In recent years, Nvidia has achieved significant success by pivoting to artificial intelligence (AI), with large language models and GPU-accelerated “premium AI PC” experiences becoming the hot trend in 2024. However, newer and smaller companies are now vying for market share, and they are not the usual suspects.

As reported by The Economist, there are noteworthy developments in the GPU field beyond Nvidia and AMD, driven by the demands of AI computing. Today’s large language models often run on setups featuring interconnected GPUs and memory, exemplified by Cerebras Systems’ innovative hardware.

Cerebras Systems Inc., founded just nine years ago, has thrived amid the AI computing boom. Its groundbreaking innovations appear to outshine Nvidia’s current-gen H100 and the upcoming GB200 die with a “single, enormous chip” boasting up to 900,000 GPU cores, such as the CS-3 chip. The Cerebras CS-3 chip, which dwarfs the GB200 with its immense size and capabilities, has been described by the manufacturer as the “world’s fastest and most scalable AI accelerator.” It is designed to “train the world’s most advanced AI models” and features a staggering 10 trillion transistors, a 4,707% increase compared to the GB200’s 208 billion.

Cerebras claims that its Wafer Scale Engine, the foundation of the CS-3, is “the chip that broke Moore’s Law,” confidently outperforming the H100 in internal benchmarks.

However, Cerebras isn’t the only newcomer making waves. Groq, a startup, is also developing hardware for AI computing. Instead of going larger, Groq has created dedicated LPUs (language processing units) designed to run large language models efficiently. The Groq LPU Inference Engine is an “end-to-end inference system acceleration system,” delivering significant performance, efficiency, and precision. It currently runs the Llama-2 70B, a large-scale generative language model, at 300 tokens per second per user. This capability is due to the LPU, which works alongside CPUs and GPUs in data centers, enabling low latency and real-time delivery.

Nvidia’s financial success in recent years has been no secret. The company was even briefly more valuable than Amazon and challenged Alphabet (Google’s parent company) in market value. Such profitability in the AI market has naturally attracted more manufacturers to enter the competition, aiming to disrupt Nvidia’s dominance.

Whether Cerebras, Groq, or smaller companies like MatX can secure a significant foothold remains to be seen. AI computing is still in its early stages, and now is the time for experimentation. Some companies will scale up their operations, while others will focus on smarter, more specialized solutions. The competition is fierce, and the race to innovate and capture market share is just beginning.

By Impact Lab