|
|
|
| EonStor GS unified storage solution now supports GPUDirect Storage technology, dramatically accelerating applications that demand extreme compute power and massive data throughput, such as AI training, HPC tasks, and real-time analytics. |
|
|
|
|
|
|
|
A Faster Route from Storage to GPU |
| In traditional architectures, the data I/O path moves from storage to the CPU, then to system memory, before finally reaching the GPU. |
|
| With support for NVMe over Fabric and NFS over RDMA, EonStor GS enables GPUDirect Storage, establishing a high-speed, low-latency data path that bypasses the CPU and system memory, enabling data to flow from GS storage to the server’s GPU. This allows GPUs to operate at maximum efficiency. |
|
 |
|
|
|
|
Benefits of GPUDirect Storage |
|
 |
Ultra-Low Latency |
|
Direct data transfers from storage to GPU minimize delays and speed up processing. |
|
|
 |
Higher Bandwidth |
|
Maximizes the utilization of storage bandwidth by bypassing the CPU, resulting in higher data transfer rates.
|
|
|
 |
Improved GPU Utilization |
|
GPUs spend less time waiting and more time computing, increasing efficiency. |
|
|
 |
Reduced Server’s CPU and Memory Overhead |
|
Prevents the CPU from becoming a bottleneck, freeing it for other system tasks. Lowers memory requirements and simplifies system configurations. |
|
|
|
|
|
|
|
EonStor GS Series with GPUDirect Storage |
|
|
|
|
|
Driving Efficiency Across Applications |
| Faster data loading accelerates demanding applications, where performance is usually limited by I/O speed, shortening job runtimes and processing cycles. |
|
 |
AI Training and Deep Learning
Accelerate model development by rapidly feeding large datasets to GPUs.
|
|
 |
High-Performance Computing
Boost performance for simulations, scientific analysis, and advanced computation.
|
|
 |
Real-Time Analytics
Instant data access and processing for finance, smart manufacturing, and logistics.
|
|
 |
Video Processing and Rendering
Faster rendering and transcoding workflows with quick read/write operations.
|
|
 |
Scientific Research
Ideal for large-scale data analysis like graph analytics.
|
|
|
|
|
| |
Unlock extreme compute power and throughput with GPUs fed directly by storage. |
|
| |
|
|
| |
|
|
|