Geopolitics overshadows VMware Explore in Barcelona, ​​AI in focus

Geopolitics overshadows VMware Explore in Barcelona, ​​AI in focus

At its annual European in-house exhibition, VMware Explore Barcelona, ​​the approximately 9,000 participants who came to the site were actually expecting news about Broadcom’s planned takeover of VMware. Broadcom’s most expensive takeover of another company to date, worth $61 billion, should have been completed at the end of October. Since Broadcom generates a significant portion of its sales in China, the Chinese government must also approve the takeover. However, due to the US government’s trade sanctions in the chip sector, it has still not approved the takeover deal and the likelihood of this happening is decreasing every day. VMware has suddenly become a pawn in current geopolitical upheavals, which is causing considerable uncertainty for VMware’s employees, partners and customers. According to analyst firm Forrester, around 20% of VMware customers are already considering turning their backs on the manufacturer in the near future.

Advertisement

AI dominates the program

On the technical side, the topic of artificial intelligence was, unsurprisingly, also in the foreground at the European edition of the in-house exhibition. VMware had already presented its “Private AI” strategy at its US VMware Explore at the end of August and has now further fleshed it out and expanded it with additional partnerships. With Private AI, the manufacturer is certainly addressing an urgent need for many customers who want to use generative artificial intelligence but are reluctant to use these services in the public cloud due to data protection and compliance concerns.

In addition to the announcements from three months ago, the manufacturer has now presented further collaborations that offer customers more freedom of choice when implementing a “private AI” environment. VMware would like to transfer its “Cloud Smart” approach to the AI ​​area. The plan is to establish itself as a platform with which customers can create their own desired environment from the wide field of AI frameworks with little effort and at the same time retain full control over their data. In the keynote, VMware presented a common reference architecture with Intel (VMware Private AI with Intel). With this, customers should be able to build their AI environment operated in their own data center according to best practices from both manufacturers – without having to use GPUs (“AI without GPUs”).

KI with Intel-Hardware

This reference architecture is based on Intel’s 4th generation Xeon processors with Advanced Matrix Extensions (AMX) AI acceleration instructions. According to Intel, they offer a tenfold performance improvement for common AI frameworks and make CPUs a GPU alternative to GPUs for certain AI tasks. In a demo, the manufacturers showed a LLAMA-2 model of Hugging Face with seven billion parameters, which they fine-tuned with data from the financial sector. In the demo, three chatbot inference sessions could run simultaneously on a single, latest-generation Xeon processor – without the use of a dedicated graphics card. Also part of the reference architecture presented is Intel’s open source framework One API, which has now been housed at the Linux Foundation under the project name Unified Acceleration Foundation (UXL). VMware also joined the UXL Foundation steering committee at the same time as the conference.

The architecture of VMware Private AI with Intel AMX at a glance.

(Image: VMware)

Intel’s goal here is clear: they want to provide an alternative to Nvidia’s proprietary and market-dominant CUDA platform. Support for Intel’s “Data Center GPU Max” GPUs will also follow shortly and offer customers more freedom of choice when selecting hardware. On the server side, manufacturers Dell Technologies, Hewlett Packard Enterprise and Lenovo will support VMware Private AI with Intel based on their announced server models with 4th generation Intel Xeons with Advanced Matrix Extensions (Intel® AMX) and Intel Max Series GPUs. VMware provides further technical information as well as configuration recommendations and benchmarks on the joint initiative with Intel for AI without GPUs here.

The next announcement from the keynote is aimed at companies that want to use IBM’s AI and data platform watsonx in a controlled on-premises environment or in the hybrid cloud managed by VMware for training and fine-tuning AI models. As part of a partnership with IBM, VMware wants to enable its customers to use watsonx AI functions to operate machine learning (MLOps) based on its Cloud Foundation software suite and Red Hat’s OpenShift. In addition, customers receive access to Hugging Face open source models selected and curated by IBM and other models in order to be able to operate their applications themselves with generative artificial intelligence (GenAI).

Bequemeres Management per VCF

On the part of VMware, the basis for all of these applications is the in-house software suite VMware Cloud Foundation (VCF), which the manufacturer has now presented in version 5.1. VCF is a bundle of VMware vSphere (currently in version 8.0 Update 2/2a), the hyperconverged storage service vSAN, VMware’s network virtualization NSX in the current version 4.1 and central management via the SDDC Manager. The SDDC Manager and other tools simplify installation, operation and maintenance of the closely interlinked products. New in VCF 5.1 are the following detailed improvements: Administrators can manage ESXi hypervisor upgrades centrally from the VMware Cloud Console. The vSAN Express Storage Architecture (vSAN ESA), which has been new since vSphere Version 8.0, is said to be significantly faster, VMware speaks of a factor of 4. AI and ML workloads can also be optimized with a combination of up to 16 GPUs per VM.

Some announcements on the sidelines of the event were also interesting. Since version 3.2, VMware NSX has built-in intrusion detection and prevention (IDS/IPS) functions, which in turn are based on the NSX Distributed Firewall implemented in the hypervisor kernel. However, SOC analysts must sift through the incidents raised by the system and trigger countermeasures. With Project Cypress, which was newly introduced at the trade fair, VMware wants to help SOC analysts with GenAI capabilities to classify and handle security incidents more quickly.

(avr)

To the home page
#Geopolitics #overshadows #VMware #Explore #Barcelona #focus