Wow, it has already been over a week since GTC and the dust is still settling. From Jensen’s keynote address until we tore down our booth it was nonstop activity. If you missed any of our announcements or sessions at the show, read on!
Our top announcement was the launch of our latest appliance for Generative AI and Large Language Model workloads, the AI400X2 Turbo. Boasting 30% more power than its predecessor, the AI400X2 Turbo offers faster performance and increased network connectivity, crucial for supporting the expanding AI workloads and GPU-accelerated processing in data centers. With impressive write speeds of 120 GB/s and read speeds of 75 GB/s per 2U appliance, the AI400X2 Turbo aims to improve ROI for GPU clusters and support Generative AI, Inference, and various AI frameworks more efficiently. The right data storage architecture is essential to effectively accelerating AI ROI and maximizing GPU utilization. The AI400X2 Turbo further solidifies DDN’s stance as a leading provider of high-efficiency, high-performance solutions for large-scale AI and language model application.
DDN also announced its support for GPU systems at NVIDIA’s largest Could Partners. With systems at leading providers like Lambda, Vultr, Scaleway, Bitdeer and others, DDN continues to expand on its success with providers building hosted solutions for or Distributed Training of LLMs, Foundation Models & Generative AI. These providers have selected DDN because of our proven track record of supplying scalable and reliable solutions that make GPU-based cloud implementations more efficient and cost-effective. To get the most out of your hosted workloads, ensure that the GPUs you are using are backed by DDN storage.
Additional announcements included DDN’s support for BlueField 3 DPUs in our EXAScaler and Infinia systems, as well as our announcement of NVIDIA as a DDN customer for the storage behind their Eos AI Supercomputer. As organizations deploy AI infrastructures in their data centers and in the cloud, the need for generative AI and AI frameworks to operate and scale reliably and cost-effectively has become increasingly important. Using these DPUs in our systems will allow us to scale further and increase the efficiency of the data path as these frameworks get larger and more complex. Our experience supporting NVIDIA’s massive environments (the Eos cluster is made up of more than 4600 GPUs) has given us the expertise to deliver efficient solutions for the largest AI systems all around the world and offer a full stack blueprint for turn-key AI infrastructure.
GTC Sessions
You can still view many of the GTC sessions on demand, for free, at the GTC site. Sessions to note include:
- DDN’s senior vice president of products, Dr. James Coomer, discussing what optimizations and instrumentation breakthroughs are needed to scale workloads for thousands of systems worldwide.
- The panel discussion, hosted by NVIDIA’s Rob Davis, features DDN’s CTO, Sven Oehme, discussing the evolving landscape of AI storage platforms and the challenges they face delivering to generative AI models.
- Vice president of marketing, Kurt Kuckein, discussing a wide range of AI storage topics with Matt Marshall of VentureBeat.
For a complete list of DDN’s sessions, check on our GTC event page.
Generative AI and large language models are igniting a revolution, but realizing their full potential for business applications requires well thought out end-to-end data center infrastructure optimization. Whether you’re training language models at scale or deploying generative AI solutions for your business or research initiatives, GTC provided extensive examples on how to optimize your full stack AI infrastructure in data centers or in the cloud. DDN and its partners are redefining what is possible in the era of accelerated computing, and we have plenty more to come!