The AI landscape is undergoing a transformative change driven by advancements in deep learning. A pivotal player in this transformation is PEAK:AiO, with its AI Data Server software. Leveraging HPE’s ProLiant DL380 Gen11 hardware, PEAK:AiO demonstrates a monumental leap in performance metrics, challenging the status quo of existing multi-node storage systems.
The AI landscape is undergoing a transformative change driven by advancements in deep learning. A pivotal player in this transformation is PEAK:AiO, with its AI Data Server software. Leveraging HPE’s ProLiant DL380 Gen11 hardware, PEAK:AiO demonstrates a monumental leap in performance metrics, challenging the status quo of existing multi-node storage systems.
A New Era of AI Dominance and Storage Challenges
As AI continues to dominate various industries, the demand for high-performance storage solutions escalates. Today’s AI-driven systems, especially those reliant on GPUs, require data throughput that far exceeds the capabilities of many legacy storage systems. This gap often results in the underutilization of costly GPU resources and extended project timelines.
The Impact of Storage on AI Performance
Storage performance is no longer a mere technical specification but a crucial factor in AI outcomes. Studies by Google and IBM underscore the dramatic improvements in AI model training and application performance achieved through high-performance storage. Delays as minimal as 1 millisecond in data access can significantly reduce the accuracy of AI models, particularly in image classification.
PEAK:AiO, designed specifically for AI, breaks away from traditional, expensive HPC and enterprise storage solutions. By optimizing for AI workloads and starting at capacities under 100TB, PEAK:AiO delivers top-tier performance at a fraction of the cost, effectively lowering the entry barrier for AI projects.
The PEAK:AiO AI Data Server transforms standard servers like the HPE ProLiant DL380 Gen11 into ultra-performance AI-NAS solutions. This approach combines the economics and stability of HPE hardware with the agility and performance required for demanding AI applications, all without locking users into proprietary storage hardware.
Unprecedented Performance Metrics
The PEAK:AiO AI Data Server achieved groundbreaking results in testing, with performance metrics usually seen in HPC multi-node storage systems. In RDMA NFS Performance tests with five clients, PEAK:PROTECT RAID6 sequential read and write speeds reached up to 118GB/sec and 18GB/sec, respectively, with 620K IOPS for random reads. Moreover, in NVMe-oF Performance tests, it achieved up to 9.5 million IOPS for random reads with a RAID5 configuration, a significant milestone for a single-node system.
GPUDirect®: Enhancing Data Movement and Access
The integration of NVIDIA® GPUDirect® technology with PEAK:AiO enhances data movement and access for NVIDIA data center GPUs. This setup allows direct reading and writing to and from GPU memory, bypassing the CPU and thus reducing latency and improving overall performance. This collaboration supports GPUDirect® for both RDMA NFS and NVMe-oF. Tests using the gdsio load generator tool demonstrated comparable bandwidth to traditional CPU data paths, highlighting the efficiency gains achievable through GPU direct data access.
Conclusion: A New Benchmark in AI Storage
PEAK:AiO has set a new benchmark in AI storage, delivering the performance of HPC multi-node storage in a single-node solution. The collaboration with HPE offers a compelling combination of performance, simplicity, and cost-effectiveness, crucial for the evolving demands of AI projects.
As AI continues to evolve, solutions like PEAK:AiO’s AI Data Server are crucial in addressing the storage challenges that come with AI’s rapid growth. Their innovative approach not only meets the current demands of AI applications but also sets the stage for future advancements in this dynamic field.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
