Ampere targets 256-core CPU

Ampere Computing has released its annual update, showcasing the company's commitment to sustainable, power-efficient computing for the Cloud and AI sectors. The announcement included a collaboration with Qualcomm Technologies, Inc. to develop a joint AI inferencing solution. This solution will leverage Qualcomm's high-performance, low-power Qualcomm Cloud AI 100 inference solutions alongside Ampere's CPUs, aiming to deliver robust AI inferencing capabilities.

Renee James, CEO of Ampere, emphasized the growing importance of Ampere's silicon design, particularly in addressing the increasing power demands of AI. She highlighted that Ampere's approach has successfully combined low power consumption with high performance, challenging the traditional notion that low power equates to low performance. James pointed out that as AI continues to advance rapidly, the energy consumption of data centers becomes a critical issue, and Ampere's solutions are designed to address this by enhancing efficiency and performance without compromising sustainability.

Jeff Wittich, Chief Product Officer at Ampere, outlined the company's vision for "AI Compute," which integrates traditional cloud-native capabilities with AI. He noted that Ampere CPUs are versatile enough to handle a range of workloads, from data processing and web serving to media delivery and AI applications. Both James and Wittich introduced the upcoming AmpereOne® platform, featuring a 12-channel 256-core CPU built on the N3 process node, which promises to significantly boost performance.

Key Highlights from Ampere's Update:

• Collaboration with Qualcomm Technologies to develop a joint AI inferencing solution featuring Ampere CPUs and Qualcomm Cloud AI 100 Ultra.
• Expansion of Ampere's 12-channel platform with a 256-core AmpereOne® CPU, delivering over 40% more performance than current market CPUs without requiring exotic cooling solutions.
• Meta's Llama 3 is now running on Ampere CPUs at Oracle Cloud, showing comparable performance to an Nvidia A10 GPU with a third of the power consumption.
• Formation of a UCIe working group as part of the AI Platform Alliance to enhance CPU flexibility through open interface technology.
• Detailed performance updates on AmpereOne®, which outpaces AMD Genoa by 50% and Bergamo by 15%, offering significant performance per watt advantages.
• AmpereOne® platform provides up to 34% more performance per rack, aiding data centers in refreshing and consolidating infrastructure to save space, budget, and power.
• Continued emphasis on sustainable data center infrastructure with solutions for retrofitting existing air-cooled environments and building new, environmentally sustainable data centers.
• Ampere CPUs capable of running a wide range of workloads, from cloud-native applications to AI, integrating seamlessly with traditional applications.
• Ampere's ongoing commitment to pioneering the efficiency frontier in computing, delivering high performance within an efficient computing envelope.

Ampere Computing has released its annual update, showcasing the company's commitment to sustainable, power-efficient computing for the Cloud and AI sectors. The announcement included a collaboration with Qualcomm Technologies, Inc. to develop a joint AI inferencing solution. This solution will leverage Qualcomm's high-performance, low-power Qualcomm Cloud AI 100 inference solutions alongside Ampere's CPUs, aiming to deliver robust AI inferencing capabilities.

Renee James, CEO of Ampere, emphasized the growing importance of Ampere's silicon design, particularly in addressing the increasing power demands of AI. She highlighted that Ampere's approach has successfully combined low power consumption with high performance, challenging the traditional notion that low power equates to low performance. James pointed out that as AI continues to advance rapidly, the energy consumption of data centers becomes a critical issue, and Ampere's solutions are designed to address this by enhancing efficiency and performance without compromising sustainability.

Jeff Wittich, Chief Product Officer at Ampere, outlined the company's vision for "AI Compute," which integrates traditional cloud-native capabilities with AI. He noted that Ampere CPUs are versatile enough to handle a range of workloads, from data processing and web serving to media delivery and AI applications. Both James and Wittich introduced the upcoming AmpereOne® platform, featuring a 12-channel 256-core CPU built on the N3 process node, which promises to significantly boost performance.

Key Highlights from Ampere's Update:

• Collaboration with Qualcomm Technologies to develop a joint AI inferencing solution featuring Ampere CPUs and Qualcomm Cloud AI 100 Ultra.
• Expansion of Ampere's 12-channel platform with a 256-core AmpereOne® CPU, delivering over 40% more performance than current market CPUs without requiring exotic cooling solutions.
• Meta's Llama 3 is now running on Ampere CPUs at Oracle Cloud, showing comparable performance to an Nvidia A10 GPU with a third of the power consumption.
• Formation of a UCIe working group as part of the AI Platform Alliance to enhance CPU flexibility through open interface technology.
• Detailed performance updates on AmpereOne®, which outpaces AMD Genoa by 50% and Bergamo by 15%, offering significant performance per watt advantages.
• AmpereOne® platform provides up to 34% more performance per rack, aiding data centers in refreshing and consolidating infrastructure to save space, budget, and power.
• Continued emphasis on sustainable data center infrastructure with solutions for retrofitting existing air-cooled environments and building new, environmentally sustainable data centers.
• Ampere CPUs capable of running a wide range of workloads, from cloud-native applications to AI, integrating seamlessly with traditional applications.
• Ampere's ongoing commitment to pioneering the efficiency frontier in computing, delivering high performance within an efficient computing envelope.

https://youtu.be/pmHnZy7AjSk

false

Read More

Edgecore intros AI Server AGS8200 powered by Intel Gaudi

Edgecore Networks has launched the AGS8200 AI server, designed to meet the needs of AI (Artificial Intelligence) and ML (Machine Learning) applications. This GPU-based server is suitable for training large language models, automation, and object classification and recognition tasks. The AGS8200 aims to enhance performance and efficiency for users in the AI and ML fields.

The AGS8200 is built with eight Intel Habana Gaudi 2 processors and dual 4th Gen Xeon Scalable Sapphire-Rapids processors. The Gaudi 2 processor features 96GB HBM2E memory and 24 NICs of 100Gbps RoCEv2 RDMA, offering strong connectivity and scalability for deep learning processes. The server's design allows for seamless training across all eight Gaudi 2 processors without needing an external Ethernet switch.

Additionally, the AGS8200 supports six QSFP-DD ports for scale-out, connecting to 400Gbps switches or 800Gbps switches via breakout cables. This makes it a practical choice for integrating into racks and clusters of Intel Gaudi 2-based nodes, optimizing AI/ML connectivity. The server is available for purchase through Edgecore Networks and its authorized distributors, making advanced AI technology accessible to businesses and research institutions.

Key Features of the AGS8200:

• Powered by the Intel Gaudi 2 AI accelerator, the AGS8200 is suitable for training and inferencing large language models (LLM).
• The Intel Gaudi 2 processors are designed for efficient heat dissipation and energy use, offering cost benefits.
• Edgecore's commitment to open and optimized hardware and software supports initiatives like OCP, TIP, and SONiC.
• The server's 24 x 100Gbps NICs and 21 x 100Gbps non-blocking connectivity support robust internal and external scale-out.
• Six QSFP-DD ports enable the AGS8200 to connect to high-speed switches, ensuring scalability in AI/ML environments.
• The AGS8200 is available for purchase, providing access to advanced AI technology through Edgecore Networks and its authorized distributors.

false

Read More

Molex releases Advanced Thermal Management report for data centers

 Molex released a report on thermal management challenges and innovations as data centers face increasing demands for high-speed data throughput. The report highlights the need to address growing power density and heat dissipation in critical servers and interconnect systems. By exploring new thermal management approaches for I/O modules, Molex aims to provide solutions that support the transition from 112 Gbps to 224 Gbps connectivity while enhancing overall system efficiency.

Doug Busch, VP & GM of Molex’s Enabling Solutions Group, emphasized the importance of integrating new cooling technologies to optimize airflow and thermal management within next-generation data centers. As the demand for faster data processing and storage escalates, the heat generated by high-performance servers also rises. Molex is pioneering innovations in both copper and optical platforms, along with power management products, to improve system cooling capabilities and energy efficiency. This includes the application of optical connectivity and modules combined with advanced cooling solutions.

The report explores the latest air and liquid cooling techniques, such as direct-to-chip liquid cooling and immersion cooling. Molex's integrated floating pedestal liquid cooling solution addresses the challenges of cooling pluggable I/O modules by allowing a single cold plate to adapt to different configurations. Additionally, Molex's Drop Down Heat Sink (DDHS) technology offers a reliable method to enhance heat transfer without compromising component integrity. By collaborating with industry leaders in the Open Compute Project, Molex is at the forefront of developing next-generation cooling technologies to meet evolving data center needs.

Highlights:

• Thermal Management Approaches: Innovations include drop-down heat sink solutions for I/O modules.
• Demand for Faster Data Processing: Rise in heat generation from high-performance servers.
• Liquid Cooling Solutions: Integration of direct-to-chip and immersion cooling within existing form factors.
• Integrated Floating Pedestal: Each pedestal moves independently, optimizing heat transfer efficiency.
• Drop Down Heat Sink (DDHS): Eliminates metal-to-metal contact, improving installation and durability.
• Industry Collaboration: Participation in the Open Compute Project to advance cooling technologies.

www.molex.com

false

Read More

MaxLinear launches 2.5G 7 & 10 port switches and 2.5G Octal-port Ethernet PHYs

MaxLinear has introduced a new family of 2.5G Ethernet switches and PHYs, featuring both 7-and 10-port switches and 8-port PHYs. These single-chip solutions are designed for consumer and industrial applications, boasting a large packet buffer, low power consumption, and an extensive feature set.

Key Features

1. Single-Chip Solutions:

• Switches: 7-and 10-port configurations.
• PHYs: 8-port configuration.

2. Power Efficiency:

• Operates at just 600mW per port.
• Total power consumption for eight PHYs is 4.8W.
• Power consumption for five PHYs is 3.6W.

3. Advanced Features:

• Wire-speed switching.
• IGMPv3 (Internet Group Management Protocol).
• MLPv1,2 (Multi-Link Point-to-Point Protocol).
• LACP (Link Aggregation Control Protocol).
• PTP (Precision Time Protocol).
• SyncE (Synchronous Ethernet).
• MACsec (Media Access Control Security).

4. Uplink Flexibility:

• Options include 10G SERDES uplink ports.

5. Versatility:

• Support for unmanaged, web-smart, cloud-managed, and industrial switches.

The new family enhances MaxLinear’s Ethernet product line, which already includes:

• 10/100 Ethernet solutions.
• 1G Ethernet solutions.
• 2.5G bridges, PHYs, and switches.

"We are excited about MaxLinear's innovative 2.5G Ethernet solutions, which provide the high data rates we require while maintaining cost-effectiveness and energy efficiency,” said Andy Yang, GM of switch BU of Tenda. 

“The inclusion of advanced features like IGMPv3, LACP, HTTPS, ACL, and LLDP in a single-chip design simplifies our network deployments and enhances overall functionality."

false

Read More

Expedera raises $20 million for AI chip tech

Expedera, a start-up based in Santa Clara, California, raised $20 million in Series B funding.

Expedera is a licensor of edge inference artificial intelligence (AI) semiconductor intellectual property (IP.

Expedera introduced its Origin NPU IP in 2021. Origin IP scales to 128 TOPS with a single engine and features highly efficient processor utilization with market-leading low-power consumption while requiring minimal external memory bandwidth. Expedera has production license contracts with multiple smartphone OEMs and numerous automotive, AR/VR/MR, and consumer device chipset makers. Its Origin IP has been deployed in over 10M devices.

The company says the funding will enable it to advance its high-performance, energy-efficient Neural Processing Unit (NPU) IP products and services and expand its global engineering footprint.

The funding round was led by indie Semiconductor, with participation from prominent repeat and new industry investors. This investment, which includes converting a previous SAFE note, brings the company's total funding to more than $47M. 

"AI is rapidly becoming one of the most important enablers for all market segments where edge computation and data intelligence is paramount, including in the automotive space," said Donald McClymont, indie co-founder and CEO. "Expedera has quickly emerged as the market leader in enabling edge inference with its Origin NPU IP. indie is excited to be on this journey with Expedera as it unlocks multiple new applications - including ADAS and sensor fusion - with class-leading edge AI performance and efficiency." 

"Since our Series A round, Expedera has dramatically increased the capabilities of our Origin hardware and software stack, added support for leading-edge neural networks, grown our customer base, and expanded into a global company," said Da Chuang, Expedera co-founder and CEO. "The addition of indie as lead among our new investors, and the continued support of our existing investors is a testament to the strength of Expedera's NPU solutions." 

https://www.expedera.com

false

Read More

Netherland's signetbreedband deploys Nokia's advanced core

Nokia has secured a significant contract with Dutch managed network solutions provider Signetbreedband, a subsidiary of team.blue, a leading digital services provider in Europe. By integrating its advanced routing technology into Signetbreedband's core network, Nokia is enhancing the provider's ability to offer reliable and robust business networking services across the Netherlands.

Nokia is instrumental in the development of Signetbreedband’s new core network, "Connectivity Core." This network features automated and scalable components and integrates various fiber, LTE, and DSL partners to create an open, aggregated network. Nokia has deployed its 7250 interconnect router (IXR) family to build a resilient, scalable, and high-performance MPLS-based core network for Signetbreedband. Additionally, the 7250 IXR routers provide termination points for all regional Points of Presence (POPs).

The deployment also includes Nokia's 7750 Service Router (SR) as the Broadband Network Gateway (BNG), which is crucial for Signetbreedband’s business service networking architecture. The BNG enables automatic authentication, authorization of network access, enforcement of bandwidth policies, and recording of usage data for accounting. This evolution allows Signetbreedband to transition from static provisioning to an automated, centralized system using advanced protocols like NETCONF for improved data configuration management.

Highlights:

• Nokia partners with Signetbreedband, enhancing their core network capabilities.
• New core network "Connectivity Core" integrates automated, scalable components with fiber, LTE, and DSL partners.
• Deployment of 7250 IXR routers provides a resilient, scalable MPLS-based core network with regional POP termination points.
• Integration of 7750 Service Router acts as Broadband Network Gateway, central to business service networking.
• Transition to advanced protocols like NETCONF for better data configuration management through automated, centralized provisioning.
• Improved reliability and performance for business network services in the Netherlands.

https://www.nokia.com

Read More