NVIDIA has dropped some major hints that the much anticipated Hopper GPU will be discussed at GTC 2022. The business published a blog post on their website titled “Hopped Up: NVIDIA CEO, AI Leaders to Discuss Next Wave of AI at GTC,” and Videocardz quickly observed that this appears to relate to the next-generation Hopper architecture.
Keep in mind that we’re dealing with Jensen here, and this isn’t the first time the charming CEO has tinkered with the press and sparked rumours.
So even a smidgeon of scepticism wouldn’t go amiss here. NVIDIA’s Hopper GPU was planned to be the company’s first MCM-based approach, with as many as two GPUs on a single device.
Given that we are currently limited by the reticle size of most EUV scanners, an MCM-based design is arguably the next stage in GPU advancement. The next logical frontier is architectural enhancements and MCM-design, and because AMD has already implemented it on the CPU front, it stands to reason that GPUs would be the next step in their grand plan – which would explain why NVIDIA would want to gain a head start and beat them to the punch.
The following are the rumoured specifications for the future NVIDIA H100 GPU (hopper flagship). These specifications are unconfirmed, and you read them at your peril. Based on the TSMC N5 technology, it is thought to have two 900mm2 dies. This amounts to 140 billion transistors in total or 70 billion per GPU. It represents an increase in density of 18% per mm squared. It will contain 268 GPU clusters and 34,304 CUDA cores in total. The memory bus will be 6144 bits, and there could be 128GB of HBM3 onboard. The TDP will very certainly be in the 500W+ range.
The architectures of NVIDIA have always been founded on computer pioneers, and this one appears to be no exception. Grace Hopper, one of the pioneers of computer science and one of the first programmers of Harvard Mark 1 and designer of the first linkers, inspired Nvidia’s Hopper architecture. She also popularized the concept of machine-independent programming languages, which led to the creation of COBOL, a pioneering high-level programming language that is still in use today. During World War II, she joined the Navy and contributed to the American war effort.