Accelerate your AI, ML, & DL workloads
with lightning-fast data storage processing
- Eliminates any compromise between performance and practicality
- Allows GPU-optimized servers to access scalable, high-performance NVMe flash storage pools as if they were local flash
- Ensures efficient use of both the GPUs themselves and the associated NVMe SSDs
- Results in higher ROI, easier workflow management and faster time to results
NVIDIA’s DGX + Excelero’s NVMesh
The perfect combination
The problem
Since GPUs can process huge amounts of data, AI storage solutions need an efficient pipeline of data to GPU memory.
- Preloading the data to GPU memory, even with pipelining, can only partially solve the problem
- Passing the data through the CPU memory subsystem creates bottlenecks and inefficiencies, leading to underutilization of GPUs
The solution
The low-latency and high-IOPS/BW benefits of Excelero’s NVMesh, combined with the massive network connectivity of Nvidia’s DGX, enable our customers to maximize GPU utilisation with a distributed and linearly scalable architecture.
- Shared storage resources across multiple GPU servers
- External NVMe capacity with local performance
- No need to copy data locally
- Minimal latency
- Linear scaling of performance
- Datasets can be larger than what can fit inside the DGX
- Nvidia GDS + Excelero RDDA means CPU-bypass all the way
- Advanced machine learning data storage
Reduced training time
Scalable, disaggregated shared storage enables DGX platforms to work on huge data sets and reduces training time from weeks to days.
Keep GPUs fed and happy
NVIDIA’s DGX nodes have massive network connectivity, allowing them to ingest as much as 200GB/s of bandwidth via 4-8 x 100Gb ports, playing a key part in the solution.
AI Storage Solution: Direct GPU-to-drive path
Excelero’s patented Remote Direct Drive Access (RDDA) functionality within our NVMesh software-defined solution bypasses the storage server CPU, providing a direct path between the application and remote NVMe drives.
In dozens of deployments with extremely demanding workloads, the approach has dramatically reduced latency and enabled high throughput for IO intensive workloads.
When NVIDIA GDS and Excelero’s RDDA are combined, they provide a direct GPU-to-drive path for the most latency-sensitive applications.