With the opening of Siggraph, Nvidia took aim at high-end of workstations with a refresh to the company’s relatively new Maximus platform. Just six months after the first shipments of the Maximus solutions, which combine a Quadro graphics card with a Tesla HPC graphics card in a single workstation, Nvidia has announced the Maximus 2.
The Maximus 2 combines a Quadro K5000 with a Tesla K20, both of which are new cards using the latest Kepler architecture and TSMCs 28nm process technology. The result is essentially a GPU accelerator with a GPU accelerator or in other terms, a high-performance computing (HPC) solution in a workstation.
The target applications for the Maximus platform are the most demanding design and simulation applications ranging from CAD/CAM design at the low-end to scientific modeling at the high-end. Although multiple Quadro GPUs can be used to accomplish a similar results, Nvidia believes that Maximus platform optimizes the solution by using different GPUs with different performance characteristics.
However, Maximus is not limited to just one Quadro and one Tesla GPU. Different configurations can be used for different workloads. Note that like other workstation GPU platforms, Maximus is designed to use the PCI Gen3 interface between the different CPU cards, which may be a limiting factor for certain multi-card configurations but is not an issue for the base one-to-one card configuration.
Although the first generation only has seven certified applications, Maximus 2 already has more than 22, a number that is likely to increase further before general availability of Maximus-based platforms.
In platforms using certified applications, which are optimized for GPU acceleration as well as other features of the GPU, the computational engine is broken down into three elements: the CPU for general system control, the Quadro for the visualization and design, and the Tesla for the parallel compute engine. This allows for design and simulation on the same system and often at or close to real-time. This provides a huge benefit to design and modeling applications that once took multiple systems and days to weeks to complete.
This rapid uptake of Maximus and certified applications further demonstrates the shift to GPU acceleration by the software industry, especially for high-performance design and modeling applications.
The shift to GPU acceleration also highlights that the GPU resources are much more important than the CPU, an argument Nvidia has made for several years, in certain applications. The Maximus platform also highlights that even with GPU acceleration, no single architecture is sufficient for all applications and heterogeneous solutions are required to meet the diverging demands of the market.
With a suggested retail price of the two cards at $5,349 ($2,249 for the Quadro K5000 and $3,100 for the Tesla K20), the performance comes at a price. However, those interested in the Maximus 2 platform typically place an even higher value of resource productivity. Unfortunately, they will still have to wait a bit longer for Maximus 2. The release of Quadro K5000 is scheduled for October and the Tesla K20 is scheduled for December.