OpenAI and AMD use AI models to speed up GPU design optimization, Anthropic reports similar results
Using AI models, OpenAI and AMD achieved up to a 14% improvement in energy efficiency, performance, and area (PPA) for AMD's GPU IP in under five months — according to OpenAI, this represents a year's worth of additional progress. Anthropic reports a similar speedup for the MI355X chip.
OpenAI, in collaboration with AMD, says it used its own AI models to achieve up to a 14% improvement in the PPA metric (power, performance, area — a chip's energy efficiency, performance, and area) for AMD's flagship GPU IP, in under five months. According to OpenAI, this corresponds to roughly a year of additional development on top of a design that AMD's engineers had already been optimizing for years. According to OpenAI, AMD has already incorporated some of these adjustments into a design intended for production.
OpenAI used the same method on Jalapeño, its first in-house inference chip. In tests on open RISC-V processor cores, the GPT-6.2 Astra model achieved the best results among the model generations. OpenAI is also collaborating with Synopsys on the GPT-Synopsys model, which uses Synopsys's EDA tools and, according to OpenAI, is meant to make AI-assisted chip design accessible to a wider range of users.
Anthropic also uses AI tools for chip design. Its Mythos model processed prompt processing for one model on AMD MI355X GPUs in three days, instead of the usual six to twelve months that Anthropic says this task normally requires. Anthropic currently runs Claude models on four chip families with the goal of optimizing the cost-to-performance ratio.
Why it matters
Shortening chip design optimization cycles from months to days or weeks can reduce hardware development costs and speed up the arrival of new GPU generations on the market, which is relevant both for chip design companies and for AI infrastructure providers dependent on the availability of high-performance hardware. For individual users of AI products, the sources do not indicate any direct impact — this concerns company-level collaboration in the semiconductor design field.
Relevant practical impact
What this means
For a business
Companies in the semiconductor design and AI infrastructure space may, thanks to AI-assisted design (e.g. the GPT-6.2 Astra, GPT-Synopsys, or Mythos models), shorten chip optimization cycles from months to days and reduce the cost of developing and operating computing hardware.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.