Anthropic says AI helped port Claude from Nvidia to AMD chips in three days
Anthropic says its internal assistant Mythos ported Claude to AMD graphics chips in about three days, a process that previously would have taken months.
Anthropic claims that artificial intelligence is weakening Nvidia's advantage in its CUDA software platform. According to the company, switching its Claude AI model to AMD graphics processors can now be done in about three days, while the same step previously took months. The claim comes solely from Anthropic itself and has not been independently verified.
According to the available information, the internal assistant Mythos carried out the porting work. Human involvement was reportedly minimal throughout the process. Anthropic has not disclosed the specific work steps involved. It is also unknown how capable and reliable the version of Claude running on AMD hardware is in operation.
The significance of the claim stems from the role of CUDA. Nvidia's software platform has long been considered one of the main reasons why developers of AI models mostly stay with Nvidia chips. Once programs have been built around CUDA, switching to other providers is difficult. Anthropic now sees this lock-in, often referred to in industry jargon as a moat, as being weakened by AI-assisted porting.
Anthropic states that Claude now runs on four families of chips. The company cites lower costs and reduced dependence on individual hardware vendors as its goals. The information provided did not describe any connection to a specific supply source or concrete supply agreements.
The information was reported by The Decoder. Further details on performance, stability and the exact porting process are not yet publicly known. The statements therefore rest entirely on the company's own explanations so far.
Source: the-decoder.com
28
