Thursday, 19 November 2026, story 6 of 7
OpenAI says its models improved AMD GPU design by up to 14%
The PPA gain took under five months, and OpenAI says AMD has put some of the changes into a design sent to manufacturing.

What happened
OpenAI says it worked with AMD to improve the power, performance and chip area (PPA) of AMD's flagship GPU IP by up to 14 percent in less than five months. According to OpenAI, that equals about a year of progress on designs AMD's engineers had already optimized for years. OpenAI says AMD has folded some of the changes into a design sent to manufacturing.
The same method also helped with Jalapeño, OpenAI's first custom inference chip, OpenAI says. In tests on open RISC-V cores, GPT-6.2 Astra gave the best results across model generations. OpenAI is also working with Synopsys on GPT-Synopsys, a model that uses Synopsys's EDA tools.
What it means
All figures here are OpenAI's own, and the words up to signal a best case. The source does not say how the gain splits across power, performance and area, and it quotes no statement from AMD. The strongest detail is that some changes reached a design sent to manufacturing, though the source does not say how many.
For builders, nothing here can be used yet. GPT-Synopsys is described only as work under way, with no date or price, and the source does not say whether GPT-6.2 Astra is available. If the claims hold up, chip design teams may face pressure to shorten their optimization cycles. Anthropic's claim of three days instead of six to twelve months for a task on AMD MI355X GPUs points the same way, but it is also the company's own number.
What to do: Treat the 14 percent as OpenAI's best-case figure, and wait for AMD's own confirmation and GPT-Synopsys release details before changing any plans.
Source
Filed under Chips and compute. Published .
"What it means" is analysis, not reporting. Found an error? Write to daily@vibitely.com.