Thursday, November 30, 2023

Broadcom’s next Trident switching silicon doubles capacity, adds neural engine

 Broadcom announced an on-chip, neural-network inference engine called NetGNT (Networking General-purpose Neural-network Traffic-analyzer) in its new, software-programmable Trident 5-X12 chip for top-of-rack (ToR) data center switches.

The new Trident 5-X12 chip provides 16.0 Tbps of bandwidth, which is double that of the Broadcom’s current Trident 4-X9. It also adds support for 800G ports, allowing direct connection to Broadcom’s Tomahawk 5, which is used as the spine/fabric in the newest compute and AI/ML data centers. 

Broadcom says its new chip is uniquely positioned to enable a 1RU data center ToR supporting 48x200G downlink ports and 8x800G uplink ports.

NetGNT works in parallel to augment the standard packet-processing pipeline. The standard pipeline is one-packet/one-path, meaning that it looks at one packet as it takes a specific path through the chip’s ports and buffers. NetGNT, in contrast, is an ML inference engine and can be trained to look for different types of traffic patterns that span the entire chip.

For example, NetGNT can look for a traffic pattern common in AI/ML workloads called “incast” in which many packet flows all converge on the same port and buffer at roughly the same time causing congestion. NetGNT can recognize this pattern in real-time as it begins and invoke congestion-control techniques to avoid degraded network performance. And since NetGNT runs in hardware at full line rate— there is no impact to throughput or latency.

“We continue to push the envelope and introduce brand-new technologies such as NetGNT to the market. We also listen very closely to our customers who have made it clear that one size does not fit all. They rely on us to deliver a broad portfolio of chips, customized for different applications,” said Ram Velaga, senior vice president and general manager, Core Switching Group, Broadcom. “Trident 5-X12 is the most power-efficient ToR (Top of Rack) on the market, while still adding cutting-edge new features that our customers have come to expect from an innovator such as Broadcom.”

Trident 5-X12 highlights

  • Uses 25% less power per 400G port than current market-leading Trident 4-X9
  • Adds support for 800G ports using Broadcom’s industry-leading 100G-PAM4 SerDes, enabling up to 4m DAC and Linear Optics
  • Supports the transition to next-generation CPU & GPU servers using 400G NICs
  • Adds NetGNT to enable new capabilities along with improved network efficiency and performance
  • Adds enhanced telemetry capabilities allowing deeper real-time insights into network operations, which can then be used to train NetGNT
  • Increases forwarding/policy scales to support ever-expanding network sizes
  • Maintains legacy API compatibility with current Broadcom switch chips
  • Fully-programmable using NPL (Network Programming Language), allowing in-field upgrades to add features such as new telemetry, security and traffic engineering
  • NPL-compatible with current Trident 4 family, which is already in use by a large number of customers to develop their own customized switch applications
  • Support for Enterprise SONiC and SAI promote rapid and seamless integration into data center operations frameworks

HPE targets AI-native architecture and hybrid cloud

Hewlett Packard Enterprise  introduced its next series of AI-native and hybrid cloud offerings for machine learning development, data analytics, AI-optimized file storage, AI tuning and inferencing and professional services. 

The solutions are based on an open, full-stack AI-native architecture and leverage HPE’s expertise in hybrid cloud, supercomputing and AI/ML software.

“With the emergence of GenAI, enterprises are quickly realizing that the data and computational demands to effectively run AI models require a fundamentally different approach to technology,” said Antonio Neri, president and CEO, at HPE. “HPE will bring its market-leading hybrid cloud, supercomputing and AI capabilities more broadly to the enterprise to enable an AI-powered transformation, where customers can develop AI models securely with their proprietary data. 

Key elements of the announcement

  • Data-first pipeline to manage public and proprietary data across multi-gen IT
  • AI lifecycle management software to accelerate workflows for training, tuning and inferencing
  • Hybrid by design to run AI anywhere from edge to cloud with data protection
  • High-performance interconnects for intelligent connectivity and traffic management for large clusters
  • Supercomputing DNA built into the entire portfolio, sustainable by design, to train the largest models
  • Open ecosystem for freedom of choice with no lock-in

HPE teams with NVIDIA on Enterprise GenAI

Hewlett Packard Enterprise announced an expanded alliance with NVIDIA to build an enterprise computing solution for generative AI (GenAI). 

These solutions integrate HPE Machine Learning Development Environment Software, HPE Ezmeral Software, HPE ProLiant Compute and HPE Cray Supercomputers with the NVIDIA AI Enterprise software suite, including the NVIDIA NeMo framework. 

Hardware encompasses rack-scale  HPE ProLiant Compute DL380a pre-configured with NVIDIA L40S GPUs, NVIDIA BlueField-3 DPUs and the NVIDIA Spectrum-X Ethernet Networking Platform for hyperscale AI. The solution was sized to fine-tune a 70 billion-parameter Llama 2 model and includes 16 HPE ProLiant DL380a servers and 64 L40S GPUs.

“Together, HPE and NVIDIA are in a unique position to deliver a comprehensive AI-native solution that will dramatically ease the journey to develop and deploy AI models with a portfolio of pre-configured solutions,” said Antonio Neri, president and CEO, at HPE. “The strategic collaboration between HPE and NVIDIA will dramatically reduce barriers for customers looking to transform their businesses with AI.”

Marvell posts Q3 sales of $1.419 billion, down 8% yoy

Marvell Technology Q3 revenue of $1.419 billion, down by 8% year-over-year, and $19 million above the midpoint of the company's guidance provided on August 24, 2023. 

GAAP net loss for the third quarter of fiscal 2024 was $(164.3) million, or $(0.19) per diluted share. Non-GAAP net income for the third quarter of fiscal 2024 was $354.1 million, or $0.41 per diluted share. Cash flow from operations for the third quarter was $503.0 million.

"Marvell delivered third quarter fiscal 2024 revenue of $1.419 billion, growing 6% sequentially, above the midpoint of guidance. Revenue from our datacenter end market grew over 20% sequentially in the third quarter, and we expect growth of over 30% sequentially in our fourth quarter," said Matt Murphy, Marvell's Chairman and CEO. "The diversification of our portfolio is serving us well, with strong growth from AI and cloud carrying us through a softening demand environment in other end markets. These dynamics are reflected in our forecast for overall revenue to be flat sequentially in the fourth quarter at the midpoint of guidance."

Q3 2023 Revenue by Segment

  • Data center $555.8 million
  • Enterprise networking $271.1 million
  • Carrier infrastructure $316.5 million
  • Consumer $168.7 million
  • Automotive/industrial $106.5 million

Vertiv supplies liquid cooling for Intel Gaudi AI Accelerator

Vertiv is providing a liquid cooling solution for the new Intel Gaudi3 AI accelerator, scheduled to launch in 2024. 

The Intel Gaudi3 AI accelerator will enable both liquid-cooled and air-cooled servers, supported by Vertiv pumped two-phase (P2P) cooling infrastructure. 

The liquid cooled solution has been tested up to 160kW accelerator power using facility water from 17°C up to 45°C (62.6°F to 113°F). The air-cooled solution has been tested up to 40kW of heat load that can be deployed in warm ambient air data centers up to 35°C (95°F). This medium pressure direct P2P refrigerant-based cooling solution will help customers implement heat reuse, warm water cooling, free air cooling and reductions in power usage effectiveness (PUE), water usage effectiveness (WUE) and total cost of ownership (TCO).

“The Intel Gaudi3 AI accelerator provides the perfect solution for a Vertiv and Intel collaboration,” said John Niemann, SVP global thermal line of business at Vertiv. “Vertiv continues to expand our broad liquid cooling portfolio, resulting in our ability to support leaders of next generation AI technologies, like Intel. Vertiv helps customers accelerate the adoption of AI quickly and reliably, while also helping them to achieve sustainability goals.”

Aviz Networks raises $10m for its open networking software

Aviz Networks, a start-up based in San Jose, California, has added Cisco Investments to its latest funding expansion, joining existing investors, Moment Ventures, Accton and Wistron. 

This funding round, now reaching $10M, will accelerate customer momentum for Aviz’s multi-vendor, GenAI-based networking stack along with adoption of SONiC based open networking.

Aviz’s stack supports the fast-growing SONiC open-source network OS (NOS) as well as multi-NOS environments and leverages the power of large-language-models (LLMs).

“We are delighted to close this latest funding round that includes Accton, Cisco Investments, Moment Ventures, and Wistron. Our vision of delivering open, cloud & AI enabled networks for all enterprises and telecoms has rapidly taken shape as we partner with customers and our vendor ecosystem,” said Vishal Shukla, Co-Founder and CEO for Aviz.

Malaysia’s Axiata picks AWS as its primary cloud

Axiata Group, a leading Asian telco and digital group headquartered in Malaysia, has selected AWS as its primary cloud provider. 

Axiata will migrate a large array of its mission-critical applications, including more than 650 services across customer service, enterprise resource planning, and human resources, to AWS by the end of 2024. 

In addition, Axiata will move 80 machine learning (ML) applications to the AWS.

“AWS has been a key collaborator in our digital transformation journey, accelerating our time to market for new services, enabling us to better serve our 174 million customers across Asia,” said Anthony Rodrigo, group chief information officer at Axiata. 

Broadcom cuts over 1800 VMware positions

Following the completion of its acquisition of VMware last week, Broadcom filed papers to state employment offices regarding the elimination of approximately 2,000 positions.

This includes 1,267 positions in California, along with 600 additional staff in the states of Washington, Colorado and Georgia.