously fast. This is ‘beating a Bugatti Veyron in a drag race by over 100mph’ fast.
The superior 3.5x memory capacity and 3x bandwidth of comparable GPU’s puts the system manufacturer ahead in high performance computing. TDP is of course a significant concern considering the power consumption at the frontier of AI. Thankfully, the arm-based “NVIDIA Grace GPU“ provides high-performance with impressive efficiency – eight petaflops of AI performance in a single server.
In addition to this raw power, the modularity of the system allows many AI superchips to work together for compute-intensive tasks.
Built on existing existing infrastructure, NVIDIA NVLink switch system and NVIDIA MGX, the firm is extremely well positioned to bear the majority of the market share in AI processing hardware.
Who was Grace Hopper?
Grace Hopper, from whom the processor takes its honorary namesake, was an extremely influential American computer programmer. Not many computer scientists rise to her level of notoriety, so what is Grace Hopper best known for?
Among several other inventions, Hopper pioneered machine-independent programming languages with FLOW-MATIC – an early iteration of COBOL. The latter is still used in some industries today.
In addition to her pioneering programming, she was rear admiral of the United States Navy in the 1980’s. As a result of her involvement in the military, to which she was initially denied acceptance, Hopper helped develop one of the worlds first general-purpose computers for use in World War II – the Harvard Mark I. An extremely impressive career, to say the least.