Yazılar

Nvidia’s New AI Chips Slash Training Times for Massive AI Models

/in /tarafından

Nvidia’s latest generation of AI chips is making significant advances in training some of the world’s largest artificial intelligence systems, according to new benchmark data released on Wednesday by MLCommons, a nonprofit organization that tracks AI system performance.

The results show a dramatic drop in the number of chips required to train large language models (LLMs), highlighting Nvidia’s growing technological lead in this critical area of AI development. While much of the financial market’s current focus is on the booming sector of AI inference—where AI models answer user queries—training remains a core competitive battleground, especially for developing next-generation models with trillions of parameters.

Blackwell Chips Outperform Previous Generations

Nvidia’s new Blackwell chips demonstrated superior performance over its previous Hopper generation. In tests involving Meta Platforms’ open-source Llama 3.1 405B model, which is complex enough to simulate some of the most demanding AI training workloads, Nvidia’s Blackwell chips completed training tasks with more than double the speed per chip compared to Hopper.

In one benchmark, a system using 2,496 Blackwell chips completed the training run in just 27 minutes. By comparison, even though more than three times as many Hopper chips were used in previous tests, they only achieved faster results due to sheer scale rather than efficiency.

Nvidia and its partners were the only ones to submit data for models of this size, giving Nvidia a clear demonstration of its leadership in training capabilities for multi-trillion parameter models.

Changing Industry Trends in AI Training

Chetan Kapoor, chief product officer of CoreWeave, which collaborated with Nvidia on the results, noted that AI companies are moving away from building vast, homogenous data centers with 100,000 or more identical chips. Instead, they are increasingly assembling smaller, specialized subsystems that handle different aspects of the training process. This modular approach allows companies to speed up training times and manage extremely large model sizes more efficiently.

“Using a methodology like that, they’re able to continue to accelerate or reduce the time to train some of these crazy, multi-trillion parameter model sizes,” Kapoor explained at a press briefing.

Global Competition Also Heating Up

While Nvidia maintains a dominant position, competitors around the world are also pushing for breakthroughs. For example, China’s DeepSeek has recently claimed it can create competitive chatbots while using far fewer chips than many U.S. rivals, adding to the growing international race for AI supremacy.

MLCommons’ report also included results from Advanced Micro Devices (AMD) and others, though Nvidia’s Blackwell system stood out in the training category.

Meta Delays Launch of Flagship ‘Behemoth’ AI Model Over Performance Concerns

/in /tarafından

Meta Platforms (META.O) is delaying the release of its much-anticipated “Behemoth” AI model, the company’s most powerful large language model (LLM) to date, amid internal doubts about its performance and readiness, according to a report by the Wall Street Journal.

Originally slated for release in April to coincide with Meta’s inaugural developer AI conference, the internal launch target was later shifted to June. Now, the launch has been postponed to fall or later, people familiar with the matter said.

Reasons for Delay:

  • Engineers at Meta are reportedly struggling to make meaningful improvements in Behemoth’s performance compared to earlier models.

  • Staff have raised questions about whether the upgrades justify a public release, suggesting the model may not yet offer a significant leap over predecessors like Llama 3 or Llama 4.

Meta has not yet commented publicly on the delay, and the Behemoth model remains unreleased as of mid-May.

Development Context:

  • Meta had previously described Behemoth as “one of the smartest LLMs in the world”, intended to act as a teacher model for training smaller, faster models.

  • In April, Meta released other variants in its LLM family, including Llama 4 Scout and Llama 4 Maverick, but did not follow through with Behemoth’s public debut.

Industry Implications:

  • The delay highlights the growing technical challenges in scaling LLMs meaningfully, especially as performance gains become harder to achieve beyond a certain model size.

  • It comes at a time when AI competitors like OpenAI, Google, and Anthropic are releasing increasingly powerful models and tools, raising competitive pressure in the LLM arms race.

Meta’s pivot may reflect a more cautious release strategy, likely aimed at avoiding backlash over underwhelming capabilities or potential AI safety concerns.

Instacart CEO Fidji Simo Joins OpenAI as Chief of Applications

/in /tarafından

Fidji Simo, CEO of Instacart and former head of Facebook, will join OpenAI later this year as its new Chief of Applications, according to OpenAI CEO Sam Altman. Simo will report directly to Altman, who retains his role at the top of the Microsoft-backed AI company.

Key Developments:

  • Leadership Transition: Simo will step down from her CEO role at Instacart, but will remain Chair of the Board to assist with a smooth transition. A new CEO, expected to be an internal promotion, will be announced shortly, Simo said in an email to employees.

  • New Role at OpenAI: As Chief of Applications, Simo will oversee the development of consumer-facing products, including ChatGPT, and will play a pivotal role in expanding OpenAI’s product ecosystem.

  • Board Connection: Simo joined OpenAI’s board in March 2023, shortly after Sam Altman was reinstated following a dramatic ouster and return in late 2023.

  • Instacart Performance: Simo has led Instacart since 2021, taking the company public in September 2023 and steering it to profitability. The firm recently issued an upbeat forecast, citing strong demand in online grocery delivery.

  • Tech Background: Before Instacart, Simo spent over a decade at Meta, serving as head of Facebook from 2019 to 2021, and currently sits on Shopify’s board.

OpenAI’s move to hire Simo comes just days after the company reaffirmed its nonprofit governance structure, dampening Altman’s push for more direct control while preserving investor confidence in its commercial trajectory.