GRIFO™ SOLVES THE PROBLEM OF SCALING THE NUMBER OF GPUs IN A SINGLE SYSTEM BY AGGREGATING MULTIPLE GPU BLADES INTO A SINGLE UNIFIED SYSTEM THAT ACTS AS SINGLE HOST WITH MULTIPLE GPUs, PROVIDING A NEW LEVEL OF PERFORMANCE TO ANY MULTI-GPU APPLICATION MAXIMIZING THE RETURN OF THE INVESTMENT.
GRIFO, THE HIGH-PERFORMANCE GPU COMPOSABLE PLATFORM THAT CHANGES THE WAY TO SCALE GPU POWER
The real composable GPU supercomputer
GPU to GPU Direct
Maximize performance in intensive GPU applications
GPU to NVMe Direct
The best GPU data access for the Maximum Application Performance
A real multi-GPU System
We are in a new age where AI and high-performance computing (HPC) are transforming our world, and new challenges are emerging that demand enormous computing resources to solve. Many applications are designed to be single host multiple GPUs, and even if they can benefit from 10s or even 100s of GPUs, they are limited by the maximum GPU number of 16 available in the high-end systems in the market. GRIFO™ solves the problem enabling single host multiple-GPU applications to scale from 20 to 100s GPUs.
While others focus on single nodes with many GPUs, A3Cube architects and delivers a new class of systems built for production-level HPC, AI and professional graphic such as large rendering applications that require scaling beyond the acceleration of a single node, without changing the application code.
Whether it’s deep learning training, signal processing, reservoir simulation, high-performance microscopy or medical image processing, the A3Cube GRIFO Box system is architected with scaling in mind.
The flexibility to choose any available model of GPU makes GRIFO™ the only solution that can meet the best price-application performance possible on the market.
IMMEDIATE BENEFIT TO YOUR INFRASTRUCTURE
GRIFO™ offers a wide range of GPU (graphics processing unit) computing platforms that are designed for High-Performance Computing (HPC), Artificial Intelligence, Machine Learning, 3D Rendering, medical imaging, GPU-Analytics (in combination with the DB-Turbo IO™ platform) and many other applications. GRIFO™ GPU platforms range from 20 GPUs to 80 GPUs inside a single-image partitionable rackmount system. Targeted to the right workload, GRIFO™ offers unique scalability features, high performance, GPU to GPU direct ultra-low latency communication, reduced rack space requirements, and lower power consumption when compared to traditional CPU-centric platforms by delivering outstanding performance in Machine Learning, 3D Rendering and modeling, Artificial Intelligence training, GPU-powered Databases and all the scalable single host multiple GPU workloads.
A3Cube GRIFO™ BOX THE COMPOSABLE GPU SYSTEM
The A3Cube Composable platform allows users to manage, scale up, merge and configure physical GPU bare-metal server systems.
Pools of computing resources, network, storage, and graphics processing elements can be deployed natively across the RONNIEE Scale™ (PCIe) fabric. Core system resources to be instantly interconnected into physical servers via RONNIEE Scale™ "in-memory" fabric and dynamically reconfigured as needed.
A powerful management software automates, orchestrates, and composes the resources across the systems permitting unique configuration flexibility.
GRIFO™ enables users to adapt and approach full utilization of the available resources to satisfy the performance associated with next-generation applications.