[Edaily Reporter Kim Seung-kwon ] JLK ( JLK, Inc.(322510)), a company specializing in medical artificial intelligence (AI), announced on the 21st that it has secured a low-cost operating environment for its proprietary medical multimodal large language model (LLM) platform through technical validation with Intel. The company’s strategy is to accelerate the expansion of its B2B business targeting medical institutions and corporate clients by reducing dependence on expensive AI servers while demonstrating commercial-grade performance.

JLK, Inc. explained that after validating its medical multimodal LLM platform, “JOOMED,” in an Intel GPU-based environment, it confirmed high operational stability in a “Intel LLM Scaler” environment equipped with two Intel Arc Pro B70 GPUs. According to the company, it also achieved comparable analytical performance in a comparative evaluation against an NVIDIA A100 80GB PCIe-based AI server.

To optimize the JOOMED platform for Intel GPUs, the company conducted a comparative validation of Intel and NVIDIA environments using the same model and parameters. It explained that it evaluated the performance of integrated analysis of medical images and clinical data under identical conditions to analyze performance and cost-efficiency by hardware type.

The validation utilized “K-MMLU-Pro,” the Korean language LLM performance evaluation standard, and 100 medical multimodal data sets developed in-house by JLK, Inc. The company stated that it configured the validation environment to ensure only hardware performance could be compared by applying the same model architecture, dataset, and evaluation criteria.

The evaluation results showed that both systems met the analytical accuracy required for the commercialization of medical AI. While NVIDIA GPUs demonstrated superior single-compute performance, the Intel GPU-based system proved competitive in terms of cost-performance ratio, which factors in total cost of ownership (TCO), the company reported.

JLK, Inc. explained that through this validation, it has secured AI infrastructure options that consider both performance and cost-efficiency. The company expects that by offering customized infrastructure with varying GPU configurations based on healthcare institutions’ budgets and IT environments, it will be able to lower the barriers to adopting medical AI.

Building on these results, the company plans to diversify the infrastructure for its JOOMED service and fully expand its B2B business targeting medical institutions and corporate clients. Its strategy is to reduce deployment costs and operational burdens by providing a variety of AI semiconductor options tailored to each customer’s environment.

A JLK, Inc. official stated, “Through Intel’s infrastructure, we have secured an environment where we can operate our proprietary medical LLM stably and cost-effectively,” adding, “We will continue to expand the scope of JOOMED’s application by providing optimized AI infrastructure even to healthcare institutions with budget constraints.”