SC25: MiTAC Computing Unveils Advanced AI Cluster and Cooling Solutions

MiTAC Computing Technology announced its participation at the Supercomputing 2025 (SC25), November 18–20, St. Louis, MO, booth 3916.

Under the theme AI Cluster Power – Cool Fast Scale Faster, the company will showcase its modular highly scalable rack infrastructure for demanding AI and HPC workloads demonstrating capabilities from single servers to cluster integration focusing on liquid cooling and energy-efficient designs. MiTAC collaborates with AMD, Broadcom, CoolIT, Intel, Kioxia, Micron, NVIDIA, Samsung and Solidigm to accelerate advanced computing and high-efficiency data centers.

AI Cluster Power. Cool Fast. Scale Faster:

• Rack-scale Innovation From Standard Architecture to AI Cluster Excellence
  At SC 2025, the firm presents its full range of rack-level solutions for cluster-scale deployments including liquid-cooled and air-cooled AI and HPC racks addressing both open architectures and traditional enterprise data centers.

48U EIA AI liquid-cooled rack MR1100 series

• AI Liquid-Cooled Rack with AMD Instinct MI355X GPUs | Optimized for Ultra-Scale AI
  The high-density 48U EIA AI liquid-cooled rack MR1100 series targets hyperscale AI training and inference. This robust solution supports 64 to 256 AI GPUs. Utilizing cold-plate technology with the latest AMD Instinct MI355X GPUs, AMD EPYC 9005 CPUs, and AMD Pensando Pollara 400 AI NICs, it ensures non-throttled AI throughput with 400/800 Gb/s network architecture.
• AI Air-Cooled Rack with AMD Instinct MI350X GPUs | Standardized Architecture with High-Speed Interconnect
  The company presents its standard EIA 45U air-cooled AI rack, MR1100A, with 4 MiTAC G8825Z5 systems using AMD Instinct MI350X/MI325X GPUs. The 800Gb/s network switch Broadcom Tomahawk 5 chipset ensures low-latency data transmission GC68C-B8056 management and TS70A-B8056 storage servers enable rapid deployment of scalable AI/HPC clusters.
• OCP ORv3 Liquid-Cooled Rack | Modular Power and Advanced Thermal Management
  The firm‘s 43OU OCP ORv3 liquid-cooled rack, MR is designed for sustainable HPC, accommodating up to 14 C2811Z5 multi-node servers powered by AMD EPYC 9005 Series processors. It integrates MiTAC Lake Erie storage, a 33kW Power Shelf, and the CoolIT 200kW CHx200+ In-Rack CDU. This modular design ensures stable, energy-efficient operation for high-density compute servers.

AI Acceleration Platforms:

• G4527G6: This 4U AI Accelerator, built on the NVIDIA MGX architecture, hosts up to 8 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. It is driven by dual Intel Xeon 6767P CPUs with Solidigm D7-P5520 SSDs, providing the speed essential for deep learning.

G4826Z5 appliance

• G4826Z5: A liquid-cooled compute node for MiTAC AI clusters, supporting up to eight AMD Instinct MI355X GPUs. It features full liquid cooling and includes the Broadcom P2200G network adapter.

• G8825Z5: A High-performance AI platform supporting up up to 8 AMD Instinct MI325X or MI350X GPUs. It also features AMD Pensando Pollara 400 AI NICs and Micron 6550 ION NVMe SSDs vital for accelerating training cycles.

HPC and Cloud Computing Frameworks:

• G4520G6: This highly adaptable server, powered by the latest Intel Xeon 6 CPUs, delivers superior performance per watt for cloud and HPC roles, leveraging Micron DDR5 DRAM. It also features extensive expansion options, including support for NVIDIA RTX PRO 6000 Blackwell Server Edition and NVIDIA Hopper GPUs.
• C2811Z5: An OCP-compliant liquid-cooled multi-node server, it uses AMD EPYC 9005 series processors. Optimized with NVMe E1.S interfaces and Micron 9550 NVMe SSDs, it guarantees sustained high-speed data delivery for intensive HPC.

Enterprise Data Solutions:

• M2810Z5: Optimized for peak storage performance by deploying Kioxia XD8 E1.S Gen5 SSDs, this server unlocks maximum bandwidth for IO-intensive database applications.
• R2520G6: A high-capacity 2U server with dual Intel Xeon 6700P series processors integrates Solidigm D7-PS1010 SSDs. PCIe 5.0 bandwidth guarantees stability for data warehousing and analytics.
• M2510G6: This high-reliability platform supports Samsung MZTL67T6HBLC-00AW7 SSDs for mission-critical enterprise settings and demanding data processing.
• R2513G6: This platform provides maximum archival capacity using Seagate EXOS M 30TB HDDs, tailored for organizations with immense data retention needs.

Success Stories Demonstrating MiTAC’s Value in AI and HPC Deployments Worldwide
At SC 2025, the company showcases real-world success cases highlighting its proven expertise in delivering comprehensive rack-level solutions and its commitment to cluster-scale integration.

• MiTAC and French CSP pioneer sustainable HPC achieving world-class PUE and operational cost reduction: Partnered with Qarnot using the Capri 3 OCP server to capture 95% of server heat for reuse, achieving a PUE of 1.01 and reducing operational costs by 50% for clients in Europe. The exhibit also highlights the integration of the Broadcom N1400GD network adapter.

• OCP deployment accelerated in US data centers featuring ORv3 platform and resource efficiency: MiTAC partnered with CTCA, a global IT solutions provider, to accelerate OCP deployment across the US data centers. This success leveraged the ORv3-compliant OCP server and advanced rack integration. The SC showcase highlights the OCP server featuring Samsung M321R8GA0EB2-CCP Memory, emphasizing faster rollouts and resource efficiency.

• Rack integration streamlines global operations for leading cloud security provider: MiTAC provided customized, ready-to-deploy rack integration services, utilizing the GC68C-B8056 server. This service streamlined operations by delivering over 380 configurations across 150 data centers worldwide, meeting 48-hour turnaround times, and enabling a GPU-first infrastructure shift within half a year, demonstrating the consistency, agility, and innovation MiTAC provides.

• MiTAC and its SDS partner maximize GPU performance by eliminating AI training bottlenecks: Collaborating with a software-defined storage (SDS) partner, MiTAC leveraged the GC68A-B8056 server to remove critical I/O constraints. The pre-validated architecture boosted GPU utilization by 35% and delivered over 200 GB/s throughput, effectively tripling the speed of AI training completion.

Resource:
MiTAC Computing SC 2025 event site