Yitan NF5288 is a low-end high performance heterogeneous server which Inspur elaborately builds for its customers.
CPU+GPU heterogeneous computing architecture
Yitan NF5288 creatively introduces the GPU computing unit, which outperforms the traditional single computing unit. It adopts the collaboration of the Intel processor core of the latest architecture and NvidiaTelsa/Kepler accelerating computation. CPU and GPU fulfill their responsibilities respectively. CPU is mainly responsible for the logic choice, judge to jump and I/O communications, while the GPU is specifically responsible for the intensive and highly parallel computation. This makes the computing resources reasonably assigned, and the computing power adequately released. In this way, the computing performance improves sharply. Specially, the collaboration of computing accelerate architecture is not against traditional technical route, but a complement and support that shows the advantages of GPU and ultimately helps the user in certain applications by solving problems in much shorter times or increasing solution scale of problem.
Virtualization solutions with better performance
Compared with the traditional GPU virtualization solutions, the specifically designed NvidiaKepler architecture-based GRID K1 and K2 can present richer effects in a virtualization environment, and at the same time, realize the hardware virtualization of GPU, which means that multiple users may share a single GPU. This does not only improve the customer density but also provide real PC performance and compatibility. Grid card supports the GPU-based virtualization solutions of Citrix and Microsoft and allows people to choose among the broad solutions that have been proved by practice in a flexible manner.
Large-scale parallel computing processing core
Compared with the multi-core CPU providing less threads working together, GPU provides thousands of threads working concurrently. This enables the system to process more information flow. For example, in game and animation rendering, GPU computing is a special kind of computing tasks, which calculate each pixel with no need to (or using method to make it not need) consider the order, so if you have a million threads, you can render all the pixels at the same time, so that all pixels can be calculated within a pixel time.
High-speed I/O switch between processing units
Each device has the dedicated connection of its own, and GPU does not need to request system bus for bandwidth. Furthermore, the data transmission efficiency can be increased to a high value. Compared with traditional PCI bus which can only achieve a one-way transmission in a certain time circle, PCI-E with dual-simplex connection can provide higher transmission rates and quality. The difference between them is similar to the difference between half-duplex and full duplex. InspurYitian NF 5588M3 adopts PCI-E 3.0 x 16 with a bandwidth of 32Gb/s.
Adoption of the latest QPI technology of Intel
To take betteruse of the collaborative effect of CPU and GPU, InspurYitian desktop supercomputer adopts Intel QPI (Quick Path Interconnect) technology, which provides a transmission rate of 8.0GT/s and makes the communication faster. QPI bus achieves a direct interconnection within the multi-core processors, and during the multi-processor operation, each process can send data to each other with no need to go through the chipset, which significantly increases overall system performance. Sandybridge architecture processor with integrated memory controller, PCI-E 3.0 graphics interface as well as the emergence of core graphics will make a further play of QPI architecture capabilities.
Better programming environment
CUDA™(Compute Unified Device Architecture) enables the CPU to solve complicated computation. It consists of CUDA instruction set architecture (ISA) and parallel computation engine inside GPU.
Developers may write C programs for the CUDA™ architecture. Since C is one of the most popular advanced programming languages, the C-coded programs can operate with super high performance on the processor that supports CUDA™.
Inspur Yitian NF5288 is the latest high-density GPU server based on dual Intel Xeon E5-2600 V2 processor platform, featuring eight hot-swap 3.5” SATA HDD cages, 1+1 80+ Platinum 1620W redundant power supply and nine PCI-E Gen3 x16 expansion slots. It delivers high density computing power, scalable expansion capability, intelligent thermal solution and green design, making it an ideal choice for applications in the HPC field of life and medical science, engineering science, financial modeling and virtualization.