Research Finder
Find by Keyword
Can On-Premises GPUs Reignite The Data Center? Oracle Thinks So.
Oracle's Compute Cloud@Customer and Private Cloud Appliance, now powered by NVIDIA L40S GPUs, aim to deliver cloud-like AI and HPC capabilities on-premises, addressing data sovereignty, performance, and cost concerns.
Key Highlights
- Oracle integrates NVIDIA L40S GPUs into its on-premises offerings.
- The solution targets AI, HPC, and graphics workloads within customer data centers.
- It offers independent scaling of compute, GPU, and storage resources.
- The platform aims to provide a balance of performance and operational efficiency.
- Oracle emphasizes data sovereignty and compliance benefits.
The News:
Oracle has enhanced its Compute Cloud@Customer (C3) and Private Cloud Appliance (PCA) with NVIDIA L40S GPUs. This integration aims to bring high performance computing, AI, and graphics processing capabilities directly into customer data centers. The move is designed to address the growing demand for on-premises solutions that offer cloud-like flexibility and scalability while maintaining data sovereignty and control. Learn more by reading the announcement blog here.
Analyst Take:
Cloud repatriation refers to the process where organizations move their data, applications, or workloads back from public cloud environments to their on-premises data centers due to various reasons like data sovereignty, cost, performance, security, or compliance. This shift necessitates a strategic perspective on workload placement, where businesses must evaluate which applications benefit from being on-premises versus in the cloud, based on factors like latency, data sovereignty, and operational costs as well as security, availability and scalability.
The need for hybrid cloud architectures have arisen as a solution to maintain flexibility, allowing companies to leverage both the scalability and innovation of public clouds while ensuring sensitive or high-performance workloads remain in-house. Hybrid clouds enable seamless integration between on-premises infrastructure and cloud services, facilitating dynamic workload placement where applications can move between environments without significant rearchitecture. This approach not only optimizes resource utilization but also enhances resilience, allowing organizations to adapt to changing business needs or regulatory landscapes while minimizing disruptions.
Against this wider backdrop Oracle's move to integrate NVIDIA L40S GPUs into its Compute Cloud@Customer and Private Cloud Appliance offerings is a noteworthy development in the evolving landscape of hybrid cloud and on-premises computing. While Oracle didn’t specifically mention repatriation, and in fact was keen to stress the connectivity with Oracle Cloud Infrastructure (OCI) its public cloud offering, it was in the back of my mind during the briefing they held for analysts. What is probably more accurate is that Oracle doesn’t care - Want your workload on OCI they have you covered, want it in your datacenter they have you covered - also crucially want it connected to your Exadata they have thought about that too. Put simply - optionality is at the core of where Oracle is going when it comes to infrastructure.
This announcement underscores a growing recognition that not all workloads are best suited for the public cloud, particularly those with stringent data residency, sovereignty, or performance requirements. The Cloud@Customer announcement aims to deliver a compelling alternative for organizations seeking to leverage the power of accelerated computing while retaining control over their data and infrastructure - all powered by NVIDIA GPUs.
What was Announced:
Oracle Compute Cloud@Customer and Private Cloud Appliance now feature NVIDIA L40S GPUs, designed to support a wide range of demanding workloads. The L40S GPU, based on the NVIDIA Ada Lovelace architecture, is designed to deliver exceptional performance for AI, HPC, and graphics. It is engineered to offer up to 91.6 teraFLOPS FP32 performance for general computing and up to 1.5 petaFLOPS Tensor Performance with sparsity for AI tasks. The platform is designed to support Large Language Models (LLMs) with up to 70 billion parameters, showcasing its potential for advanced AI applications.
The solution is architected to offer flexible configurations. Customers can scale compute, GPU, and storage resources independently, allowing them to tailor the system to their specific needs. The base rack configuration is designed to provide a substantial foundation, starting with a significant number of OCPUs and expandable storage capacity. The GPU expansion rack, with each node equipped with four L40S GPUs, aims to provide the necessary horsepower for computationally intensive workloads. The platform also features high speed networking with 800 Gbps connectivity, designed to facilitate seamless data transfer and communication between nodes. Each GPU node is designed to include a substantial amount of high-speed memory and network bandwidth. Power consumption is a key consideration, with each GPU compute node potentially drawing a significant amount of power. Put simply, customers can start small and grow over a four year commitment period.
Oracle has designed these offerings to integrate seamlessly with its existing ecosystem. The platform is compatible with Oracle Exadata, enabling enhanced vector search and retrieval augmented generation (RAG) workloads. Oracle made a point to stress this heavily in the analyst briefing so expect to see more about this in coming releases, and rightly so.This integration aims to streamline the development and deployment of AI-powered applications that leverage the power of Oracle Database. Moreover, the platform is designed to be compatible with applications developed for Oracle Cloud Infrastructure (OCI), ensuring workload portability and reducing the complexities of hybrid cloud deployments.
The announcement highlights several key use cases for the new GPU-accelerated platform. These include AI/ML workloads such as real-time fraud detection, AI model inferencing, and feature extraction. The platform is also designed to be well suited for sectors like financial services, telecom, retail, and healthcare, where high computational power is often required without the need for cloud data offloading. Oracle emphasizes the importance of data sovereignty, arguing that on-premises deployment ensures compliance with data residency and regulatory requirements. I am particularly interested to see an opinionated architectural template for Cerner, the healthcare patient record system Oracle acquired a few years back, as this would make sense for large hospitals and healthcare providers.
One thing not mentioned in the launch material is that Oracle has offered half rack configurations in the past for Exadata and when I pressed Oracle they did acknowledge that upon request they would be able to accommodate Exadata and Cloud@Customer configurations in a single rack. Why this is not being positioned as an option for the smallest of clients is something I will push on, as it makes sense for those smaller deployments where data sovereignty augmented by AI is key.
Looking Ahead
The integration of NVIDIA L40S GPUs into Oracle's on-premises offerings represents a significant step in the evolution of hybrid cloud computing and one that makes perfect sense given the data sovereignty concerns of some enterprises looking to deploy LLMs. It directly addresses the needs of organizations that require high performance computing capabilities while maintaining control over their data and infrastructure. Based on my analysis of the market, this move could resonate strongly with industries facing increasing regulatory scrutiny and those with highly sensitive data or certain geographies with sovereign regulatory frameworks. The key trend that I am going to be looking out for is how customers leverage this platform to develop and deploy cutting-edge AI applications, particularly in areas like generative AI and real-time analytics and crucially how they couple these to Exadata solutions.
When you look at the market as a whole, the announcement signals a growing recognition that a one-size-fits-all approach to cloud computing is no longer sufficient. Organizations are increasingly demanding greater flexibility and choice in how they deploy and manage their workloads. Oracle's approach, which emphasizes hybrid cloud and on-premises solutions, is a testament to this trend - both with Cloud@Customer and Exadata. When you couple this approach with consistent pricing approaches across the on-premises offerings and OCI the customer has optionality, without restrictions.
Going forward I am going to be closely monitoring how Oracle performs on delivering a seamless and integrated experience across its cloud and on-premises offerings. Based on what I am observing, the success of this strategy will depend on several factors, including the ease of use of the platform, the cost effectiveness of the solution, and the strength of the use cases (specifically for its suite of software offerings like NetSuite and Cerner) that Oracle can build around it. It will be interesting to see how competitors such as HPE with its Private Cloud AI, and the likes of Lenovo and Dell respond to this move and whether it triggers further innovation in the hybrid cloud space.
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.