Nvidia‘s latest chip could help transform the commerce industry by allowing AI applications to run faster and more efficiently, experts say.
of B200 GPUwas announced Monday (March 18) during a keynote by the CEO at Nvidia’s annual developer conference. jensen fana computer chip that can deliver up to 20 pieces of data. petaflops Thanks to 208 billion transistors, the processing power of FP4 is significantly increased.
Additionally, Nvidia said that the GB200, which integrates two B200 GPUs and one Grace CPU, can deliver up to 30x performance improvements for large-scale language model (LLM) inference tasks. GB200 is his Nvidia debut chip blackwell AI graphics processor series. This setup is designed to be significantly more efficient, potentially reducing costs and energy consumption by up to 25 times compared to his previous H100 model.
“The B200’s ability to analyze vast amounts of data allows companies to more accurately predict customer demand,” said Lars Nijman, the company’s chief marketing officer. CUDO computingsaid in an interview with PYMNTS. “This improves inventory management and reduces the risk of out-of-stocks and overstocks.”
Faster processing power
According to Nvidia, training a 1.8 trillion parameter model previously required 8,000 parameters Hopper GPU and 15 megawatts of power. However, with the new Blackwell architecture he only needs 2,000 GPUs and consumes only 4 megawatts of power. On the GPT-3 LLM benchmark with 175 billion parameters, the GB200 demonstrated a 7x performance improvement over his H100 and his 4x improvement in training speed.
benjamin leeA University of Pennsylvania engineering professor said in an interview with PYMNTS that the B200 will change AI through improved power efficiency. He pointed out that training AI models is expensive because the team has to pay for his GPU and also pay for the electricity to run it.
“Improving power efficiency directly translates into lower operating costs,” he said.
The new chip can perform twice as many calculations per second as the previous generation by cutting the accuracy of calculations in half.
“Researchers have long studied how changes in accuracy affect efficiency and performance, and the B200 is a pretty significant demonstration of this idea,” said Lee. .
According to Lee, the B200 builds the largest chips possible and uses a high-speed network to interconnect pairs of those chips. This network facilitates more efficient coordination of computations between the two chips. The efficiency of the GPU in receiving data for computation is highly dependent on the performance of this interconnect network.
Lee said Nvidia’s main advantage over competitors persists in its software ecosystem designed to run AI workloads on GPU hardware. Compilers in this ecosystem enable researchers to quickly deploy models to the latest generation GPUs. This innovation allows you to take advantage of increased energy efficiency.
“Another key advantage for NVIDIA continues to be its high-performance networking that allows GPUs in large clusters to communicate with each other quickly and efficiently,” he added. “Combining these advantages enables efficient AI for models with up to trillions of parameters.”
Impact of new chips on commerce
The new chip’s capabilities may translate into real-world results.
Nyman said the B200 can provide a more personalized, efficient and secure shopping experience for both businesses and consumers. He noted that the processor’s real-time price optimization capabilities enable dynamic price adjustments based on factors such as demand, competition and customer behavior, allowing businesses to maximize profits while remaining competitive. .
The B200 could also be important in enhancing security and fraud detection, Nyman said. By analyzing transactions in real-time, processors can identify suspicious patterns that may indicate fraudulent activity, helping businesses prevent financial loss and protect customers from fraud.
Nyman emphasized that the B200’s high processing speed opens up new possibilities for personalization and customer profiling. The processor also enables real-time analysis of customer browsing behavior, giving merchants a window to engage with shoppers.
“B200’s capabilities allow us to create highly targeted advertising assets that deliver the right message to the right customer at the right time, adjusting based on browsing behavior,” said Nyman. Masu.
Nyman said he also envisions the B200 powering virtual shopping companions. The processor enables the creation of a virtual assistant that accompanies customers through their online store, helping them select outfits, compare products, and answer questions, providing a more personalized and engaging shopping experience. Become.
Thanks to new chips, AI products could become cheaper and more widely available in the future.Founder Abdullah Ahmed Calm data operationtold PYMNTS that this is due to something called “cheap reasoning.”
Ahmed said that starting in 2024, more companies could start using chips like the B200 and basic AI models that are sufficient for their needs. Then you can focus on improving and selling your product.
This means that even small businesses may be able to turn to AI products without the need for very expensive GPUs. This could reduce the cost of developing and using AI tools, making them more affordable for businesses and consumers.
“However, this is dependent on Nvidia’s ability to maintain lead times reasonably in the face of overwhelming demand,” he said. “B200 is focused on the prevalence of large-scale language models, such as his GPT4 in OpenAI, which powers ChatGPT.”
