NVIDIA plans to introduce a recent artificial intelligence (AI) semiconductor on the annual developer conference 'GTC 2024' held in San Jose, California, USA from the 18th to the twenty first (local time). Regarding this, not only related industries but additionally the stock market are expressing great expectations.

Reuters reported on the 14th that NVIDIA is predicted to introduce the next-generation GPU 'B100' for AI and high-performance computing (HPC) applications at GTC next week.

After introducing B100 within the next-generation GPU release roadmap following 'H100' in October last yr, NVIDIA briefly revealed B100 performance in a document covering H200 GPU for AI and HPC early this yr. The graph shown by the corporate shows that it has achieved tremendous performance improvement in comparison with the present A100 and H100 GPUs, raising expectations.

B100 is a GPU based on 'Blackwell', the successor architecture to 'Hopper', and is predicted to start shipping at the top of this yr using the TSMC 3nm process. It is usually expected to be the primary GPU to utilize a chiplet design.

Improvements in B100 GPU performance may be expected just through the use of 3-nanometer process technology. If a chiplet design is adopted here, the variety of transistors and execution units are expected to extend dramatically although power consumption increases. In other words, it’s an unprecedented ‘high-cost, high-performance’ structure.

Moreover, because it uses faster memory than the H100 or H200, performance improvement is predicted. Although the particular price is not going to be revealed, the B100 is more likely to be dearer than its predecessor, which is sold for greater than $20,000 (about 26.6 million won).

In conclusion, it is predicted that through B100, we’ll see the biggest generational performance improvement in NVIDIA GPUs in recent times.

NVIDIA called CEO Jensen Huang's GTC keynote speech “a transformative moment in AI” and emphasized, “Don't miss CEO Jensen Huang's keynote speech.” This GTC keynote speech will reveal the capabilities of the Blackwell-based B100 computing GPU. It’s believed to be a touch that it would be demonstrated.

Actually, the A100 GPU based on the Ampere architecture was announced on the GTC keynote in May 2020, and the H100 GPU based on the Hopper architecture was unveiled on the GTC keynote in March 2022.

Meanwhile, this event is the primary face-to-face event because the pandemic, and 16,000 persons are expected to attend, double the number from five years ago. Greater than 300,000 people from world wide are expected to participate online.

This yr's event is attracting great attention in that it’s being held with NVIDIA leading the Recent York stock market's strength on account of the AI ​​craze and becoming probably the most notable corporations. The evaluation is that the expectations around and the burden on Nvidia can be enormous.

Bloomberg said, “With some evaluating that the rise in Nvidia's stock price is excessive, this event could be very necessary for the long run direction,” and added, “Some are comparing it to the Apple Developer Conference.”

As well as, Bank of America analysts called this event 'AI Woodstock' and raised Nvidia's stock price goal from $925 (about 1.2 million won) to $1,100 (about 1.5 million won) ahead of the event. Woodstock is an American rock festival held in 1969, and has established itself as a legendary event symbolizing the hippie era.

Reporter Park Chan cpark@aitimes.com