According to Nvidia’s latest blog analysis, Blackwell GPUs cost nearly double per hour compared to Hopper generation, yet deliver 35-fold lower per-token inference costs. Using DeepSeek-R1 as a test model, Blackwell (GB300 NVL72) rents at $2.65 per GPU per hour versus Hopper’s $1.41, but single GPU throughput jumps from 90 to 6,000 tokens per second. This 65x throughput gain reduces per-million-token costs from $4.20 to $0.12.
The $0.12 figure assumes full software optimization including FP4 low-precision inference and multi-token prediction (MTP). Without MTP enabled, per-million-token costs reach approximately $2.35, dropping to $0.11 with it active, demonstrating a 21x optimization impact from that feature alone.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
OpenAI Announces GPT-5.5-Cyber Cybersecurity Model Launch in Coming Days
According to Cointelegraph, Sam Altman announced OpenAI will launch GPT-5.5-Cyber, a cybersecurity model, in the coming days. The model will be made available to critical cybersecurity defenders, with OpenAI planning to establish trusted access mechanisms in collaboration with government entities to
GateNews3m ago
Google DeepMind CEO Predicts AGI Arrives by 2030, Giving Companies Four Years to Prepare
According to Sequoia Capital's video series featuring Demis Hassabis, Google DeepMind's CEO predicts artificial general intelligence (AGI) will arrive by 2030. Hassabis, a 2024 Nobel Prize winner in chemistry, views this timeline as a strict business reality that forces leaders to rethink product de
GateNews7m ago
OpenAI to Operate as Perpetually Low-Profit Company, Signs 20-Year Power Deal
According to Odaily, OpenAI CEO Sam Altman stated at Stripe Sessions that OpenAI aims to become a perpetually low-profit yet massive and fast-growing infrastructure company, offering products similar to an "intelligent meter" that users can purchase to automate business operations or embed into
GateNews8m ago
Forefront Tech Completes $100M IPO Pricing, to List on Nasdaq
According to ChainCatcher, Forefront Tech Holdings Acquisition Corp completed a $100 million IPO pricing on April 30, set to list on Nasdaq under ticker FTHAU. The proceeds will be used to explore merger and acquisition opportunities in blockchain, artificial intelligence, fintech, and other
GateNews45m ago
Ex-Meta Executive Raises Funds for AI Startup Flourish at $2.5B Valuation
According to Bloomberg, Thomas Reardon, former Meta executive who led work on Meta's Neural Band, is raising funds for Flourish, a new AI startup focused on low-power AI systems. Lux Capital and Google Ventures have agreed to be major investors, with backers already committing $500 million at a $2.5
GateNews47m ago
OpenAI Traces Goblin Problem to Nerdy Personality Reward Signal, Goblin Mentions Spike 175% in GPT-5.1
According to OpenAI, the company identified the root cause of the "goblin" problem that plagued GPT models from GPT-5.1 onwards. A reward signal used to reinforce the "Nerdy" personality trait encouraged outputs containing fantasy creature references, with 76.2% of the training dataset showing
GateNews1h ago
