Inside The Metrics Behind OpenAI’s Up To 14% AMD Chip Design Gains
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Inside The Metrics Behind OpenAI’s Up To 14% AMD Chip Design Gains on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

OpenAI said on November 18 that its models rewrote parts of AMD chip designs, producing up to 14% better power, performance and area against heavily optimized baselines. The figure is OpenAI’s reported best result; the companies have not published block-level measurements or an independent audit, and AMD has disclosed no figures of its own.

OpenAI said November 18 that its models rewrote parts of designs for AMD processors, reporting up to 14% improvement in power, performance and area, or PPA, on AMD’s flagship GPU design. The company said some model-generated changes had been included in a design sent for manufacturing, but the headline figure is a best-case result whose blocks, average gain and independent verification have not been made public.

The results appeared in an OpenAI post titled “Advancing chip design with OpenAI models.” OpenAI’s account, as reported by ANI and other outlets, describes several separate projects rather than one uniform test. For the AMD GPU, the company compared model-generated changes with heavily optimized baselines and reported gains of up to 14% across PPA. It did not specify which design blocks produced that peak or how representative it was of the chip as a whole.

OpenAI also said AMD had verified single-digit PPA gains in key blocks of its next flagship CPU. According to OpenAI’s description, AMD had already put some model-generated changes into a recent design sent for manufacturing, with other changes placed in design files for future products. AMD CEO Lisa Su praised the work on X, but AMD has not released its own measurement breakdown.

In a separate effort on OpenAI’s custom Jalapeño inference chip, the company said model-proposed changes reduced standard-cell area by about 13% in one unit and 10% in another. OpenAI said those changes were validated after the first chips came back. Reports differ on the name of one unit, describing it as a warp engine or vector unit. OpenAI also reported testing five models on the open-source PicoRV32 and Ibex processor cores, with 9,881 scored runs; its post said GPT-6.2 Astra, released the day before, scored highest.

The design work targeted register-transfer-level code, known as RTL, which describes a chip’s behavior cycle by cycle. OpenAI said its models proposed alternative implementations, while Synopsys design tools checked whether rewritten code preserved the original behavior and assessed PPA. The models therefore generated candidate changes; conventional verification and design tools performed the reported checks.

At a glance
reportWhen: Announced November 18, 2026
The developmentOpenAI reported that its models improved selected AMD chip-design metrics, with a peak gain of up to 14% in power, performance and area.

AI Moves Into Chip Design

If the reported changes survive manufacturing and later testing, they could help chip teams improve designs without relying solely on engineers to manually rewrite circuit logic. Gains in power, performance and area can affect how much computing a chip delivers, how much electricity it consumes and how much silicon it requires. Even improvements confined to selected blocks may matter in products built at scale, though the disclosed figures do not establish the effect on an entire processor.

The development also illustrates a feedback loop: AI models are being used to change hardware designed to run AI systems. OpenAI president Greg Brockman described the AMD work as “roughly an additional year of progress” achieved in less than five months using the company’s models, according to a post quoted by Stocktwits. That is OpenAI’s characterization of the pace of progress, not an independently measured timetable or proof that chip development overall has been shortened by a year.

For readers evaluating the claim, the distinction between a local design improvement and a product-level gain is central. A best result in one block does not mean the whole GPU is 14% faster, uses 14% less power or is 14% smaller. The company has not disclosed enough detail to calculate those broader effects.

Products Worth Considering

Logitech MX Creative Console with 9 Customizable Keys, Graphite

Logitech MX Creative Console with 9 Customizable Keys, Graphite

Dell XPS Tower 8960 Desktop, Intel Core i7-13700, NVIDIA GeForce RTX 4060, 16GB DDR5, 512GB SSD

Dell XPS Tower 8960 Desktop, Intel Core i7-13700, NVIDIA GeForce RTX 4060, 16GB DDR5, 512GB SSD

As an affiliate, we earn on qualifying purchases.

From RTL Code to Silicon

RTL is an early, consequential layer of chip development. Engineers use it to describe how a circuit’s components operate and exchange data. Synthesis software converts that description into logic gates; later design stages place components and connect them before manufacturing. A change that looks promising at the RTL stage still has to work through those subsequent steps.

According to OpenAI’s account, Synopsys tools checked that candidate rewrites preserved the original behavior and evaluated their PPA trade-offs. That verification matters because a faster or smaller circuit is not useful if it changes the intended computation. OpenAI’s reported Jalapeño examples included a multiplier redesign that used about 15% less cell area in synthesis while reducing latency from two clock cycles to one, and an adder redesign that reduced cell area by about 15% at the same two-cycle latency, as described by Inside AI.

The claims come amid a commercial relationship between the companies. The source material says OpenAI holds a warrant, issued in October 2025, for up to 160 million AMD shares, with vesting tied to AMD chip deployments. That financial connection does not establish that the engineering results are wrong, but it makes clear why independent measurements and fuller disclosures would help readers assess them.

“Su praised the results.”

— Lisa Su, AMD CEO, on X, as quoted by Stocktwits

The Missing Measurement Details

The reported 14% is an “up to” figure. OpenAI has not identified the GPU blocks behind the peak, their share of the full design, the average result, or the precise design stage at which the measurement was taken. The source reports say the baseline was heavily optimized, but do not provide enough detail for outside researchers to reproduce the comparison.

No independent audit of the AMD results has been published in the material available for this report. AMD’s confirmation, as relayed by OpenAI, is limited to single-digit gains in key blocks of its next flagship CPU. AMD has not issued a public technical breakdown, and the source material does not clarify how many changes were in the design sent for manufacturing or whether they have been tested in production silicon.

There are also differences in reporting on the name of one Jalapeño unit, and no complete set of model-by-model results for the open-source-core tests is provided here. The reports agree on the central claims, but readers cannot independently check the companies’ underlying data, tool settings or full test methods from the figures supplied.

Verification After Manufacturing

The next substantive evidence would be detailed results from AMD or OpenAI showing which design blocks changed, how PPA was measured, and whether the reported gains persisted through later implementation and silicon testing. OpenAI says some changes were included in a design sent for manufacturing; the available account does not identify a publication date for test results from that design.

Readers can also look for AMD’s own technical disclosures on the next flagship CPU and GPU, including any figures that distinguish local block gains from whole-chip performance, power use or area. Until those details or an independent assessment are available, the 14% figure should be read as OpenAI’s reported best case, not as a verified improvement across an AMD product.

Key Questions

What does OpenAI’s 14% figure measure?

OpenAI reported up to 14% improvement in power, performance and area on AMD’s flagship GPU design, compared with what it described as heavily optimized baselines. It did not publish an average or say which blocks produced the peak result.

Does this mean AMD’s whole GPU is 14% better?

No. The reported number is a best-case result for unspecified design blocks. OpenAI has not said that the whole GPU is 14% faster, uses 14% less power or is 14% smaller.

Who checked the model-generated designs?

OpenAI said Synopsys design tools verified that rewritten RTL preserved the original behavior and assessed design trade-offs. The source material does not describe an independent audit of the results.

Has AMD confirmed the reported gains?

OpenAI said AMD verified single-digit PPA gains in key blocks of its next flagship CPU. AMD CEO Lisa Su praised the work publicly, but AMD has not published its own figures or a detailed measurement report in the source material.

What happens next?

OpenAI says some changes went into a design sent for manufacturing. Further assessment depends on technical disclosures or test results showing how those changes perform after implementation and, where applicable, in manufactured chips.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Silvaco To Accelerate Physics-Based Digital Twins For Semiconductor Design And Manufacturing Using NVIDIA AI And Accelerated Computing

Silvaco announces plans to enhance physics-based digital twins for semiconductor design by leveraging NVIDIA AI and accelerated computing technologies.

AI, Buyers And The Wage Gap: Inside Bangladesh’s Garment Worker Protests

Bangladesh garment workers demand higher pay after a wage-board meeting was postponed. Talks are due to resume Nov. 23 amid protests and factory closures.

Robot AI In The Spotlight After OpenAI Splits Its Team And Runway Opens Praxis-1

OpenAI has folded robotics teams into other groups, while Runway released Praxis-1, an open-weight robot model built for faster control.

How Rubin HPC Could Scale Nvidia GPU Power For Science By 2028

Nvidia says Rubin HPC will have more than 50,000 CUDA cores and target science and engineering workloads. It is expected in 2028.