More performance in less space: Dell announces new servers for AI workloads

Dell Technologies has announced new servers and storage systems for AI workloads. Customers will also be able to use them for conventional workloads such as databases, systems of records (ERP and CRM), edge applications and containerized tasks. Dell refers to this as a so-called disaggregated infrastructure, which combines computing power, storage, and networking into shared and customizable resource pools for all types of workloads. In addition, this infrastructure should allow for great scalability and offer higher availability.

New servers require up to 80 percent less space

The US manufacturer's portfolio includes the four PowerEdge systems, R470, R570, R670 and R770. They are equipped with Intel Xeon 6 P-Core processors. The R470 and R570 models are single-socket systems, while the R670 and R770 are dual-socket systems. Compared to previous models, the new systems require less space, with the R770 saving up to 80 percent per 42U rack. Dell also promises 3.35 times the performance with 2.57 times the number of cores. Furthermore, the new servers are based on the modular DC-MHS architecture, a standard server design of the Open Compute Project, which is intended to facilitate integration into existing infrastructures.

In the storage sector, Dell announced a new generation of its ObjectScale platform. This includes the X560, which, according to the company, can store exabytes of object data and achieve eight times the density of previous generations. Furthermore, the XF960 is designed for AI workloads and is said to offer twice the throughput per node compared to other systems. The new PowerScale Series scale-out storage includes 122 terabytes of SSDs in a 2U node configuration, doubling all-flash capacity.

Dell PowerScale also has updated archive and hybrid nodes in the H710, H7100, A310 and A3100 models. They are designed to improve latency and performance through a revised computing module for HDD-based platforms. The PowerStore systems now use the company's own AIOps software for AI-supported analyses and stronger zero-trust security. In addition, the backup portfolio of the PowerProtect series now includes the DD6410 system with a storage capacity of between 12 and 256 terabytes, which is said to provide 91 percent faster recovery. Dell promises around 61 percent faster backups with its All-Flash Ready Node with a capacity of up to 220 terabytes.

Due to VMware costs: Backups for Hyper-V and OpenShift

Dell has developed a hybrid cloud platform based on ObjectScale for the operation of AI data lakes. It contains a range of new functions, including multi-site federation, copy-to-cloud, geo-replication, global namespaces and data management. The hardware manufacturer has also collaborated with cloud provider Wasabi. Dell also responded to the current changes in the area of virtualization. “Many customers have clarified that they are looking for alternatives to VMware in light of recent events,” said David Noy, Vice President of Product Management, referring to the price increases of VMware products. As a result, Microsoft Hyper-V and Red Hat OpenShift can be archived on Dell ObjectScale.

Dell is not alone in its push for hardware for AI workloads. Nvidia CEO Jensen Huang also sees a high demand. “There are a trillion dollars worth of data centers around the world, but they were all built for the previous generation of IT and now need to be fully converted to the new way of using them,” he said at a Dell event last year. As a result, all OEMs are pressing ahead with the corresponding developments. At its Partner Summit last fall, for example, Cisco presented special AI offerings based on its Universal Computing System and a validated design.

In February, Cisco then announced an agreement with Nvidia to jointly offer a wide range of AI network options. Also in February, Hewlett Packard Enterprise introduced new versions of its Gen12 ProLiant servers with Intel Xeon 6 processors. And while all OEMs praise the intensive and cooperative partnership with Nvidia, a new competitor is looming from the GPU manufacturer: at the recent GPU Technology Conference, the company presented an exaflop-capable rack server system based on the NVL72 GB200 system that can process compute-intensive workloads such as AI applications with high performance.

(dmk)