DriveNets

Watch now! 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗽𝗲𝗿𝗳𝗲𝗰𝘁 𝗳𝗮𝗯𝗿𝗶𝗰 𝗳𝗼𝗿 𝗔𝗜 🎦 drivenets.com/resources/vide… @DriveNets and @EdgecoreNetwork explore how to build a lossless, congestion-free AI fabric using Ethernet — delivering InfiniBand-class performance at a fraction of the cost and driving the utilization of your GPUs up and to the right... What you'll learn! How Fabric Scheduled Ethernet (FSE) and the DriveNets OS on Edgecore platforms enables: ▶️ Predictable, lossless performance for AI clusters ▶️ High-scale fabrics built on standard Ethernet ▶️ Support for any GPU or NIC ▶️ Major reductions in AI infrastructure and operational costs Trim millions of dollars off the costs associated with building and maintaining a large AI cluster, with any GPU or NIC.

Stop touching boxes! Zero-touch provisioning isn’t new. But making it cluster-wide, consistent, and reliable at scale is where things get interesting. medium.com/dn-techbites/s… In this new TechBites, Brad Riapolov breaks down how we: 🔹 Move from per-device bring-up to cluster-level initialization 🔹 Enforce state consistency across all nodes 🔹 Eliminate configuration drift before it even starts The result: faster deployment, fewer human errors, and predictable outcomes across large-scale systems - not just individual boxes. 👉 Follow #DNTechBites for real-world engineering insights, protocol breakdowns, and networking architecture innovations

May The Fourth Be With...You(r Network)! drivenets.com/blog/may-the-f… “May the Force be with you” became the inspiration for the official Star Wars holiday, #MayTheFourth. But how does this connect to high-scale networks? Well, there are a number of lessons to be learned from the galaxy far far away…even for networks.

Let's meet in Vegas! DriveNets is proud to be a Bronze Sponsor at Dell Technologies World 2026! ​ get.drivenets.com/dell-tech-worl… Stop by Booth 506 to see how we’re helping organizations to implement an open, Ethernet-based alternative to InfiniBand that’s built for the most demanding AI workloads. ​​ #DellTechWorld #DellTechnologies

Planning to attend the ONUG #AINetworkingSummit? Come and see DriveNets! 📅 Stand 41 May 13-14, 2026 - Frisco|Dallas Join our session! 📅 May 14 | 12:30-12:40 pm CT Sani Ronen, DriveNets Director of AI Networking, will break down the building blocks of AI clusters and how the most critical parts can be optimized for peak performance. onug.net/events/the-mos…

Networking is 10% of the cost — but 80% of the headache. In @AvidThink Data Center Networking in 2026 report, Dudy Cohen, VP Product Marketing at @drivenets, shares insights into why the network is a critical part of AI infrastructure. "Get the network wrong and you are burning GPU cycles, which is like burning money." Key takeaways: 🔹 Ethernet has closed the performance gap with InfiniBand. NCCL benchmark results: DriveNets' production results on a 512-GPU H200 cluster showed up to 18% better than InfiniBand - and up to 15% better than Spectrum-X - across all-reduce, all-gather, all-to-all, and reduce-scatter operations. 🔹 Two metrics that matter most for AI infrastructure builders: → Time to First production Token (TTFT) - how fast you go from investment to production → Cost per million tokens - marginal cost at scale, which a well-utilized network directly impacts 🔹 Scheduling is what makes Ethernet viable for AI workloads DriveNets offers two approaches: → Fabric scheduling Ethernet: packets are broken into cells, sprayed across the fabric, and reassembled at the destination with zero packet loss → Endpoint scheduling Ethernet: NICs manage traffic using network telemetry. Data Center Networking in 2026 👉 Read the full interview and report insights:  nextgeninfra.io/2026-dcnetwork…

DriveNets welcomes Gali Ross-Hasson as VP of Business Development & Alliances – AI! Gali joins us from @nutanix where she led global business development and strategic partnerships with companies including @AMD @nvidia @IBM @intel and Red Hat. Before that, she consulted at @McKinsey on cloud and IT strategy, and earlier founded and scaled her own venture. She brings a powerful combination of startup experience, strategic insight, and ecosystem leadership — and a deep understanding of enterprise technology and AI infrastructure. We're thrilled to have her on board as we build the future of AI networking. Wishing Gali a warm welcome to the @drivenets team!

Let's meet in Vegas! DriveNets is proud to be a Bronze Sponsor at Dell Technologies World 2026! ​ Want to meet our team? ​get.drivenets.com/dell-tech-worl… Stop by Booth 506 to see how we’re helping organizations to implement an open, Ethernet-based alternative to InfiniBand that’s built for the most demanding AI workloads. ​ #DellTechWorld #DellTechnologies

Is the network the hidden bottleneck holding back AI cluster performance? In this post published in the @ONUG_ blog, @drivenets' Sani Ronen explores the shift from manual integration to a Full-Stack AI Cluster Orchestration model. Read the full breakdown 👉 onug.net/blog/optimizin… Why Full-Stack Orchestration? 🔹 Provisioning: Automate cluster bring-up from hardware discovery to RDMA enablement in one step. 🔹 Benchmarking: Validate the entire AI fabric under real conditions, not just GPUs in isolation. 🔹 Operations: Use real-time telemetry to correlate network events directly with GPU performance. Don't let your expensive accelerators sit idle due to "invisible" network bottlenecks. Planning to attend the @ONUG_ AI Networking Summit? Come and see DriveNets! 📍 Stand 41 🗓 March 13-14, 2026 - Frisco|Dallas Hear our presentation! 🎤 The Most Important Part of Your AI Cluster Isn’t What You Think 📅 May 14 | 12:30-12:40 pm CT onug.net/events/the-mos… Sani Ronen will break down the building blocks of AI clusters and how the most critical parts can be optimized for peak performance.

Celebrating Israel's 78th independence day 🇮🇱

Today on Yom HaZikaron, we stop to honor the fallen and the heroes of Israel. At DriveNets, we were moved by the experience of Maayan Mulla, a Lieutenant Colonel in the Engineering Special Forces. Living in India when October 7th occurred, Maayan left his life and business behind to return and lead his unit. On the 6th night of Hanukkah, he shared the harrowing account of an ambush that tragically claimed the life of 20-year-old Uriah Yaakov z"l on his birthday. Despite being hit himself and riddled with over 140 pieces of shrapnel, Maayan’s first instinct was to care for his team, providing life-saving medical aid and commanding his unit while severely wounded. Maayan’s spirit stands as an inspiration, guided by his powerful motto: "You don't need to experience death up close in order to appreciate life." His journey serves as a powerful testament, reminding us that even in the face of immense pain, our spirit can thrive. We honor the memory of the fallen, salute the bravery of survivors like Maayan, and thank him and salute the bravery of heroes like Maayan, and thank them all our defenders for their sacrifice and service. 🕯️

Is Ethernet really suited for very large GPU clusters in AI data centers? With NextGenInfra.io, Dudy Cohen, VP of Product Marketing at @drivenets, explains how the network has become the critical factor in AI data center success. drivenets.com/resources/vide… Learn how Ethernet is evolving to meet the demands of AI infrastructure - through advancing speeds (800G to 3.2T), co-packaged optics (CPO) integration, and scheduling layers that address its inherent lossiness. With GPU clusters scaling from 8,000 to nearly one million units, hear how DriveNets' scheduled fabric delivers superior time-to-first-token performance and lower cost-per-million-tokens compared to proprietary alternatives.

Join us at the ONUG AI Networking Summit in Frisco | Dallas onug.net/ai-networking-… Building an AI cluster is much more complex than simply racking and stacking hardware. Poor network performance can leave GPUs idle and increase the cost per token. Cluster builders need a networking solution that delivers high performance at any scale, supports diverse use cases and hardware choices, and is simple to operate. Our AI Ethernet alternative to InfiniBand: DriveNets AI Fabric With a full-stack Ethernet solution for backend, frontend, and storage connectivity in one package, it includes hardware, software, and services designed to support any AI architecture, use case, or GPU type.

Join us at the ONUG AI Networking Summit in Dallas onug.net/ai-networking-… Building an AI cluster is much more complex than simply racking and stacking hardware. Poor network performance can leave GPUs idle and increase the cost per token. Cluster builders need a networking solution that delivers high performance at any scale, supports diverse use cases and hardware choices, and is simple to operate. Our AI Ethernet alternative to InfiniBand: DriveNets AI Fabric With a full-stack Ethernet solution for backend, frontend, and storage connectivity in one package, it includes hardware, software, and services designed to support any AI architecture, use case, or GPU type.

Why are the world's largest AI teams moving away from InfiniBand? Performance isn't the problem. Proprietary lock-in, specialist expertise requirements, inflexible lead times, and a 20%-per-year maintenance bill are. DriveNets and Edgecore Networks Corporation explore the various approaches to creating high performance, high scale AI-compatible networks. drivenets.com/resources/vide…

Inference test results of AMD MI355X single node and multi node clusters DriveNets optimized AMD cluster achieved higher total throughput and significantly faster time to first token (TTFT) compared to results from SemiAnalysis’ InferenceX 👉 drivenets.com/blog/system-le…

Download our new white paper!​ drivenets.com/resources/whit… Learn how to build distributed AI infrastructure with unprecedented interconnect efficiency and performance - freeing enterprises and NeoCloud providers from the power and space limitations of a single site deployment.​ Scaling AI workloads beyond a single data center is becoming the default path for many enterprises and neocloud providers. ​ As deployments extend across multiple sites, the network is the critical make-or-break layer that determines overall performance.

Ethernet actually beat InfiniBand in real AI cluster production environment?​ Yes!​ ​drivenets.com/blog/ethernet-… As AI models scale, raw GPU power isn’t the limiting factor anymore - the network is. When GPUs spend time waiting on data, costs explode and performance drops. This is why networking architecture is becoming the decisive factor in large-scale AI infrastructure.​ Tested in a real-world production environment, Learn how Ethernet -done differently - can meet (and exceed) AI training demands at scale.

New on TechBites! What’s actually holding your network back from scaling—technology, or outdated assumptions? medium.com/dn-techbites/g… In this post, Brad Riapolov takes a deep dive into why traditional architectures struggle to keep up with modern demands - and how SRv6 is reshaping what’s possible. Instead of layering complexity on top of legacy designs, SRv6 enables a fundamentally different approach to traffic engineering, programmability, and service delivery. This is Part 2, check out Part 1: medium.com/dn-techbites/u… 👉 Follow #DNTechBites for real-world engineering insights, protocol breakdowns, and networking architecture innovations

Happy Easter from DriveNets As spring arrives, Easter reminds us of the joy found in fresh beginnings, simple moments, and time spent with those who matter most. Wishing all our customers, partners, and friends across the globe a season filled with health, happiness, and bright new opportunities. Wishing you a happy Easter and a joyful spring! 🌷

Your GPU cluster is only as fast as its network — and most networks aren't ready. AI training stalls when the network stalls. Packet drops force GPU waits. Even 0.1% packet loss can slash GPU utilization by 30%+. Networking — not compute — is the silent bottleneck. In this webinar, DriveNets and Edgecore Networks Corporation explore the various approaches to creating high performance, high scale AI-compatible networks. drivenets.com/resources/vide…