AMD turns to AI startups to inform chip, software design
SAN JOSE (Reuters) -Advanced Micro Devices has forged close ties to a batch of artificial intelligence startups as part of the company's effort to bolster its software and forge superior chip designs.
As AI companies seek alternatives to Nvidia's chips, AMD has begun to expand its plans to build a viable competing line of hardware, acquiring companies such as server maker ZT Systems in its quest to achieve that goal.
But to build a successfulline of chips also requires a powerful set of software to efficiently run the programs built by AI developers. AMD has acquired several small software companies in recent weeks in a bid to boost its talent, and it has been working to beef up its set of software, broadly known as ROCm.
"This will be a very thoughtful, deliberate, multi-generational journey for us," said Vamsi Boppana, senior vice president of AI at AMD.
AMD has committed to improve its ROCm and other software, which is a boon to customers such as AI enterprise startup Cohere, as it results in speedy changes and the addition of new features.
Cohere is focused on building AI models that are tailored for large businesses versus the foundational AI models that companies like OpenAI and others target.
AMD has made important strides in improving its software, Cohere CEO Aidan Gomez said in an interview with Reuters. Changing Cohere's software to run on AMD chips was a process that previously took weeks and now happens in only "days," Gomez said.
Gomez declined to disclose exactly how much of Cohere's software relies on AMD chips but called it a "meaningful segment of our compute base" around the world.
OPENAI INFLUENCE
OpenAI has had significant influence on the design of the forthcoming MI450 series of AI chips, said Forrest Norrod, an executive vice president at AMD.
AMD's MI400 series of chips will be the basis for a new server called "Helios" that the company plans to release next year. Nvidia too has engineered whole servers in part because AI computations require hundreds or thousands of chips strung together.
OpenAI's Sam Altman appeared on stage at AMD's Thursday event in San Jose, and discussed the partnership between the two companies in broad terms.
Norrod said that OpenAI's requests had a big influence on how AMD designed the MI450 series memory architectureand how the hardware can scale up to thousands of chips necessary to build and run AI applications.
The ChatGPT creator also influenced what kinds of mathematical operations the chips are optimized for.
"(OpenAI) has given us a lot of feedback that, I think, heavily informed our design," Norrod said.
(Reporting by Max A. Cherney in San JoseEditing by Shri Navaratnam)
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
The Star
10 hours ago
- The Star
Voice of doubt in Nvidia's H20 chips grows louder in China
US chip giant Nvidia faces increasing mistrust in China over its H20 chips, after the company secured a licence from Washington to export the tailor-made processors by agreeing to pay 15 per cent of its revenues to the US government. On Sunday, a Chinese state media outlet labelled Nvidia's H20 chip as unsafe and urged Chinese buyers to boycott the artificial intelligence processors designed for the Chinese market. Despite Nvidia's repeated denials of 'back-door' vulnerabilities, Yuyuan Tantian, associated with state broadcaster China Central Television, said the H20 chip could not be trusted because of the US government's alleged efforts to embed 'back doors' in the chips. 'When a chip is neither environmentally friendly nor advanced or safe, as consumers, we certainly have the option not to buy it,' the outlet said in a commentary. It suggested that the H20 chip could be remotely controlled through hardware and software exploits, citing Washington's attempts in 1992 to implement surveillance via chip back doors and a recent bill proposed in May requiring US chipmakers to incorporate tracking features. Nvidia's chief security officer David Reber Jnr published a blog post last week titled: 'There are no back doors in Nvidia chips. No kill switches. No spyware.' Chinese state media's allegations come as the US Department of Commerce has begun issuing licences to Nvidia for exporting H20 chips to China. Nvidia and rival Advanced Micro Devices agreed to pay the US government 15 per cent of their revenues from chip sales to China as part of a deal with the Trump administration to secure export licences, the Financial Times reported on Sunday. Yuyuan Tantian's article reflects Nvidia's struggle to establish trust in China, which accounted for 13 per cent of its revenues in the last financial year despite escalating technology rivalry between Beijing and Washington. Last month, the Cyberspace Administration of China (CAC) summoned Nvidia for discussions about the risks associated with potential tracking and remote control functions of its H20 chips. In an opinion piece titled 'How can we trust you, Nvidia?' earlier this month, the People's Daily, the official newspaper of the Central Committee of the Communist Party, called on the US chipmaker to provide proof of its security measures to regain the trust of Chinese customers. In response, Nvidia said it 'does not have 'back doors' in our chips that would give anyone a remote way to access or control them'. The H20, an AI graphics processing unit released by Nvidia last year specifically for China to comply with US export restrictions, has become a focal point in the tech war between the two countries after Washington lifted a sales ban on the chip last month. While the resumption of H20 sales is expected to provide a near-term boost to China's AI development, Nvidia faces challenges amid Beijing's scrutiny and a push from domestic firms to adopt home-grown processors, such as those developed by Huawei Technologies. The CAC's review served partly 'as a reminder to domestic firms to avoid over-reliance on a single chip and to always be prepared to use alternatives', Chinese semiconductor research firm ICwise said in a note last week, which has since been deleted. The firm also said that China's interest in the H20 chip was likely to 'continue to wane' as the country's chip sector advanced. - SOUTH CHINA MORNING POST
The Star
11 hours ago
- The Star
Anthropic offers AI chatbot Claude to US government for $1
FILE PHOTO: Anthropic logo is seen in this illustration taken May 20, 2024. REUTERS/Dado Ruvic/Illustration/File Photo (Reuters) - Anthropic said on Tuesday it will offer its Claude AI model to the U.S. government for $1, joining a growing list of artificial intelligence startups proposing lucrative deals to win federal contracts. This comes days after OpenAI's ChatGPT, Google's Gemini and Anthropic's Claude were added to the government's list of approved AI vendors. ""America's AI leadership requires that our government institutions have access to the most capable, secure AI tools available," CEO Dario Amodei said. Rival OpenAI had announced a similar offer last week, wherein ChatGPT Enterprise was made available to participating U.S. federal agencies for $1 per agency for the next year. (Reporting by Arsheeya Bajwa in Bengaluru)
Free Malaysia Today
13 hours ago
- Free Malaysia Today
Nvidia, AMD reach deal to give US a cut of China AI chip sales
AMD's shares gained less than 1% to US$173.05 in New York yesterday, while Nvidia's shares were little changed. (Exness pic) NEW YORK : Nvidia Corp and Advanced Micro Devices Inc (AMD) have agreed to pay 15% of their revenues from Chinese AI chip sales to the US government in an unusual, legally questionable deal that reflects the Trump administration's willingness to soften export controls in exchange for financial payouts. 'Nvidia plans to share 15% of the revenue from sales of its H20 AI accelerator in China,' US President Donald Trump said in a briefing with reporters yesterday. 'AMD will deliver the same share from MI308 revenues,' a person familiar with the situation said, asking not to be identified discussing internal deliberations. Trump said he'd originally told Nvidia that he wanted a 20% cut for the US if he cleared H20 sales to China, but eventually settled for a 15% share. The two negotiated 'a little deal,' he said. The chip companies' consent to cede part of chip sales underscores the urgency of their desire to cater to customers in the world's second-largest economy. The US government has cut off sales of the most capable AI chips, arguing that China might use them for military purposes, instead allowing shipment of only pared-back products. However, the Trump administration has consistently relented on trade conditions in exchange for financial concessions — and in this case officials said the chips are not the most advanced, playing down their national security implications. There's no guarantee the arrangement with Nvidia and AMD will succeed. Trade experts said it's vulnerable to legal challenges because it could be construed as an export tax, something that's not allowed under the constitution. The chipmakers themselves said it will take months to revive production of the parts — assuming Chinese customers even opt for dated components. China's government, meanwhile, has grown increasingly hostile to the idea of Chinese firms deploying the H20 and is unlikely to warm to the idea of a chip tax. Yuyuantantian, a social media account affiliated with state-run China Central Television that regularly signals Beijing's thinking about trade, on Sunday slammed what it described as security vulnerabilities and inefficiencies of Nvidia's chip. AMD shares gained less than 1% to US$173.05 at 3.25pm in New York yesterday. Nvidia shares were little changed. 'Both Nvidia and AMD already said they would start shipping to China, so that market reaction already happened,' said Jay Goldberg, an analyst at Seaport Global Securities. 'The big question is exactly when they're going to start delivering to China again, especially now that there are strings attached,' Goldberg said. 'The deal also threatens to undermine the US argument that some trade controls are necessary to safeguard national security,' said Jacob Feldgoise, a researcher at the DC-based Center for Security and Emerging Technology. 'This seeming quid pro quo is unprecedented from an export-control perspective. 'The arrangement risks invalidating the national security rationale for US export controls,' he added. When the Trump administration first decided to grant export licenses to Nvidia and AMD last month, treasury secretary Scott Bessent said exports of H20 chips were part of trade talks with China and were used as 'a negotiating chip,' White House AI adviser David Sacks emphasised at the time that the product wasn't 'the latest and greatest'. Trump reiterated these points at the briefing yesterday, calling Nvidia's H20 'an old chip' and hailing the company's latest Blackwell chip as the 'super duper advanced' one. He signaled though that he'd be open to negotiate another deal with Nvidia CEO Jensen Huang to sell a scaled-back version of the most advanced Blackwell chip to China, too. 'I think he's coming to see me again about that,' Trump said. An Nvidia spokesman said the company follows US export rules, adding that while it hasn't shipped H20 chips to China for months, it hopes the regulations will allow US companies to compete in China. AMD similarly said in a statement yesterday that it's adhering to all US export-control laws. The US government has meanwhile begun approving export licenses for the chips. AMD's initial license applications have been cleared, the company said yesterday. The Financial Times earlier reported on the revenue-sharing deal. Trump has targeted chipmakers in the past week with a series of declarations that were light on specifics and rattled companies from Silicon Valley to Asia. On Wednesday, Trump threatened 100% tariffs on imported chips, unless companies also made investments on US soil. On examination, though, those new tariffs would apply to almost no one since most major chipmakers appear to be covered by existing investments or separate trade deals. On Thursday, Trump called on Intel Corp CEO Lip-Bu Tan to resign, describing the Malaysia-born entrepreneur as 'highly conflicted' without giving details. Tan, who sent a letter to employees assuring them that he had engaged with the administration, is expected to meet with Trump yesterday, a person familiar with the situation said. The Wall Street Journal was first to report on the meeting. Tan has been targeted by Republican senator Tom Cotton over historical business ties to China. Huang has lobbied long and hard for the lifting of restrictions, arguing that walling China off will only slow the spread of American technology and encourage local rivals such as Huawei Technologies Co. In yet another sign that his message is getting through to the White House, Trump described Huang yesterday as a 'great', 'very brilliant guy'. The tax is expected to funnel some capital to the US — but not an enormous amount in relative terms. Both Nvidia and AMD have said it'll take time to ramp back up production of their China-specific products — even if order levels return to previous levels, which is uncertain. Nvidia raked in US$4.6 billion of revenue from the H20 in the fiscal quarter ended April 27 — days after new restrictions on shipping the AI accelerator to China were imposed. It also said it had been unable to ship US$2.5 billion of H20 China revenue in that period because of the new rules. That implies it would have got more than US$7 billion in H20 sales to China during the period. If it can return to that level, the US government will stand to get about a billion dollars a quarter from its deal. AMD could generate US$3 billion to US$5 billion of 2025 revenue if restrictions were lifted, Morgan Stanley estimates. Chinese alternatives such as Huawei's Ascend chips now account for 20% to 30% of domestic demand, it reckoned. 'The US government clearly needs the money given its deficits and eagerness to collect tariffs,' said Vey-Sern Ling, managing director at Union Bancaire Privee in Singapore. 'But the complication is China's accusations about H20 chips containing back doors, which could be a negotiation tactic to highlight that the country is not 'hard up' for US chips,' Ling added. For its part, Nvidia emphasised that its H20 chip is 'not a military product or for government infrastructure'. China has 'ample supply of domestic chips,' the company said in an emailed statement. 'It won't and never has relied on Americans chips for government operations, just like the US government would not rely on chips from China,' it added.
