NVIDIA's new Blackwell platform has earned top marks in the latest InferenceMAX v1 benchmarks, proving itself as a leader in both performance and energy efficiency for artificial intelligence (AI) systems.
So what makes this newsworthy? InferenceMAX v1 is an independent benchmark test that evaluates how much it actually costs to run different AI models in real-world situations. NVIDIA Blackwell took the lead in every important category, showing that it delivers more power and does so with less energy and lower costs than other options.
NVIDIA’s GB200 NVL72 system is especially impressive. A company that spends 5 million dollars on this system could earn 75 million dollars in revenue by running AI models, which means getting 15 times more back than what they invested. Even more, the Blackwell B200 chips now power through tasks much faster and more efficiently, driving down costs to just two cents for one million AI tokens, which is five times cheaper than before.
Performance is also about speed. The NVIDIA B200 GPU can process as many as 60,000 AI tokens every second on certain tasks. For users, that could mean getting 1,000 tokens each second, making chatbots and AI assistants more responsive than ever. These advancements come thanks to NVIDIA’s continuous software updates and partnerships with the open-source community to keep everything running smoothly and quickly.
But why do these results matter for beginners or businesses? Modern AI is used for much more than just answering questions. It now does complex reasoning and handles more steps in each task, asking a lot more from computer systems. Thanks to Blackwell’s improvements, companies can deploy AI faster, at larger scale, and get more value for their money.
Behind these achievements are cutting-edge features. The Blackwell architecture uses a new format to store numbers efficiently without sacrificing accuracy. Multiple GPUs can be linked together to work as a giant, powerful brain. Software like NVIDIA TensorRT LLM further boosts speed by predicting more information at once, meaning less waiting for users.
Another important detail is energy efficiency. For companies running large AI factories, saving energy means saving money and doing more with the same amount of power. Blackwell delivers ten times more results per unit of energy compared to previous systems, making it much more sustainable.
Overall, NVIDIA Blackwell proves that with smart hardware and software working together, AI can be both fast and cost-effective. This paves the way for more businesses to use AI in real-time operations, turning data into valuable insights and decisions. Open benchmarks like InferenceMAX help everyone see which systems truly deliver, guiding smarter investments in the rapidly growing world of artificial intelligence.
Original article and image: https://blogs.nvidia.com/blog/blackwell-inferencemax-benchmark-results/
