In the realm of artificial intelligence, ChatGPT stands as a testament to the power of modern hardware. This AI chatbot, a product of collaboration between OpenAI and Microsoft, is powered by an impressive array of NVIDIA V100 and A100 GPU clusters. These GPUs, specifically designed for AI and analytical applications, are the backbone of the ChatGPT infrastructure.
The NVIDIA A100 GPU, the primary hardware component of ChatGPT, is not your average gaming GPU. It lacks a display out, signifying its specialized role in AI applications. This GPU comes in two versions: a PCI Express version and an SXM4 version. The latter is more prevalent in data centers due to its ability to handle a higher electrical power load. The SXM4 version of the A100 can utilize up to 500 watts, leading to superior performance.
NVIDIA V100 Tensor Core
The NVIDIA V100 Tensor Core currently stands at the apex of data center GPUs, constructed specifically to escalate the capabilities of Artificial Intelligence, High-Performance Computing (HPC), the emerging field of data science, and graphics rendering. Utilizing the cutting-edge NVIDIA Volta architecture as its powerhouse, this GPU is available in two memory configurations of 16 and 32GB.
It astounding capacity is such that it delivers the equivalent performance of up to 32 Central Processing Units (CPUs) concentrated in a single GPU. This technological marvel empowers data scientists, researchers, and engineers by relieving them from the burden of optimizing memory usage, thus enabling them to channel their effort and time into crafting groundbreaking AI innovations.
In an era where AI is deployed everywhere from recognizing speech patterns to training virtual personal assistants (VPAs) and teaching autonomous cars to drive, data scientists are confronting increasingly intricate challenges. Addressing these complications calls for the creation and training of deep learning models that are complex and vast, all within a reasonable timeframe.
With an impressive arsenal of 640 Tensor Cores, the V100 elevates to the rank of the world’s first GPU to shatter the 100 teraFLOPS (TFLOPS) barrier, signifying a milestone in deep learning performance. Its subsequent generation – the NVIDIA NVLink serves to connect multiple V100 GPUs at lightning speeds of up to 300 GB/s, spawning the world’s most formidable computing servers. AI models that had previously strained computing resources for weeks can now be trained within a matter of days. This remarkable decrease in training time unfolds a whole new panorama of AI challenges that were previously deemed unsolvable, therefore broadening the horizons of AI’s problem-solving domain.
What hardware is used to power ChatGPT
These GPUs are interconnected with a high-speed NV-link, enabling the GPUs on a single board to function as a single, large GPU. A single NVIDIA HGX A100 unit, housing eight A100 GPUs, is capable of running ChatGPT. However, due to the vast number of users, additional processing power is required to ensure seamless operation.
The exact number of GPUs used in the ChatGPT project remains undisclosed by OpenAI and Microsoft. However, estimates suggest that around 30,000 A100s are in operation. The training process of the AI model likely required around 4,000 to 5,000 GPUs, but accommodating 100 million users necessitates approximately six times more GPUs.
Microsoft’s investment in AI
Microsoft’s investment in this system is believed to be in the hundreds of millions of dollars, with daily operational costs reaching several hundred thousand dollars. Microsoft is currently integrating the newer NVIDIA H100 GPUs into its Azure Cloud AI service, which outperforms the A100’s performance by a factor of six and adds FP8 support. This upgrade will allow more people to use ChatGPT and other AI services, and will enable Microsoft to train more complex language models.
In addition to GPUs, ChatGPT also utilizes CPUs for tasks less suited for GPUs, such as loading data and running the chat interface. Storage is another crucial component, with SSDs or cloud storage typically used to store the massive datasets and models. A high-speed network, usually provided by a dedicated data center, is essential for ChatGPT to communicate with users and other systems.
The NVIDIA V100 GPU, a high-performance GPU designed for data centers, and its successor, the NVIDIA A100 GPU, are key components of the ChatGPT hardware configuration. SSDs, which are faster than traditional hard drives, and cloud storage, hosted on remote servers, are used to store the datasets and models that power ChatGPT.
The hardware used to power ChatGPT is continually evolving as new technologies emerge, allowing ChatGPT to become more powerful and efficient over time. This AI chatbot is a testament to the power of modern hardware and the potential of artificial intelligence.
Source & Image Credit: NVIDIA
Filed Under: Guides, Top News
Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, togetherbe may earn an affiliate commission. Learn about our Disclosure Policy.