The increasing demand for artificial intelligence hardware directly affects not only graphics cards but also memory and storage. The products shown at the SK Hynix booth at Computex 2026 are positioned especially for artificial intelligence servers, accelerator systems, data center infrastructures and professional systems requiring high bandwidth. Among the solutions exhibited by SK Hynix, CMM-DDR5, SOCAMM2 and HBM3E memories used on NVIDIA GB300 stand out.
High-speed memory solutions located close to the processor and accelerators are becoming increasingly critical for the operation of artificial intelligence models. Memory Demand in Artificial Intelligence Hardware is Increasing. One of the companies that stands out in the artificial intelligence era is SK Hynix. As the demand for graphics cards increases, the price side of memory, storage and especially high-capacity memory solutions also increases.
The main reason for this situation is that powerful GPUs alone are not enough to run artificial intelligence models. In order to process data quickly in large models, memory solutions with high capacity, low latency and located close to the processing units are required. The products shown at SK Hynix’s stand also focus on the infrastructure side for this need. CMM-DDR5 and SOCAMM2 Solutions ExhibitedCMM-DDR5 second generation module and 192 GB SOCAMM2 memory solutions were also exhibited at Computex 2026.
On the CMM-DDR5 side, 256 GB capacity, CXL 3.2 support, PCIe Gen5 connection and DDR5 6400 Mbps speed are highlighted. On the SOCMM2 side, the 192 GB capacity and LPDDR5X-based structure attracts attention. This form factor is positioned for lower power consumption and more compact structure. This structure is especially important for systems that require intense memory capacity in artificial intelligence infrastructures.
SK Hynix HBM3E is Used on NVIDIA GB300. HBM3E memories have 24 Gb capacity, 9.6 Gbps speed, x1024 I/O structure and bandwidth over 1.2 TB/s. On the HBM3E side, 36 GB capacity and high bandwidth play a critical role in artificial intelligence operations in terms of fast access to memory and parallel data transfer. When it comes to artificial intelligence factories (AI Factory), the scalability of artificial intelligence infrastructures is important.
The purpose of these systems is to increase token generation performance by providing fast access to memory and to create more efficient artificial intelligence infrastructures. The solutions demonstrated by SK Hynix here target data centers, artificial intelligence servers and large-scale accelerator platforms rather than direct end-user systems. For this reason, the focus of the products is on capacity, bandwidth, power efficiency and memory placement close to the processing units.
Comments
You can write your views about this story. Comments may be moderated according to site settings.