Dell PowerEdge M7725 and XE9712 servers, together with integrated Rack 7000 servers
Efficiency and innovation are fueled by open standards in the dynamic data center industry. By 2029, it is projected that businesses globally would have invested $422.55 billion on AI infrastructure, growing at a compound annual growth rate of 44% during the next five years. This work is being driven by the Open Compute Platform (OCP) community, which focuses on open-source hardware. Dell's Integrated Rack Scalable Systems (IRSS), which offer dependable, adaptable solutions designed to meet the specific needs of large-scale AI deployments, are an embodiment of these concepts.Integrated Rack Scalable Systems (IRSS) from Dell: A Perfect Fit for Large-Scale AI and Dense Computing
Introducing the new Integrated Rack 7000 (IR7000), PowerEdge M7725, and PowerEdge XE9712. These new solutions provide a seamless and powerful solution for contemporary data centers as AI workloads increase the need for technology that provides dense compute that is energy-efficient and will support future technological advancements. Dell's IRSS enables data center operations and complies with the Open Compute Project's rack specifications while offering enhanced scalability and energy efficiency.
Dell's Integrated Rack Scalable Systems (IRSS) Are Ideal for Dense Computing and Large-Scale AI
The 21-inch Open Rack Version 3 (Orv3) based Dell Integrated Rack 7000 (IR7000) rack architecture features liquid cooling for high TDP GPUs and CPUs and dense compute. Before ORV3, direct liquid cooling (DLC) manifold connections were proprietary and incompatible when trying to combine servers with several DLC OEMs in a single rack. Dell's next-generation, quick-connect modular infrastructure is the industry standard for training massive AI models.Important characteristics of the PowerEdge M7725, IR7000, and XE9712
Flexible power: AC power shelves with a current capacity of 33kW to 264kW per rack and future plans to deploy 480kW.Future-ready: 44OU and 50OU systems with higher, wider sleds based on OCP requirements will handle several generations of CPU architectures.
Efficient: Consumes and dissipates high power and thermal exhaust, while offering smooth power scaling and approximately 100% heat absorption.
Optimizes Front IO: Custom configurations may provide for flexible storage or I/O in a 21-inch box design.
Dense computing was considered in the design of the Dell PowerEdge M7725. The PowerEdge M7725 supports two AMD EPYC 5th Generation processors, which offer 37% more instructions per clock (IPC) for high performance computing (HPC) than the earlier generations. A single IR7000 can host up to 72 M7725 server nodes and 27,000 cores per rack. The M7725 offers improved serviceability with front IO, cable-free power, and DLC connectivity. The M7725 with IR7000 Rack has room neutral (almost 100%) heat collection capability by utilizing Dell's own rear door heat capture technology.
Scalable: Each 50OU rack can accommodate up to 72 nodes and 27,000 cores. AMD 5th Gen Processors have an IPC that is 37% greater for HPC applications.
High Speed IO: An additional OCP slot enables host administration connectivity, and two PCIe Gen5 x 16 IO slots per node offer the fastest data transfer rates (supporting Ethernet and InfiniBand options).
Power & Cooling: To effectively power and cool two 500W processors in a small form factor, use the IR7000 Rack Infrastructure.
An NVIDIA GB200 NVL72 solution is added to the Dell AI Factory with the Dell PowerEdge XE9712. Building on the success of the XE9680 with 8-way HGX GPUs, the XE9712 offers faster LLM performance with 72 GPUs running as a single rack. Dell's extensive deployment of super PODs with complete networking across racks will be supported by its turnkey rack-scale deployment services, supply chain, and logistics.
- 25 times the efficiency of the H100
- Thirty times quicker than the H100, real-time trillion parameter LLM inference
- The efficient heat control offered by liquid-cooled architecture enables higher performance and faster data processing.
0 Comments