This Is The Most Expensive Graphics Card In The World

The post-pandemic volatility in supply and demand, as well as the global chip shortage, have made graphics card prices skyrocket. With the growing work-from-home trend and the popularity of PC gaming, more consumers are snatching up graphics cards than before. Because of supply bottlenecks from imports and the global chip shortage, companies cannot meet the demand. Crypto miners and scalpers stockpiled graphics cards, leading to more scarcity and a steep rise in the prices of cards as well.

Modern graphics cards are also incredible powerhouses that can handle everything from intensive 4K, ray-tracing in games to parallel processing for artificial intelligence. AI training is where you find the most high-performance, blistering-fast GPUs because they're required to analyze big data and handle deep learning workloads on massive data models and sets (via Nvidia). These bleeding-edge GPUs cost more to make, so they are more expensive.

With the accelerated rise of complex AI like ChatGPT and AI-powered art, the demand for sophisticated hardware to develop said AI systems is also rising. So unsurprisingly, the most expensive graphics card on the market is designed — not for gaming or everyday work — but AI inference, training, and high-performance computing.

Meet the fastest graphics card in the world — the Nvidia A100. At least when you're talking pure benchmark figures, the A100 is as fast as a GPU can be (via Tom's Hardware). The GPU is sold as part of Nvidia's DGX 100 system designed for AI data centers. The DGX 100 is a stack of eight Nvidia A100 cards, but it's optimized to scale to thousands of GPUs to handle anything thrown at it (via Nvidia). The latest-gen A100 has a whopping 80GB of memory with a bandwidth of up to 2 terabytes per second. When paired via Nvidia's NVLink bridges, the bandwidth and computing power go up considerably (via Nvidia). 

Using MIG technology (multi-instance GPU), it can be split into seven GPU instances to run seven different workloads in parallel. 

The GPU is built on an advanced 7 nm TSMC fabrication process, so it packs more processing power while reducing power usage. Almost 7000 32-bit and 3500 64-bit CUDA cores, coupled with over 400 Tensor cores, handle all the number-crunching. 

The DGX A100 system retails for hundreds of dollars plus service charges. But even the Nvidia A100 graphics card goes for upwards of $30,000, making it the most expensive GPU on the market (via NeoBits).