Showing posts with label DPU. Show all posts
Showing posts with label DPU. Show all posts

Tuesday, May 10, 2022

Intel unveils IPU roadmap with ASIC and FPGA designs

Intel unveiled its IPU roadmap extending through 2026, featuring new FPGA + Intel architecture platforms (code-named Hot Springs Canyon) and the Mount Morgan (MMG) ASIC, as well as next-generation 800GB products. The discussion also a look at Intel's open-source software foundation, including the infrastructure programmer development kit (IPDK), which builds upon SPDK, DPDK and P4.

In terms of the timeline, Intel's roadmap includes:

  • 2022: Mount Evans, the company's first ASIC IPU; and Oak Springs Canyon, Intel’s second-generation FPGA IPU shipping to Google and other service providers.
  • 2023/24: introduction of 400 Gbps IPUs, code-named Mount Morgan and Hot Springs Canyon,
  • 2025/26: introduction of 800 Gbps IPUs

Here are some details on the 200 Gbps and 400 Gbps IPUs:

Mount Evans -- the code name for Intel’s first ASIC IPU, architected and developed with Google Cloud

  • Hyperscale-ready, it offers high-performance network and storage virtualization offload while maintaining a high degree of control.
  • Provides a programmable packet processing engine enabling use cases like firewalls and virtual routing.
  • Implements a hardware accelerated NVM storage interface scaled up from Intel Optane technology to emulate NVMe devices.
  • Deploys advanced crypto and compression acceleration, leveraging high- performance Intel Quick Assist technology.
  • Can be programmed using commonly deployed, existing, software environments, including DPDK, SPDK; the pipeline can be configured utilizing P4 programming.
  • Shipping is expected to begin in 2022 to Google and other service providers; broad deployment is expected in 2023.

Oak Springs Canyon -- the code name for Intel’s 2nd generation FPGA-based IPU platform built with the Intel Xeon D and the Intel Agilex FPGA, the industry’s leading FPGA in power, efficiency, and performance.

  • Network virtualization function offload for workloads like open virtual switch (OVS) and storage functions like NVMe over fabric and RoCE v2
  • Standard yet customizable platform that enables customers to customize their data path and their solutions with FPGA and Intel Xeon-D with software like Intel Open FPGA Stack, a scalable, source-accessible software and hardware infrastructure
  • Programmable using commonly deployed existing software environments, including DPDK and SPDK, which have been optimized on x86.
  • A more secure, high speed 2x 100 gigabit Ethernet network interface with the hardened crypto block
  • VirtIO support in Hardware for Native Linux support

Mount Morgan -- a next-generation ASIC IPU expected in 2023/2024.

Hot Springs Canyon -- a next-generation FPGA-based IPU platform expected in 2023/2024.

https://www.intel.com/content/www/us/en/newsroom/home.html

Intel demos Xeon + Tofino switch + Mount Evans IPU

As part of the Intel Innovation event this week, Intel demonstrated an Intelligent Fabric based on its Xeon Scalable processors and next-generation Xeon D processors, Tofino 3 programmable switching silicon and new "Mount Evans" infrastructure processing unit (IPU). The idea is to leverage P4 programming across all 3 processing platforms for use cases such as near real-time telemetry and analytics with the Intel Deep Insight Network Analytics...

Google collaborates on Intel's ASIC-based infrastructure processor

Intel and Google Cloud announced a deep collaboration to develop an ASIC P4-programmable infrastructure processing unit (IPU).Code-named “Mount Evans,” this open solution supports open source standards, including an infrastructure programmer development kit (IPDK) to simplify developer access to the technology in Google Cloud data centers. Machine learning, large-scale data processing and analytics, media processing, and high-performance computing...

Intel rolls FPGA-based Infrastructure Processing Unit (IPU)

Intel outlined its vision for the infrastructure processing unit (IPU), a programmable network device that intelligently manages system-level infrastructure resources by securely accelerating those functions in a data center.In a video, Guido Appenzeller, chief technology officer with Intel's Data Platforms Group says the idea is to cleanly separate the processing of client workloads from workloads of the cloud service provider.Intel cites several...

Tuesday, April 19, 2022

Fungible launches a partner program

Fungible is launching a partnership program for its composable data center solutions.

The Fungible Partner Exchange (FunPX) partner program allows resellers, distributors, and technology providers worldwide to connect, collaborate, and grow their composable infrastructure capabilities to cloudify the world’s data centers.

“It’s my pleasure to announce the FunPX partner program. We recognize the importance of the channel. Our partners have been instrumental in fueling Fungible’s success and are a critical component in reaching our growth targets,” said Brian McCloskey, Chief Revenue Officer at Fungible. “We are investing in the tools, people, and processes to enable our partners to grow with us by providing immediate access to the essential information they need to win and help our mutual customers realize the benefits that the Fungible portfolio can bring to their infrastructure.”

https://www.fungible.com/partners

Fungible leverages DPU to DPU to centralize GPUs across Ethernet

Fungible introduced a means by which data centers could centralize their existing GPU assets into a single resource pool to be attached to servers on demand.The Fungible GPU-Connect (FGC) solution leverages the company's DPU to dynamically compose GPU and CPU resources across an Ethernet network.Instead of dedicated GPUs sitting idle most of the time, data centers can provide new users with access to the GPU pool, making greater use of existing assets....


Tuesday, December 21, 2021

NVIDIA's BlueFiled DPU claims performance record: 41.5 million IOPS

NVIDIA is reporting a new performance benchmark for DPUs:  two BlueField-2 data processing units reached 41.5 million input/output operations per second (IOPS) — more than 4x more IOPS than any other DPU.

The BlueField-2 DPU delivered record-breaking performance using standard networking protocols and open-source software. It reached more than 5 million 4KB IOPS and from 7 million to over 20 million 512B IOPS for NVMe over Fabrics (NVMe-oF), a common method of accessing storage media, with TCP networking, one of the primary internet protocols.

To accelerate AI, big data and high performance computing applications, BlueField provides even higher storage performance using the popular RoCE network transport option.

In testing, BlueField supercharged performance as both an initiator and target, using different types of storage software libraries and different workloads to simulate real-world storage configurations. BlueField also supports fast storage connectivity over InfiniBand, the preferred networking architecture for many HPC and AI applications.

Testing Methodology

The 41.5 million IOPS reached by BlueField is more than 4x the previous world record of 10 million IOPS, set using proprietary storage offerings. This performance was achieved by connecting two fast Hewlett Packard Enterprise Proliant DL380 Gen 10 Plus servers, one as the application server (storage initiator) and one as the storage system (storage target).

Each server had two Intel “Ice Lake” Xeon Platinum 8380 CPUs clocked at 2.3GHz, giving 160 hyperthreaded cores per server, along with 512GB of DRAM, 120MB of L3 cache (60MB per socket) and a PCIe Gen4 bus.

To accelerate networking and NVMe-oF, each server was configured with two NVIDIA BlueField-2 P-series DPU cards, each with two 100Gb Ethernet network ports, resulting in four network ports and 400Gb/s wire bandwidth between initiator and target, connected back-to-back using NVIDIA LinkX 100GbE Direct-Attach Copper (DAC) passive cables. Both servers had Red Hat Enterprise Linux (RHEL) version 8.3.

https://blogs.nvidia.com/blog/2021/12/21/bluefield-dpu-world-record-performance/

Tuesday, October 5, 2021

Marvell debuts 5nm Prestera switch and 5nm OCTEON 10 5nm DPU

Marvell announced the availability of its 5nm Prestera DX 7321 Ethernet switch, the industry’s first 5nm 50G PAM4 device for the carrier market.

The new Marvell Prestera switch is designed for 5G fronthaul and edge applications. With the addition of the 5nm Prestera device, Marvell's carrier-optimized switch portfolio now comprises four Ethernet switches that scale port speeds from 1Gbps to 400Gbps with aggregate bandwidths ranging from 200Gbps to 1.6Tbps. 

Marvell also announced sampling of its 5nm OCTEON 10 DPU family, which will include hardware accelerators to address workloads required by 5G, cloud, carrier and enterprise data center applications. OCTEON 10 offers a mix of compute, hardware acceleration, data path bandwidth, and I/O including PCIe 5.0 and DDR5. 

Marvell’s 5G network solutions also include Alaska Ethernet PHYs and a broad portfolio of optical solutions for edge, metro and long-haul applications. In addition, Marvell’s COLORZ II 400ZR/ZR+ coherent optical modules are enabling carriers with high-capacity, long-reach DWDM connectivity directly from Ethernet switches.

“Carriers are requiring optimized high-performance, low-power and secured next-generation 5G data infrastructure solutions that enable new high-value applications,” said Raghib Hussain, president, Products and Technologies at Marvell. “Our 5nm switch and DPU set a new bar for power, performance and footprint, providing the breakthrough technology needed to fulfill the potential of 5G.”

https://www.marvell.com/

Wednesday, September 29, 2021

Fungible enhances its Storage Cluster for NVMe over TCP

Fungible, a start-up pursuing DPU accelerated data center computing, announced new capabilities enabling its Fungible Storage Initiator (SI) cards installed in standard servers to access NVMe over TCP (NVMe/TCP).

Fungible claims its accelerators deliver the world’s fastest and most efficient implementation of NVMe/TCP. The enhancements bring security and usability capabilities for the entire data platform.


The Fungible Storage Initiator solution is delivered on Fungible’s FC200, FC100 and FC50 cards. Each of these cards is powered by the S1 Fungible DPU, and a single FC200 card is capable of delivering a record breaking 2.5 million IOPS to its host. These cards, and the Fungible Storage Cluster, are managed by Fungible Composer, which orchestrates the composition of disaggregated data center resources on demand.  

“With our high-performance and low-latency implementation, Fungible’s disaggregated NVMe/TCP solution becomes a game changer. Over the last five years, we have designed our products to support NVMe/TCP natively to revolutionize the economics of deploying flash storage in scale-out implementations,” said Eric Hayes, CEO of Fungible. “In addition to industry leading performance, our solutions offer more value and the highest levels of security, compression, efficiency, durability and ease of use. At Fungible, we continue to disrupt the traditional rigid models by disaggregating compute and storage using available industry standards like NVMe/TCP.”

https://www.fungible.com/

Fungible ships its disaggregated NVMe storage platform powered by DPUs

Fungible, a start-up based in Santa Clara, California, unveiled a disaggregated data storage platform powered by its own Fungible Data Processing Unit (DPU).

The new Fungible Storage Cluster delivers 15M IOPS in a 2RU form factor, scaling linearly to 300M IOPS in a single 40RU rack, and extending further to many racks. The company says its high-performance design improves $/IOPS by at least 3x compared to existing software-defined storage solutions by consolidating workloads and increasing utilization of storage media.

The Fungible Storage Cluster comprises a cluster of Fungible FS1600 storage target nodes connected over a standards-based IP network and the Fungible Composer software. The FS1600s implement the data path for storage while the Fungible Composer performs control and management functions. This clean separation of functions results in higher performance, better scalability and better reliability. Each FS1600 storage target node is powered by two Fungible F1 DPUs and packs 24 standard NVMe SSDs delivering an aggregate of 15M IOPS in a 2RU form factor.

Notably, the Fungible Storage Cluster has been validated with IBM Spectrum Scale, delivering more than 80M read IOPS/PB.

“Today, we demonstrate how the breakthrough value of the Fungible DPU is realized in a storage product,” said Pradeep Sindhu, CEO and Co-Founder of Fungible. “The Fungible Storage Cluster is not only the fastest storage platform in the market today, it is also the most cost-effective, reliable, secure and easy to use. This is truly a significant milestone on our journey to realize the vision of Fungible Data Centers — where compute and storage resources are hyperdisaggregated and then composed on-demand to dynamically serve application requirements.”

“Innovations in data center infrastructure have occurred largely within the silos of compute, storage and networking,” said Raj Yavatkar, CTO at Juniper Networks. “Fungible has broken down these silos delivering end-to-end value with Fungible DPU enabled servers interconnected by TrueFabric, a truly ground-breaking networking technology, and software composable for on-demand provisioning. This approach will serve as a blueprint for future data centers from core to edge.”



Wednesday, July 21, 2021

Fungible appoints Eric Hayes as CEO

 Fungible has appointed Eric Hayes as its new CEO and member of Fungible’s Board of Directors, succeeding Pradeep Sindhu, who has served as Executive Chairman and CEO since Fungible’s inception. Sindhu will continue his role as Executive Chairman and assume the role of Chief Development Officer, where he will lead the engineering teams responsible for the company’s products and solutions.

Hayes most recently served as the Senior Vice President and General Manager of the High-Speed Connectivity business unit at Inphi where he led the company’s multi-hundred-million-dollar PAM4 DSP business. Prior to joining Inphi he held multiple senior leadership positions in marketing and general management at Marvell, Cavium and Broadcom. 


“It has been a great privilege for me to lead this extraordinarily talented group of dedicated individuals to invent the DPU and bring to market industry leading products that exploit its unique capabilities. The DPU is a new category of microprocessor destined to become a key building block of data centers as the industry embraces data-centric computing. My new role allows me to focus on technology: taking the learnings from our first generation of DPUs and applying them to the next and further enhancing our already industry leading products,” said Pradeep Sindhu, Co-Founder, Executive Chairman and Chief Development Officer of Fungible. 

“There are tremendous opportunities for Fungible to radically transform the global data center industry in the coming years, thanks to the great work of Pradeep and the team,” said Eric Hayes, CEO of Fungible. “I can’t express how truly inspired I am to join Fungible at this pivotal time. While many other companies continue to invest in faster CPUs and GPUs, the real bottleneck to achieving performance at scale remains the inability to efficiently disaggregate CPUs, GPUs and storage over a high performance standards-based network. The market is ripe for disruption, and Fungible’s DPU is the only technology capable of solving this problem.”


Thursday, June 17, 2021

Intel rolls FPGA-based Infrastructure Processing Unit (IPU)

Intel outlined its vision for the infrastructure processing unit (IPU), a programmable network device that intelligently manages system-level infrastructure resources by securely accelerating those functions in a data center.

In a video, Guido Appenzeller, chief technology officer with Intel's Data Platforms Group says the idea is to cleanly separate the processing of client workloads from workloads of the cloud service provider.

Intel cites several advantages for an IPU architecture. First, cloud service providers will be able to process their workloads very efficiently on silicon designed for the tasks. The cloud service provider will be able to rent out 100% of available CPU resources to the clients. The clients are now able to fully control the CPUs including with their own hypervisors. In addition, local disk storage can be enhanced or replaced with virtual storage connected to the network. 

The first instances of Intel's FPGA-based IPU platforms are already shipping. Intel will be rolling out additional FPGA-based IPUs and integrated ASICs as well. "As a result of Intel’s collaboration with a majority of hyperscalers, Intel is already the volume leader in the IPU market with our Xeon-D, FPGA and Ethernet components,” said Patty Kummrow, vice president in the Data Platforms Group and general manager of Ethernet Products Group at Intel. “The first of Intel’s FPGA-based IPU platforms are deployed at multiple cloud service providers and our first ASIC IPU is under test.”

https://youtu.be/ahChxDyl8t4

https://www.intel.com/content/www/us/en/newsroom/news/infrastructure-processing-unit-data-center.html





Monday, April 12, 2021

NVIDIA outlines roadmap for BlueField data processing units (DPUs)

NVIDIA outlined a roadmap for the next two generations of its BlueField data processing unit (DPU) chips.

The current generation, BlueField-2, is now shipping and features dual 100Gb/s Ethernet or InfiniBand network ports and up to eight Arm cores. The BlueField-2 DPU includes accelerators for software-defined storage, networking, security, streaming, line rate TLS/IPSEC cryptography, precision timing for 5G telco and time-synchronized data centers and other cloud infrastructure services.

BlueField-3, which is expected to sample in the first quarter of 2022, will deliver 400GbE/NDR networking performance. It features 10x the accelerated compute power of the previous generation, with 16x Arm A78 cores and 4x the acceleration for cryptography. BlueField-3 is also the first DPU to support fifth-generation PCIe and offer time-synchronized data center acceleration. The company says BlueField-3 will enable real-time network visibility, detection and response for cyber threats and acts as the monitoring, or telemetry, agent for NVIDIA Morpheus, an AI-enabled, cloud-native cybersecurity platform, also announced today.



BlueField-3 takes advantage of NVIDIA DOCA, NVIDIA'S new data-center-on-a-chip software platform for building software-defined, hardware-accelerated networking, storage, security and management applications running on BlueField DPUs. Released today and available for download, DOCA includes a runtime environment to create, compile and optimize applications for the BlueField DPU; orchestration tools to provision, update and monitor thousands of DPUs across the data center; as well as libraries, APIs and a growing number of applications, such as deep packet inspection and load balancing.



BlueField-4 was shown on a slide with a projected 2024 horizon. It is expected to scale to 800GbE performance.

“Modern hyperscale clouds are driving a fundamental new architecture for data centers,” said Jensen Huang, founder and CEO of NVIDIA. “A new type of processor, designed to process data center infrastructure software, is needed to offload and accelerate the tremendous compute load of virtualization, networking, storage, security and other cloud-native AI services. The time for BlueField DPU has come.”

“Red Hat continues to collaborate with NVIDIA as part of an open ecosystem that accelerates innovation while providing access to the latest hardware innovations for composable infrastructure,” said Chris Wright, chief technology officer of Red Hat. “We recognize the need to develop advanced solutions for network security and automation and are excited to support BlueField DPUs and the NVIDIA Morpheus AI framework via Red Hat Enterprise Linux, Red Hat OpenShift, industry-leading containers and Kubernetes-powered hybrid cloud platform.”

“Our mutual customers are racing to harness the power of AI for enterprise applications,” said Lee Caswell, vice president of marketing for the Cloud Platform Business Unit at VMware. “The vision of enterprise infrastructure powered by the VMware Cloud Foundation and to be certified with the newly announced NVIDIA BlueField-3 DPU shows customers a path to improved application performance, a consistent operating model across virtualized and bare-metal environments, along with a new model for delivering zero-trust security without compromising performance.”

NVIDIA also disclosed that it is using its BlueField DPUs to accelerate its GeForce Now online gaming service.



NVIDIA acquires Mellanox - focus on Next Gen Data Centers

NVIDIA completed its $7 billion acquisition of Mellanox Technologies. The deal was originally announced on March 11, 2019. NVIDIA says that by combining its computing expertise with Mellanox’s high-performance networking technology, data center customers will achieve higher performance, greater utilization of computing resources and lower operating costs. “The expanding use of AI and data science is reshaping computing and data center architectures,”...


Monday, March 1, 2021

Marvell to supply OpenRAN Distributed Unit design for Facebook's Evenstar

 Marvell  will join the Evenstar program and work with Facebook Connectivity to provide a 4G/5G OpenRAN Distributed Unit (DU) design for Evenstar based on its OCTEON Fusion baseband processors and Arm-based OCTEON multi-core digital processing units (DPUs). 

The Evenstar program is a collaborative effort focused on building a general-purpose RAN reference architecture for 4G and 5G networks in the OpenRAN ecosystem. 

Specifically, Marvell will supply a fully integrated DU reference board featuring the OCTEON Fusion-O baseband, providing 4G and 5G PHY layer processing and an OCTEON DPU to run software functions. Facebook Connectivity will collaborate with Marvell to enable software operations on this solution and encourage multiple third parties to port protocol stack software as well. The DU supports up to 16 downlink layers at 100 MHz channelization with 10 Gbps downlink and 5Gbps uplink performance. The goal is to have Evenstar DU equipment ready for network operator trials next year.


“Marvell demonstrated its commitment to OpenRAN by introducing our O-RAN Platform in December 2020 based on our industry-leading OCTEON Fusion architecture,” said Raj Singh, executive vice president of the Processors Business Group at Marvell. “We also realize that the key to success – not just for our platform, but for the industry as a whole – involves collaboration such as this Evenstar DU engagement with Facebook Connectivity and other partners. We look forward to delivering a high performance, cost-optimized DU design that supports the capacity and interoperability requirements necessary to deploy versatile, next-generation networks by the operator community.”

Tuesday, December 8, 2020

Marvell debuts O-RAN solution based on Octeon DPU

Marvell introduced a 5G O-RAN portfolio consisting of silicon, software and hardware reference designs spanning the radio unit (RU), distributed unit (DU) and centralized unit (CU) with Ethernet connectivity between these network elements.

Marvell’s O-RAN platform includes its OCTEON Fusion baseband processor family that is currently shipping in volume to tier 1 OEMs. Marvell’s newly announced OCTEON Fusion-O processors maintain the full capabilities of the OCTEON Fusion family including 15 Gbps capacity and support for 200MHz channelization, while adding specific O-RAN functions such as eCPRI compression/decompression and split 7.2 interfaces with integrated fronthaul. Additionally, the platform includes key reference software, allowing OEMs, ODMs and network operators to rapidly bring O-RAN compatible products to market.

mMIMO RU

A 32T32R, 8 DL layer, 100MHz channel radio unit.

  • Hardware reference design in partnership with Analog Devices
  • Reference lower L1 and beamforming software
  • O-RAN management plane software
  • O-RAN fronthaul interface
  • Analog Devices Transceivers and DFE software

vDU and DU

The OCTEON Fusion-O DU hardware reference designs are available as either a virtualized DU (vDU) PCIe-based offload accelerator or an integrated DU. 

  • Dual mode 5G and LTE
  • Integrated fronthaul for 7.2 split
  • 16DL/8UL layer support
  • Reference LTE and 5G L1 software

CU

In addition to support for vDU server configurations, the Marvell O-RAN platform also provides support for our industry-leading L2/L3 OCTEON processors supporting a fully integrated standalone CU implementation.

  • Dedicated L2/L3 silicon
  • Virtualized L2/L3 support
  • Partnership protocol stack ports from industry leaders

https://www.marvell.com/

Tuesday, October 27, 2020

Fungible ships its disaggregated NVMe storage platform powered by DPUs

Fungible, a start-up based in Santa Clara, California, unveiled a disaggregated data storage platform powered by its own Fungible Data Processing Unit (DPU).

The new Fungible Storage Cluster delivers 15M IOPS in a 2RU form factor, scaling linearly to 300M IOPS in a single 40RU rack, and extending further to many racks. The company says its high-performance design improves $/IOPS by at least 3x compared to existing software-defined storage solutions by consolidating workloads and increasing utilization of storage media.

The Fungible Storage Cluster comprises a cluster of Fungible FS1600 storage target nodes connected over a standards-based IP network and the Fungible Composer software. The FS1600s implement the data path for storage while the Fungible Composer performs control and management functions. This clean separation of functions results in higher performance, better scalability and better reliability. Each FS1600 storage target node is powered by two Fungible F1 DPUs and packs 24 standard NVMe SSDs delivering an aggregate of 15M IOPS in a 2RU form factor.

Notably, the Fungible Storage Cluster has been validated with IBM Spectrum Scale, delivering more than 80M read IOPS/PB.

“Today, we demonstrate how the breakthrough value of the Fungible DPU is realized in a storage product,” said Pradeep Sindhu, CEO and Co-Founder of Fungible. “The Fungible Storage Cluster is not only the fastest storage platform in the market today, it is also the most cost-effective, reliable, secure and easy to use. This is truly a significant milestone on our journey to realize the vision of Fungible Data Centers — where compute and storage resources are hyperdisaggregated and then composed on-demand to dynamically serve application requirements.”

“Innovations in data center infrastructure have occurred largely within the silos of compute, storage and networking,” said Raj Yavatkar, CTO at Juniper Networks. “Fungible has broken down these silos delivering end-to-end value with Fungible DPU enabled servers interconnected by TrueFabric, a truly ground-breaking networking technology, and software composable for on-demand provisioning. This approach will serve as a blueprint for future data centers from core to edge.”

Fungible announces its DPU for scale-out data centers

Fungible, a start-up based in San Jose, California, unveiled its Fungible Data Processing Unit (Fungible DPU), a microprocessor optimized for data interchange and data-centric computation in scale-out architectures. Fungible describes its DPU as the "third socket" in data centers, complementing the CPU and GPU, and delivering significant gains in performance, footprint and cost efficiencies for next-generation, scale-out networking, storage, security,...

Fungible raises $200 million for Data Processing Units (DPUs)

Fungible, a start-up based in Santa Clara, California, closed $200 million in Series C financing for its efforts to create an entirely new category of programmable processor. The Fungible Data Processing Unit (DPU) aims to deliver an order of magnitude improvement in the execution of data-centric workloads. The company sees its DPU as a fundamental building block for next-generation data centers. Fungible has not yet announced its first products....

Video: Scaling-out Data Centers with the Fungible Data Processing Unit

Fungible was founded in 2015 to revolutionize the performance, economics, reliability, and security of scale-out data centers. In this video, Pradeep Sindhu, Co-Founder and CEO of Fungible, shares observations about scale-out data centers and the key innovations of Fungible’s Data Processing Unit (Fungible DPU™) which has been positioned as the “third socket” in data centers, complementing the CPU and GPU. https://youtu.be/spJAOn_y21A...


Monday, October 5, 2020

NVIDIA unveils DPUs and Data-Center-Infrastructure-on-a-Chip SDK

 NVIDIA has begun sampling its second-generation family of BlueField data processing unit (DPUs) for accelerating data center infrastructure.

The new BlueField-2 DPU is designed to offload critical networking, storage and security tasks from server CPUs. The company says a single chip can deliver the same data center services that could consume up to 125 CPU cores. This frees up valuable CPU cores to run a wide range of other enterprise applications.

NVIDIA’s current DPU lineup includes two PCIe products:

  • The NVIDIA BlueField-2 DPU, which features all of the capabilities of the NVIDIA Mellanox ConnectX-6 Dx SmartNIC combined with Arm cores. Fully programmable, it delivers data transfer rates of 200 Gbps and accelerates key data center security, networking and storage tasks, including isolation, root trust, key management, RDMA/RoCE, GPUDirect, elastic block storage, data compression and more.
  • The NVIDIA BlueField-2X DPU, which includes all the key features of a BlueField-2 DPU enhanced with an NVIDIA Ampere GPU’s AI capabilities that can be applied to data center security, networking and storage tasks. Drawing from NVIDIA’s third-generation Tensor Cores, it is able to use AI for real-time security analytics, including identifying abnormal traffic, which could indicate theft of confidential data, encrypted traffic analytics at line rate, host introspection to identify malicious activity, and dynamic security orchestration and automated response.

BlueField-2 DPUs are sampling now and expected to be featured in new systems from leading server manufacturers in 2021. BlueField-2X DPUs are under development and are also expected to become available in 2021.

NVIDIA says its BlueField DPUs are being adopted by leading server manufacturers worldwide, including ASUS, Atos, Dell Technologies, Fujitsu, GIGABYTE, H3C, Inspur, Lenovo, Quanta/QCT and Supermicro.

NVIDIA is developing a novel data-center-infrastructure-on-a-chip (DOCA) architecture and programming that is intended to be analogous to its CUDA environment for GPUs. A new NVIDIA DOCA software development kit enables developers to rapidly create applications and services on top of NVIDIA BlueField DPUs. he DOCA SDK provides industry-standard open APIs and frameworks, including Data Plane Development Kit (DPDK) and P4 for networking and security and the Storage Performance Development Kit (SPDK) for storage. 

  • NVIDIA is also participating in VMware's recently announced Project Monterey to integrate SmartNICs with VMware Cloud Foundation.
  • Red Hat plans to offer support for BlueField-2 DPUs with Red Hat Enterprise Linux and Red Hat OpenShift, components of Red Hat’s open hybrid cloud portfolio, which is used by 95 percent of the Fortune 500.
  • Canonical announced support of BlueField-2 DPUs and DOCA in its Ubuntu Linux platform, the most popular operating system among public clouds.
  • Check Point Software Technologies, a leading cybersecurity provider, is integrating BlueField-2 DPUs into its technologies, which more than 100,000 organizations worldwide use to protect themselves from cyberattacks.




http://www.nvidia.com


Wednesday, August 19, 2020

Video: Scaling-out Data Centers with the Fungible Data Processing Unit

Fungible was founded in 2015 to revolutionize the performance, economics, reliability, and security of scale-out data centers.

In this video, Pradeep Sindhu, Co-Founder and CEO of Fungible, shares observations about scale-out data centers and the key innovations of Fungible’s Data Processing Unit (Fungible DPU™) which has been positioned as the “third socket” in data centers, complementing the CPU and GPU.

https://youtu.be/spJAOn_y21A