loading
loading
Recently, I was in Seoul, South Korea, for the ASUS AI Tech 2026 event. This is ASUS’s event where it showcases its new AI servers, racks, and offerings, along with its suppliers, partners, and customers. At the event, I moderated a panel discussion, so I thought it was worth an article and a short video. Of course, ASUS paid for my travel, so we have to say this is sponsored. ASUS AI Tech 2026 Seoul, South Korea In the lobby, there were all kinds of AI servers. That included NVIDIA HGX Rubin NVL8 solutions with eight Rubin GPUs. ASUS showed off a number of these, including an awesome-looking liquid-cooled version that is a massive change from Ampere, Hopper, and Blackwell generation designs. This generation flips the GPU orientation on the baseboard to enable more efficient liquid cooling. NVIDIA also brings a new generation of NVLink Switch onboard, with high-speed scale-out NVIDIA networking to enable larger clusters. This design also lets the eight GPUs fit in 1U instead of taking 6U or more to air-cool the HGX baseboard, making them much denser. ASUS uses this plus another chassis segment for the CPUs, SSDs, DPUs, and more to fit a huge amount of compute into much denser rack footprints than we have seen previously. The ability to use either hybrid cooling or all-liquid cooling in these nodes helps them fit into data centers with different capabilities. While the individual GPUs may use more power than a Hopper-generation GPU, for example, the higher-performance Rubin GPUs mean that the onboard NVL8 topology has more compute resources and memory connected, minimizing data movement. In turn, that means that not only is each GPU faster, but the NVL8 domain is significantly faster and can hold more data, so that way more communication can stay within the node instead of having to go out to the scale-out fabric. ASUS also brought trays from its NVIDIA GB200/GB300 NVL72 and Vera Rubin NVL72 solutions that it calls the ASUS AI Pod. It was neat seeing the different liquid-cooled ORv3 trays on the floor. The all-liquid-cooled design increases efficiency, which ultimately determines how much performance you can get from a data center. That was a key theme among the day’s speakers. You could also see the generational shift toward moving fans out of the chassis and liquid-cooling everything in the rack in the newer Vera Rubin NVL72 design. That helps increase compute density and maximize performance within a given power footprint. We previously showed you how they are tested in our ASUS Thermal Lab Tour 2026 Testing AI Servers. If you want to check that out, you can find the short video for that one here: Next, let us get to the storage and our panel discussion.
document fetched 2026-09-28 02:54 · canonical https://servethehome.com/asus-ai-tech-2026-seoul-south-korea