Artificial Intelligence (AI) vehicles are driving change in how businesses manage and handle their data. Traditional storage systems feel clunky and outdated, built for simpler times and fewer users. Modern AI needs real-time, large-scale data access, something old storage systems just can’t keep up with, as they tend to add layers of complexity that stifle AI performance. Data has to journey through multiple stages before it gets to the GPU, the workhorse behind AI models.
Wenn es darum geht, dieses Dilemma zu lösen, ist das Technologieunternehmen Cloudian der Vorreiter. Die MIT-Absolventen Michael Tso und Hiroshi Ohta sind Mitbegründer des Unternehmens, das für seine skalierbare, KI-freundliche Speicherplattform bekannt ist, die den Datenfluss von der Speicherung zu den Verarbeitungseinheiten rationalisiert. Die Architektur umfasst parallele Berechnungen, die Latenzzeiten und Komplexität effektiv reduzieren und direkte Hochgeschwindigkeitsübertragungen von der Speicherung zur GPU oder CPU ermöglichen. Dieses Design gewährleistet eine nahtlose Skalierbarkeit und vereinfacht gleichzeitig die Entwicklung von KI-Tools im kommerziellen Maßstab.
Tso’s mantra when it comes to AI is straightforward: “It’s all about the data.” As he rightly puts it, minor increments in data aren’t enough to improve AI performance. We’re talking about needing 1,000 times more data, stored and processed in a way that doesn’t necessitate moving it around. Cloudian’s platform uniquely allows data to be computed as it’s ingested, removing the need for time-consuming transfers and hence making real-time AI operations possible.
Tso’s passion for data channels back to his time at MIT in the 90s. There, under the mentorship of Professor William Dally, he delved into parallel computing and later worked on distributed networking systems with Associate Professor Greg Papadopoulos. His journey continued at Intel, where he contributed to data synchronization algorithms and helped ignite the ringtone download industry. These experiences, coupled with his adventures at the startup Inktomi and as a co-founder of Gemini Mobile Technologies, equipped him with the necessary insight to chart new territory in data storage.
Cloudian’s journey began in earnest with the rise of cloud computing in the late 2000s. Tso observed a significant bottleneck: the growth of data exceeded the pace of network speeds. Big data, he says, is a lot like gravity: it’s hard to move, which means the cloud solution needs to come to it. This idea led to the foundation of Cloudian in 2012, with its prime focus being distributed, cloud-compatible storage solutions.
Cloudian didn’t initially forsee AI as its primary consumer, but as AI started taking up more of the data usage space, Cloudian’s architecture proved to be a perfect solution. The company’s object storage solution is ideal for handling unstructured data. A significant upgrade in July transformed raw data into a vector form, ready for immediate action by AI models, making engines for search, recommendation, and AI assistants more potent.
In a strategic partnership with NVIDIA, Cloudian has seamlessly integrated its storage systems with NVIDIA’s powerful GPUs. “GPUs,” explains Tso, “are only practical when they’re fed data non-stop.” By embedding AI functions directly into its storage, Cloudian was able to process data closer to where it’s collected which resulted in lower latency and energy costs, faster AI calculations and better efficiency.
Heute unterstützt Cloudian rund 1.000 Organisationen weltweit, darunter Automobilhersteller, Gesundheitsdienstleister, Regierungsbehörden und Finanzunternehmen. Die National Library of Medicine und die National Cancer Database nutzen Cloudian zur Speicherung komplexer Datensätze, die für die KI-gestützte Forschung unerlässlich sind. Gleichzeitig setzt ein großer Automobilhersteller sein KI-Modell ein, um vorherzusagen, wann seine Fabrikroboter gewartet werden müssen.
Tso’s vision for the future of data storage focuses on making GPUs more effective by removing any layers impeding the data path. Cloudian’s AI-first storage strategy is aiding businesses in achieving this goal, helping to transform their raw data into a real-time resource advancing next-generation intelligence.
Den vollständigen Artikel finden Sie im Original unter MIT-Nachrichten.
Diese Website verwendet Cookies.