At the GTC conference on March 18th, Nvidia CEO Jensen Huang announced a groundbreaking plan to boost the processing speed of DeepSeek's AI by 30 times with its newly released software, Dynamo. This move is Nvidia's response to the market disruption caused by DeepSeek's R1 AI program. DeepSeek's technology significantly reduces the computational requirements for each query, leading to investor concerns about Nvidia's future prospects.
Nvidia's Dynamo software can distribute AI inference tasks across up to 1000 GPUs for parallel processing, dramatically increasing query throughput. Huang stated that using the new Blackwell chip, Dynamo delivers 30 times the processing power on the same architecture. This increased capacity allows AI data centers to process more tokens per second, leading to higher overall revenue.
Specifically, Dynamo software can increase processing throughput at a cost of $1 per million tokens. This means service providers can choose to handle more client queries simultaneously or provide more powerful services to individual users, charging higher fees. Nvidia calls this high-performance AI service an "AI factory."
To further enhance DeepSeek's performance, Nvidia also released a tuned version of DeepSeek R1 on HuggingFace. This version uses fewer computational bits, namely "FP4," significantly improving performance without substantially reducing model accuracy.
Besides Dynamo, Nvidia also unveiled the latest version of the Blackwell chip, "Ultra," boasting increased memory capacity and overall performance. Furthermore, Nvidia officially launched the DGX Spark, a smaller personal computer for AI developers, and updated the configuration of the DGX Station desktop computer.
At the conference, Nvidia also showcased numerous other new products and collaborations, including improved network switches and software models for medical robots, demonstrating Nvidia's continued innovation in AI and computing.
Key Highlights:
🌟 Nvidia launched Dynamo software, boosting DeepSeek AI processing speed by 30 times.
💰 Dynamo enables service providers to process client queries more efficiently, increasing revenue.
🖥️ The new Blackwell "Ultra" chip and DGX Spark computer were also officially released at the conference.