 "NVIDIA launches new Blackwell B200 AI superchip, claims 30 times more power than their current flagship H100")
NVIDIA, a frontrunner in AI hardware, has taken the tech world by surprise after it announced a completely new platform out of nowhere The latest platform, dubbed Blackwell is a family of new generation of AI chips, that promise significantly better AI computing performance, and better energy efficiency.
Revealed by CEO Jensen Huang at the company’s annual GTC event in San Jose, the new Blackwell series of AI chips promises unparalleled speed and efficiency, marking a significant leap forward in how artificial intelligence machines work
Named in homage to mathematician David Harold Blackwell, NVIDIA’s Blackwell chips boast a remarkable performance upgrade over their predecessors, the H100 chips. According to Huang, Blackwell chips are between seven and 30 times faster than the H100, while consuming a fraction of the power — about 25 times less, to be precise. Leading the pack are the new Blackwell B200 GPU and GB200 “superchip.”
According to Nvidia, the latest B200 GPU boasts a remarkable 20 petaflops of FP4 power, leveraging its 208 billion transistors. Additionally, Nvidia claims that pairing two of these GPUs with a Grace CPU in the GB200 configuration can deliver LLM inference workloads with a performance boost of up to 30 times, potentially achieving significantly greater efficiency. Nvidia asserts that this setup could reduce costs and energy consumption by as much as 25 times compared to the H100.
This breakthrough technology is poised to redefine the landscape of AI applications, considering how resource-hungry they can get.
Huang emphasized the pivotal role of Blackwell GPUs in driving what he termed the “new Industrial Revolution,” highlighting the transformative potential of generative AI. With support from leading companies across various sectors, NVIDIA aims to unleash the full capabilities of AI, revolutionizing industries and driving innovation.
The technological prowess of Blackwell chips lies in their unprecedented performance capabilities. With speeds of up to 20 petaflops, Blackwell outstrips the H100’s performance by a staggering margin—quadrupling the computational power.
This quantum leap is facilitated by the integration of 208 billion transistors, a substantial increase compared to the H100’s 80 billion. NVIDIA achieved this feat by interconnecting two expansive chip dies, enabling lightning-fast communication at speeds of up to 10 terabytes per second.
The significance of NVIDIA’s contributions to the AI landscape is underscored by endorsements from industry titans. In a testament to the indispensability of NVIDIA hardware, CEOs including Sam Altman of OpenAI, Satya Nadella of Microsoft, and Sundar Pichai of Alphabet lauded Blackwell’s performance advancements. Elon Musk, CEO of Tesla, echoed this sentiment, hailing Blackwell as a catalyst for innovation in AI computing.
While NVIDIA has yet to disclose pricing details for Blackwell chips, the demand for its technology remains insatiable. Despite the hefty price tags associated with NVIDIA’s offerings, tech companies view access to these chips as a badge of honour. Startups have even raised funds solely on the basis that they have access to NVIDIA’s H100 AI GPUs. With delivery wait times stretching up to 11 months, securing NVIDIA’s AI chips has become a top priority for firms seeking to maintain a competitive edge in the AI landscape.
As NVIDIA continues to push the boundaries of AI computing, the unveiling of Blackwell heralds a new era of innovation and progress, propelling the industry towards unprecedented heights.
(With inputs from agencies)
