Search results
Speed-of-Light Data Movement Between Storage and the GPU
GPUs have long been compute monsters. But for a new class of applications with huge datasets and low computational intensity, like generative neural
network (GNN) training and retrieval-augmented generation (RAG), GPU-initiated communication from O(100K) GPU threads enable them to also be data
access monsters. We'll frame the problem of storage IO, tie it to specific usage models and classes of applications, present breakthrough successes, and op…
An Introduction to the OPI (Open Programmable Infrastructure) Project
A new class of cloud and datacenter infrastructure is emerging into the marketplace. This new infrastructure element, often referred to as Data Processing
Unit (DPU), Infrastructure Processing Unit (IPU) or xPU as a general term, takes the form of a server hosted PCIe add-in card or on-board chip(s), containing
one or more ASIC’s or FPGA's, usually anchored around a single powerful SoC device. The OPI project has been created to foster the emergence of an op…
Accelerating Generative AI – Options for Conquering the Dataflow Bottlenecks
Workloads using generative artificial intelligence trained on large language models are frequently throttled by insufficient resources (e.g., memory, storage,
compute, or network dataflow bottlenecks). If not identified and addressed, these dataflow bottlenecks can constrain Gen AI application performance well
below optimal levels. Given the compelling uses across natural language processing (NLP), video analytics, document resource development, image proce…
DPUs, K8s, ML & Ops: The future of compute
The datacenter is evolving. DPUs are fast becoming the third pillar of enterprise computing in addition to the CPU and GPU. At the same time, the increasing
complexity in enterprise software has led to the rise of new tools like Kubernetes and Operators. This talk touches upon why the DPU makes sense and how
best to use this new class of devices. We then demo Ubuntu running a lightweight upstream Kubernetes (microK8s) on the DPU, and discuss the challenges …
BlueField Partner’s DPU Storage Solutions and Use Cases
NVIDIA BlueField data-processing units (DPUs) with NVIDIA DOCA are very good at accelerating many applications, but they're great at accelerating
networked storage solutions. With multiple up to 200 Gb network ports and many lanes of gen4 PCIe, it's the perfect solution to sit between high-performance
NVMe solid-state drives and high-performance network ports. Add to that the ease of development with DOCA and line speed NVMe over fabrics offloads, li…
NVMe/NVMe-oF and DPUs: The Technologies Driving the Next Storage Revolution
Join key storage industry players, and learn more about storage industry trends and how those new technologies are driving the next storage revolution. The
explosion of data, driven by demanding services and applications using artificial intelligence (AI), machine learning (ML), image processing and predictive
analytics, is putting a lot of pressure on Cloud and Edge Data Center’s networking and storage to feed the demanding compute intensive resources. New Fl…
Introduction to SmartNIC DPUs for the data centre
Softwarisation has been continuously influencing the data centre during the past two decades, from compute to networking to storage, culminating in
software-defined data centres (SDDC). For example, networking functions, such as virtual routing and switching done in software on a hypervisor CPU have
become a norm. However, performance optimisation has been a growing concern as well. To address it, a trend was adopted recently by the major device …
Accelerating Enterprise Cybersecurity with Software-defined DPU Firewall
To keep up with the explosion of data, data centers are deploying high-speed networks from 25G to 100G. While network speed has been increasing, security
functions such as next-generation firewalls (NGFW) need to keep up with higher traffic loads. Software-defined NGFW offers the flexibility and agility to build
modern data centers; however, scaling them for performance, efficiency, and economics has been a challenge. The need for a software-defined, hardware…