NVIDIA launches codesign strategy to optimize memory use in AI

NVIDIA unveils a new codesign strategy to optimize memory use in AI. The technique includes NVFP4, CMX and NVL72, which tune efficiency at every level of the hierarchy. The GeForce RTX 5090, which is already sold out, is priced between US$ 7,857 (converted from EUR 7,000) and US$ 11,224 (converted from EUR 10,000).

The AI crunch has shifted the production bottleneck from chips to electrical power. The company invested US$ 3,000,000,000 in Meta, which plans to use more of the company's hardware for its AI agents. The Radeon RX 9070 XT has grown and ranks 21st on the list of most popular GPUs. Full HD resolution reaches 50% share among PC users. NVIDIA is facing pressure from the rise of larger AI models and longer contexts, which demand more efficient memory use. Demand for AI hardware is affecting the gaming market, with graphics cards such as the RTX 5090 selling out.

For those who use NVIDIA accelerators, this means greater memory efficiency without the need for additional hardware. The new strategy may also affect demand for AI chips, since efficiency improves with fewer resources. NVIDIA has already released information about the use of AI agents in factory operations and about technology to optimize GPUs in Kubernetes, which reinforces the consistency of the approach.

AI pressure shifts chip production priorities

The AI crunch, which demands larger models and longer contexts, is pressuring the company to seek more efficient solutions. At the same time, NVIDIA is investing in partnerships, such as with Meta, which plans to use more of its solutions for its AI agents. The new strategy may affect the gaming market by reducing the availability of GPUs for games, but it may also bring benefits to companies that use NVIDIA accelerators. The pressure for memory efficiency is shifting chip production priorities, with a focus on electrical power rather than just manufacturing capacity.

New approach shifts resource priorities in AI

The new approach tunes efficiency instead of relying on isolated components, measuring memory use at every level of the hierarchy. This reduces waste in computing processes and has redefined resource priorities in AI systems.

With information from blogs.nvidia.com, the-decoder.com and multiplayer.it.

Prices converted to US dollars from the original currency at the reference rate of 2026-11-05 (European Central Bank).

Text written with AI from the sources listed and checked by the newsroom's automated rules. How we work

• #hardware
• #software