AMD's New AI Technology Could Reduce RAM Usage by 75%

AMD's new AI technology could reduce server RAM requirements by 75% while maintaining performance across demanding workloads.

Hardware by Nahe Yan on  Nov 18, 2026

The ongoing memory crisis has raised DRAM costs, pressuring companies that rely on large memory capacities to run their systems. However, AMD may have found a way to reduce memory requirements without sacrificing much performance. The company's new AI-based technology allows SSD storage to work alongside system RAM by predicting which data a system will need next.

AMD claims the approach can reduce physical RAM requirements by up to 75% while retaining around 80% of the original performance in certain workloads. AMD has announced a new solution to the memory crisis, and it's already being used in real-world deployments. The idea starts with a simple question: What if systems didn't need so much expensive RAM in the first place?

AMD AI Technology Can Reduce RAM Usage

AMD Introduces AI Technology

The technology comes from a company AMD acquired earlier this year. In simple terms, it lets a much cheaper SSD partially replace system RAM. This could help companies operate systems with less physical memory while maintaining performance. You might be thinking that this sounds like a page file. It's similar, but the difference matters.

A page file moves data to your drive when RAM runs out. The problem is speed. Even the fastest NVMe drives respond much more slowly than DDR5 memory. Every time your system needs something that isn't in RAM, it has to wait, and performance drops. AMD's approach tries to avoid that delay by using AI to predict what data the system will need next. It then moves that data from the SSD back into RAM before the system requests it.

Instead of stopping to fetch the required data, the system finds it already available in memory. The result is the capacity of a much larger memory setup without paying for nearly as much DRAM. The irony is hard to ignore. AI is one reason memory has become so expensive, and now AI might help address the problem.

But does it actually work? AMD's benchmarks show some interesting results. In one chip design workload, AMD maintained 95% of the original performance with half the RAM. Even after cutting physical RAM by 75%, the system still retained around 80% of its original performance. AI workloads showed even better results.

AMD claims 75% lower RAM usage with limited performance loss.

In a test running five AI agents simultaneously, AMD maintained 99% of the original performance with half the RAM. Even with just 25% of the original memory capacity, performance remained around 85%. These are AMD's own benchmarks so that real-world results will vary depending on the workload and system configuration.

However, the results indicate that systems could operate with significantly less physical RAM while retaining much of their original performance. This is particularly relevant for workloads that require large memory capacities. If companies can reduce their DRAM requirements without experiencing a major performance drop, they could potentially lower hardware costs and allocate their budgets elsewhere.

One important detail to understand. Right now, AMD's technology is designed for server chips rather than consumer PCs. However, that doesn't mean it is irrelevant to the consumer hardware market. Servers account for a significant share of global memory demand. Data centers require large amounts of DRAM to operate their infrastructure, particularly as AI workloads continue to expand.

FAQ

Q. How can AMD's new technology reduce RAM requirements by 75%?

A. AMD's technology uses AI to predict which data a system will need next and moves that data from an SSD into RAM before it is requested. This lets systems run with less physical memory while retaining much of their original performance.

Q. How much performance can AMD's technology retain with 75% less RAM?

A. According to AMD's benchmarks, a chip design workload retained around 80% of its original performance after physical RAM was reduced by 75%. In an AI workload running five agents simultaneously, performance remained around 85% with just 25% of the original memory capacity.

Q. Is AMD's new AI memory technology available for consumer PCs?

A. No. The technology currently targets server systems. However, reducing data center memory requirements could eventually ease DRAM demand and reduce pressure on the broader memory market.

Q. How much could AMD save with its new memory technology?

A. AMD projects savings of more than $60 million in memory expansion costs. The company has already deployed the technology across 145 of its servers and plans to expand that number to around 2,000 by the end of next year.

Nahe Yan

Editor, NoobFeed

Gaming Hardware Updates

No Data.