GPU manufacturer Nvidia has been at the forefront of the generative AI wave and is also one of its biggest beneficiaries. The company, which produces the graphics processing units that power models like ChatGPT, showed a mind-boggling forecast of $11 billion for the next quarter, a whopping 50 per cent higher than Wall Street’s estimates of $7.15 billion.
The company also reported a revenue of $7.19 billion for the first quarter of fiscal 2024, against analysts’ prediction of $6.52 billion. Earnings per share at $1.09 surpassed consensus expectations of $0.92. NVIDIA’s founder, in a blog, announced that the firm is boosting production of its AI chips to meet the surging demand.
As an impact, AI stocks rose after market trading hours adding almost $300 billion in market capitalisation. Taking another step towards becoming the 7th company in the world to reach a trillion-dollar market cap, NVIDIA and SoftBank have partnered to drive generative AI applications in Japan. SoftBank will use NVIDIA’s GH200 Grace Hopper superchip in its data centers to develop 5G/6G applications, enabling cost-effective and energy-efficient generative AI and wireless applications. The collaboration aims to advance areas such as autonomous driving, AI, AR, VR, and digital twins.
Ruling AI With an Iron Fist
NVIDIA has come a long way to capture the chip market like no other. It is the original GPU maker which introduced the first GPU for personal computers, the GeForce 256, in 1999. After Rajat Raina, Anand Madhavan, and Andrew Y. Ng published a research paper in 2009, the idea that training AI algorithms is ideal for Gpus took shape—because the majority of training tasks, like matrix multiplication, are well-suited for parallel processing.
NVIDIA capitalised on this and in order to reinforce its dominant position in the market, the company introduced the DGX series of AI supercomputers. These supercomputers are specifically designed for AI applications and utilise the advanced Volta GPU architecture. With the addition of the Titan V and Quadra GV100, this series expands the availability of AI computing power to both businesses and individual consumers.
Additionally, the CUDA API provided by NVIDIA and the compatibility of their GPUs with AI workloads established these chips as indispensable tools for training AI models.
NVIDIA’s dominance in the AI market is evident, as demonstrated by its extensive use in training models like ChatGPT and Stable Diffusion. OpenAI utilised over 10,000 NVIDIA H100 and A100 GPUs for ChatGPT, while Stable Diffusion required around 200,000 GPU hours with the NVIDIA A100 GPU. Additionally, major cloud providers such as AWS and Azure have partnered with NVIDIA to create large-scale clusters of their GPUs specifically for enterprise training purposes. The company has solidified its position in AI training with the introduction of their latest H100 series GPUs.
But competitors like AMD have joined the AI accelerator market with their latest AMD Instinct AI accelerator, which combines CPU and GPU technologies for enterprise use. However, AMD’s consumer GPUs are not as well-suited for AI applications as NVIDIA’s GPUs because they lack an equivalent API to NVIDIA’s CUDA. While AMD has created the ROCm open software platform for machine learning, it is not as mature or widely integrated as CUDA, which has undergone extensive development over the years. NVIDIA’s GPUs have better integration with popular AI tools like TensorFlow and PyTorch, resulting in fewer issues and bugs in practical work settings.
Application Before the AI Wave
To reduce dependence on NVIDIA, tech giants like Google and Amazon have developed custom chips for AI workloads, such as AWS’s Inferentia for inference tasks and Google’s Tensor Processing Unit for TensorFlow.
But NVIDIA is a seasoned veteran present even before GPUs application in training AI algorithms was a thing, Tesla was using NVIDIA’s DRIVE PX 2 AI computing platform as an in-vehicle supercomputer for their self-driving cars. All Tesla Motors vehicles manufactured from mid-October 2016 were using DRIVE PX 2. This powerful system, providing over 40 times more processing power than its predecessor, runs Tesla’s neural network for processing vision, sonar, and radar data. NVIDIA has a longstanding collaboration with Tesla, with their CEO, Jensen Huang, being both a Tesla owner and a supplier of Tegra chips.
On the other hand, Google introduced A3 GPU VMs, powered by NVIDIA H100 Tensor Core GPUs to accelerate the training and inference of complex machine learning models. These VMs utilise Google’s advanced networking technology, including 200 Gbps IPUs for high network bandwidth and low latency.
Even as tech giants enter the AI compute market, these chips do not hold a candle to NVIDIA’s general purpose GPUs. While these chips may surpass its GPUs in specific workloads, NVIDIA’s GPUs are more versatile and suitable for a broader range of applications.
The post The Rise And Rise of NVIDIA appeared first on Analytics India Magazine.
