While some fast-growing technologies like artificial intelligence (AI) have been garnering much media attention of late – and we are always reading about concepts such as Big Data analytics – you may not realize that such computing capabilities are now available cost-effectively by utilizing the technology found in computer gaming consoles. Complex computing applications in science and engineering, such as fluid dynamics and 3D modeling, can now be processed more quickly using Supermicro GPU Servers – an affordable hardware.
Not long ago, complex mathematical modeling, simulation, and large-scale data manipulation required a massively expensive supercomputer. Today, the humble graphical processor aka GPU Solutions we are familiar with from real-time computer games is being increasingly used to perform much more complex tasks.
Graphical processor units (GPUs) were initially developed as an efficient way to manipulate computer graphics. Unlike general-purpose CPU’s, GPU has a highly parallel processor structure boosting its efficiency to handle complex algorithms which require processing of large volume of data parallel. High-performance GPU cards like NVidia GTX 1080, Tesla or Titan V can be installed in a rack-mount server, creating a server for compute-intensive applications.
The power of GPU servers
It is becoming increasingly common to use high computing GPU servers to run individual compute kernels, turning the parallel computational power of a modern graphics accelerator into general-purpose computing power, as opposed to being used solely to do graphical operations. For math-intensive or high-power operations or calculations or for data which require massive graphics operations, high-performance GPU computing servers are your ultimate go-to option. Using GPU Solution it is now possible for business enterprises, universities and research institutions to realize what would have been termed ‘supercomputing’ capabilities only a few years ago at a fraction of the cost of historical supercomputers.
Because GPUs offer the ability to simultaneously perform large numbers of parallel tasks, it makes them suitable for any high-complexity computing application involving the manipulation of large amounts of data. In contrast, conventional CPUs are better at sequential processing involving branching and decision processes. Applications that may have taken days to run on a conventional CPU-based system, when re-coded to take advantage of one or more GPUs, can have their runtime reduced to minutes.
For example, a typical Intel Core i7-6700K (Skylake) CPU has four cores and is capable of around 200 GFLOPs/s (floating-point operations per second). In contrast, a commodity GPU such as the NVidia GTX1080 has 2560 cores and can perform at well over 5000 GFLOPs/s. Installing multiple GPU cards in a GPU server increases the processing power proportionally further.
Applications for GPUs
Some common applications that are best performed on GPU architectures can be found in scientific research, engineering simulations, image processing and ‘big data’ analytics.
Science and engineering
Computational fluid dynamics (CFD), molecular dynamics, genome sequencing, mechanical simulation, electromagnetic field interactions, quantum electrodynamics, and digital twin modeling of complex engineered systems.
Registration, interpolation, feature detection, recognition, filtering
Databases, sorting and searching, data mining, machine learning, artificial intelligence (AI).
Example of applications that benefit well from parallel GPU architectures include:
- ANSYS (engineering design and simulation)
- Fluent (fluid dynamics)
- ANSYS Maxwell (electromagnetic field simulation)
- NVidia CUDA (general purpose programming)
- Tensorflow (machine learning)
- AutoCAD (3D design)
Digicor offers a range of rack systems from Supermicro that can be configured as GPU servers with commodity GPU cards.
- Supermicro SYS-1029GQ-TRT: Rack-mount 1 RU server supporting up to four GPU cards.
- Supermicro SYS-7049GP-TRT:Standalone (tower) server supporting up to six GPU cards.
- Supermicro SYS-4029GP-TRT: Rack-mount 4 RU server supporting up to eight GPU cards.
The ability to run consumer or commodity GPUs allows the end user to balance cost and computing power as needed, and to customize the system to their needs – whether that be scientific research, engineering design, and simulation, or business data analytics.