SA-21/06, C-11, Asapur, Varanasi, UP support@kamaldnp.com +91 8318013442
Technology

OpenAI's AI Models Rewrote AMD Chip Circuits: Up to 14% PPA Gains, Some Changes Already in Manufacturing

KAMALDNP Raj Kamal
November 21, 2026
3 Min Read
3 Views
OpenAI's AI Models Rewrote AMD Chip Circuits: Up to 14% PPA Gains, Some Changes Already in Manufacturing

OpenAI reported on 18 November 2026 that its AI models improved AMD's flagship GPU designs by up to 14% across power, performance and area (PPA) — relative to baselines AMD's engineers had already heavily optimised. The models rewrote SystemVerilog RTL, the code that describes chip circuits, with Synopsys tools verifying that each change preserved the original behaviour. AMD has already included some model-generated changes in a chip design sent for manufacturing, and verified single-digit PPA gains in key blocks of its upcoming flagship CPU.

What the 14% figure does and does not mean

The 14% figure is a combined PPA number. The individual contributions from power, speed and area were not disclosed. It is not a 14% improvement in overall GPU performance, and it should not be read as one.

Key facts

ItemDetail
Reported18 November 2026
PPA improvementUp to 14% across power, performance and area (combined figure)
BaselineAMD's flagship GPU designs, already heavily optimised by AMD's engineers
What the models changedSystemVerilog RTL — the code describing chip circuits
VerificationSynopsys tools checked that each change preserved the original behaviour
ManufacturingSome model-generated changes already included in a chip design sent for manufacturing
Flagship CPUSingle-digit PPA gains verified in key blocks of AMD's upcoming flagship CPU
OpenAI's framingRoughly a year of chip-design progress in under five months
First use of the techniqueOpenAI's own first custom inference chip, Jalapeño — standard-cell area cut by about 13% in its vector unit and 10% in its matrix unit
Next stepOpenAI and Synopsys are now training a specialised GPT-Synopsys model

What this means / what to watch

The striking part of the report is the baseline: these gains came on top of designs AMD's engineers had already heavily optimised, and some of the model-generated changes have already passed into a chip design sent for manufacturing. The same technique had first been used on OpenAI's own first custom inference chip, Jalapeño, where it cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit. The next thing to watch is the specialised GPT-Synopsys model that OpenAI and Synopsys are now training — and whether further PPA figures, with the power, speed and area contributions broken out, are disclosed.

A note on verification

The facts above are verified from the sources named in the research file: Economic Times / ANI, DataCenterDynamics, Analytics India Magazine, Gadgets Now, Silicon Report, bpdata and WCCFTech. The 14% figure is reported as a combined PPA number in those sources; no source disclosed the individual power, speed and area contributions, and none presented it as a 14% improvement in overall GPU performance.

FAQs

Did AMD's GPUs get 14% faster?
No. The 14% figure is a combined PPA number — power, performance and area taken together. The individual contributions were not disclosed, and it is not a 14% improvement in overall GPU performance.

What did the AI models actually change?
The models rewrote SystemVerilog RTL, the code that describes chip circuits. Synopsys tools verified that each change preserved the original behaviour.

Are any of these changes in a real chip?
Yes. AMD has already included some model-generated changes in a chip design sent for manufacturing, and verified single-digit PPA gains in key blocks of its upcoming flagship CPU.

What is Jalapeño?
Jalapeño is OpenAI's first custom inference chip, where this technique was first used. It cut standard-cell area by about 13% in the chip's vector unit and 10% in its matrix unit.

Source: Economic Times / ANI; DataCenterDynamics; Analytics India Magazine; Gadgets Now; Silicon Report; bpdata; WCCFTech. Last checked: 21 November 2026.

Share this post

Comments (0)

Share your question or experience about this syllabus. Your comment will appear after review.

No comments yet — be the first to share your thoughts!

🤖 KAMALDNP AI Assistant

Ask about courses, website & more

Hi! I'm KAMALDNP AI Assistant. Ask me about our courses, website features, or anything else! 😊
Please wait 5s...
🤖 KAMALDNP Assistant ✕

Request human callback:

🔐 Login Required
✕

Please log in to download. You will stay on this page after logging in!

New here? Register now →