VAST Data is the first storage provider to be integrated with Cisco’s Nexus 9000 Ethernet switches in the networking giant’s Nexus HyperFabric for Ethernet AI stacks connected to Nvidia GPU farms.
Cisco has developed programmable and high-bandwidth Silicon One ASIC chips for its fastest network switches and routers. The June 2023 G200 product, for example, with 512 x 100GE Ethernet ports on one device, supports 51.2 Tbps and features advanced congestion management, packet-spraying techniques, and link failover. The Silicon One hardware is available as chips or chassis without Cisco software. Cisco has developed its Ethernet AI Fabric, using Silicon One, which is deployed at three hyperscale customers.
Nexus HyperFabric AI clusters, using Nexus switches, are designed to help enterprises build AI datacenters using Nvidia-accelerated computing, which includes Tensor Core GPUs, BlueField-3 DPUs, SuperNICs, and AI Enterprise software, through Cisco networking and the VAST Data Platform. This VAST system also supports BlueField-3 DPUs. VAST has certified Cisco Nexus Ethernet-based switches with its storage, delivering validated designs. Cisco customers can monitor and correlate storage performance and latency using VAST’s APIs, pulling both network and storage telemetry back to the Nexus HyperFabric.
Renen Hallak, CEO and co-founder of VAST Data, said: “This is the year of the enterprise for AI. Traditionally, enterprises have been slower to adopt new technologies because of the difficulty of integrating new systems into existing systems and processes. This collaboration with Cisco and Nvidia makes it simple for enterprises to implement AI as they move from proof of concept to production.”
Jonathan Davidson, Cisco EVP and GM for its networking, added: “With an ecosystem approach that now includes VAST Data and Nvidia, Cisco helps our enterprise customers tackle their most difficult and complex networking, data and security challenges to build AI infrastructures at any scale with confidence.”
The Cisco Nexus HyperFabric features:
- Cisco cloud management capabilities to simplify IT operations across all phases of the workflow.
- Cisco Nexus 9000 series switches for spine and leaf that deliver 400G and 800G Ethernet fabric performance.
- Cisco Optics family of QSFP-DD modules to offer customer choice and deliver super high densities.
- NVIDIA AI Enterprise software to streamline the development and deployment of production-grade generative AI workloads
- NVIDIA NIM inference microservices that accelerate the deployment of foundation models while ensuring data security, and are available with NVIDIA AI Enterprise
- NVIDIA Tensor Core GPUs starting with the NVIDIA H200 NVL, designed to supercharge generative AI workloads with performance and memory capabilities.
- NVIDIA BlueField-3 data processing unit DPU processor and BlueField-3 SuperNIC for accelerating AI compute networking, data access and security workloads.
- Enterprise reference design for AI built on NVIDIA MGX, a modular and flexible server architecture.
- The VAST Data Platform, which offers unified storage, database and a data-driven function engine built for AI.
The BlueField-3 DPUs can also run security services like the Cisco Hypershield, which enables an AI-native, hyperdistributed security architecture, where security shifts closer to the workloads needing protection.
VAST and Cisco claim they are simplifying network management and operations across all infrastructure endpoints. Congestion management with flow control algorithms, and visibility with real-time telemetry are provided by the private cloud (on-premises) managed Cisco Nexus Dashboard.
Select customers can access the VAST Data Platform with Cisco and Nvidia in beta now, with general availability expected in calendar Q4 of 2024.
Comment
Cisco has developed a FlashStack AI Cisco Verified Design (CVD) with Pure Storage for AI inferencing workloads using Nvidia GPUs. A similar FlexPod AI CVD does the same for NetApp. In February, Cisco said more Nvidia-backed CVDs will be coming in the future, and it looks like the VAST Data example has just arrived.
We understand that HyperFabric will support AI training workloads. This leads us to understand that a higher-speed Silicon One chip is coming, perhaps offering 76.5 Tbps or even 102.4 Tbps.
In effect, it looks like Cisco has developed a VAST Stack AI type offering which could potentially pump data faster to Nvidia GPU farms than existing arrangements with NetApp and Pure Storage.
