Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

OpenAI details model-driven chip design: 13% area savings on Jalapeño, 14% PPA gain on AMD GPUs

OpenAI described how its models cut standard-cell area by roughly 13% in Jalapeño's warp engine and 10% in its matrix unit, and delivered up to 14% PPA improvement over heavily optimized baselines on AMD's flagship GPU IP. It also reported GPT-6.2 Astra leading evaluations on two open-source RISC-V cores and a new partnership with Synopsys to train GPT-Synopsys on EDA tools.

Published

OpenAI published a detailed account on November 18 of how its frontier models improved the physical design of its own inference chip, Jalapeño, and of AMD silicon. It is the clearest public evidence so far that models can contribute net-new gains to production hardware, not just accelerate code writing.

The post says engineers began using GPT-4o on Jalapeño in early 2025, and even that early model found non-trivial room for PPA improvements. The team estimates that model-proposed changes reduced standard-cell area by about 13% in the warp engine and 10% in the matrix unit, savings that were silicon-proven during the chip's bring-up.

The circuit-level detail is concrete: one multiplier redesign used roughly 15% less cell area in synthesis while dropping latency from two clock cycles to one, and a separate adder redesign cut area by about 15% while holding the same two-cycle latency. The models explored both alternative logic implementations and how to split work across pipeline stages, sharing hardware and trimming intermediate storage.

The AMD results go further. By rewriting SystemVerilog RTL, OpenAI's model delivered up to a 14% PPA improvement against heavily optimized baselines on AMD's flagship GPU IP, with Synopsys tools confirming that each rewrite preserved the block's behaviour. OpenAI says the work compressed roughly a year of progress into less than five months. AMD has already placed some changes in a design sent for manufacturing and added others to production design files for future versions.

On the CPU side, AMD verified single-digit percentage PPA gains in key blocks of its upcoming next-generation flagship CPU using designs written by the model. OpenAI says the model reasoned about RTL at the level of an expert microarchitect, rewriting circuits that perform calculations, control operations, or move data while weighing power, performance and area trade-offs.

To measure generational progress, OpenAI tested successive models on two open-source 32-bit RISC-V cores: PicoRV32, a compact core for FPGAs, and Ibex, the embedded processor inside OpenTitan's security chip. At a comparable budget of about 112,000 tokens per run, GPT-6.2 Astra posted the highest average score on both cores and GPT-6.1 Sol outscored GPT-5.6 Sol, across 9,881 scored runs.

The methodology also carries limits worth reading closely. Designs scored zero if they failed simulation checks or exceeded 16 times the original core's area, and when models designed a core from written requirements they could omit features such as performance counters, which may explain part of the scoring advantage. OpenAI notes a more production-like test would lock those features, tighten the area cap and check timing after physical synthesis.

Next up is a partnership with Synopsys to train GPT-Synopsys, a model meant to drive EDA tools the way expert designers do. The signal is less about any single chip than about direction: AI is moving from suggesting code to rewriting production silicon, which puts verification and accountability at the centre of the next phase.

Why it matters

If model-written RTL can reliably clear verification and reach production, AI moves into the most expensive and expertise-bound part of chip development, potentially shortening design cycles and reshaping how EDA tools are used.

OpenAISemiconductorsAMD
Back to realtime news

Nearby Updates

All

11/18, 08:00

Reson8 open-sources Rumble-2-Flash, a 0.5B speech model for 40+ languages

Reson8 released Rumble-2-Flash on November 18 under Apache 2.0: a 0.5B-parameter speech recognition model covering more than 40 languages that runs natively in Transformers and vLLM. It posts a 6.96% average error rate on the Open ASR Leaderboard's multilingual track, beating Meta's 15x larger omniASR-LLM 7B and matching a 30B Nemotron model on a 302-hour call-centre benchmark.

11/18, 08:00

OpenAI and Groupe Figaro partner to bring Le Figaro journalism into ChatGPT

OpenAI and Groupe Figaro announced a strategic partnership that makes Le Figaro's current reporting and archive content discoverable inside ChatGPT through summaries and excerpts with links to the original stories. Both sides said the publisher keeps full editorial independence and control over its journalism.

11/18, 08:16

Nvidia says record revenue run won't stop, even amid leadership turmoil

Nvidia said its string of record-breaking revenue is not about to end, even as the company navigates the fallout from founder and CEO Jensen Huang's departure, TechCrunch reported. The report attaches a caveat to the optimism: the projection is framed as holding for now.

11/18, 07:37

Nvidia's interim CEO says chip sales to China have stopped, TechCrunch reports

TechCrunch reported on November 17 that Nvidia's interim CEO has said the company is no longer selling chips to Chinese customers, framing the statement against a leadership upheaval. The story asks what the decision will cost but gives no revenue figure or timeline.