Researchers on the Korea Superior Institute of Science and Expertise (KAIST) have developed energy-efficient NPU expertise that demonstrates substantial efficiency enhancements in laboratory testing.
Their specialised AI chip ran AI fashions 60% sooner whereas utilizing 44% much less electrical energy than the graphics playing cards at present powering most AI methods, primarily based on outcomes from managed experiments.
To place it merely, the analysis, led by Professor Jongse Park from KAIST’s Faculty of Computing in collaboration with HyperAccel Inc., addresses probably the most urgent challenges in trendy AI infrastructure: the big vitality and {hardware} necessities of large-scale generative AI fashions.
Present methods similar to OpenAI’s ChatGPT-4 and Google’s Gemini 2.5 demand not solely excessive reminiscence bandwidth but in addition substantial reminiscence capability, driving corporations like Microsoft and Google to buy lots of of hundreds of NVIDIA GPUs.
The reminiscence bottleneck problem
The core innovation lies within the staff’s method to fixing reminiscence bottleneck points that plague present AI infrastructure. Their energy-efficient NPU expertise focuses on “light-weight” the inference course of whereas minimising accuracy loss—a vital stability that has confirmed difficult for earlier options.
PhD scholar Minsu Kim and Dr Seongmin Hong from HyperAccel Inc., serving as co-first authors, offered their findings on the 2025 Worldwide Symposium on Pc Structure (ISCA 2025) in Tokyo. The analysis paper, titled “Oaken: Quick and Environment friendly LLM Serving with On-line-Offline Hybrid KV Cache Quantization,” particulars their complete method to the issue.
The expertise centres on KV cache quantisation, which the researchers establish as accounting for most reminiscence utilization in generative AI methods. By optimising this part, the staff permits the identical stage of AI infrastructure efficiency utilizing fewer NPU units in comparison with conventional GPU-based methods.
Technical innovation and structure
The KAIST staff’s energy-efficient NPU expertise employs a three-pronged quantisation algorithm: threshold-based online-offline hybrid quantisation, group-shift quantisation, and fused dense-and-sparse encoding. This method permits the system to combine with present reminiscence interfaces with out requiring adjustments to operational logic in present NPU architectures.
The {hardware} structure incorporates page-level reminiscence administration methods for environment friendly utilisation of restricted reminiscence bandwidth and capability. Moreover, the staff launched new encoding methods particularly optimised for quantised KV cache, addressing the distinctive necessities of their method.
“This analysis, by joint work with HyperAccel Inc., discovered an answer in generative AI inference light-weighting algorithms and succeeded in growing a core NPU expertise that may remedy the reminiscence drawback,” Professor Park defined.
“By means of this expertise, we applied an NPU with over 60% improved efficiency in comparison with the most recent GPUs by combining quantisation methods that scale back reminiscence necessities whereas sustaining inference accuracy.”
Sustainability implications
The environmental affect of AI infrastructure has turn out to be a rising concern as generative AI adoption accelerates. The energy-efficient NPU expertise developed by KAIST presents a possible path towards extra sustainable AI operations.
With 44% decrease energy consumption in comparison with present GPU options, widespread adoption might considerably scale back the carbon footprint of AI cloud providers. Nonetheless, the expertise’s real-world affect will depend upon a number of components, together with manufacturing scalability, cost-effectiveness, and business adoption charges.
The researchers acknowledge that their resolution represents a major step ahead, however widespread implementation would require continued growth and business collaboration.
Trade context and future outlook
The timing of this energy-efficient NPU expertise breakthrough is especially related as AI corporations face growing stress to stability efficiency with sustainability. The present GPU-dominated market has created provide chain constraints and elevated prices, making various options more and more enticing.
Professor Park famous that the expertise “has demonstrated the opportunity of implementing high-performance, low-power infrastructure specialised for generative AI, and is predicted to play a key function not solely in AI cloud information centres but in addition within the AI transformation (AX) atmosphere represented by dynamic, executable AI similar to agentic AI.”
The analysis represents a major step towards extra sustainable AI infrastructure, however its final affect might be decided by how successfully it may be scaled and deployed in industrial environments. Because the AI business continues to grapple with vitality consumption issues, improvements like KAIST’s energy-efficient NPU expertise supply hope for a extra sustainable future in synthetic intelligence computing.
(Picture by Korea Superior Institute of Science and Expertise)
See additionally: The 6 practices that guarantee extra sustainable information centre operations
Need to be taught extra about cybersecurity and the cloud from business leaders? Take a look at Cyber Safety & Cloud Expo going down in Amsterdam, California, and London.
Discover different upcoming enterprise expertise occasions and webinars powered by TechForge right here.