GPU computing power cluster
this depends on your computing application hardware. If computing is purely dependent on CPU or GPU computing power
for the computing power of CPU, it is the superposition of pure CPU's double precision computing power, and it is the same for GPU
in addition, we should consider whether your computing software supports GPU computing power. If a pure CPU costs 1 Pb, then the budget is much larger. GPU is different. A card such as titanx has a single precision computing power of 11T flops.
first of all, can CPU remove cache like GPU? no way. There are two key factors for GPU to get rid of cache: the particularity of data (high alignment, pipeline processing, not conforming to localization assumption, rarely writing back data), and high speed bus. For the latter problem, CPU is subject to the backward data bus standard, which can be changed in theory. For the former problem, it is very difficult to solve in theory. Because the CPU to provide versatility, it can not limit the type of processing data. That's why GPU can never replace CPU
secondly, can CPU add many cores? no way. First, the cache takes up the area. Secondly, the CPU needs to increase the complexity of each core in order to maintain cache consistency. In addition, in order to make better use of cache and deal with data that are not aligned and need a lot of write back, CPU needs complex optimization (branch prediction, out of order execution, and some vectorization instructions and long pipeline simulating GPU). Therefore, the complexity of a CPU core is much higher than that of GPU, and the cost is higher (not that the etching cost is high, but the complexity reces the film rate, so the final cost will be high). So CPU can't add core like GPU
as for the control ability, the current situation of GPU is worse than CPU, but it is not an essential problem. However, control like recursion is not suitable for highly aligned and pipeline processed data, which is essentially a data problem.
Cluster is divided into four parts: computing node, storage node, management node and cluster accessories. Computing node is the node responsible for computing. In terms of GPU cluster, CPU and GPU card (or Phi card) are used for computing. Generally, only one hard disk is installed as the system disk. The computing node is connected to the storage node through Infiniband network (hereinafter referred to as IB network) to complete the reading and storage of computing data
in 2014, the mainstream speed of IB network is 40Gb / s, which translates into 8GB / s, that is, in theory, an 8GB USB flash disk full of data can be transmitted in one second. Storage node, as the name suggests, is a pile of hard disks to store data. Management node, as the name suggests, is responsible for management, generally a terminal. Cluster accessories include IB network (IB switch, wire, etc.), Gigabit Ethernet network (Gigabit Ethernet switch, wire, etc.), rack (cabinet), PDU (socket of advanced point), etc. OK, the parts are complete. Now how to assemble them? See the figure below:

GPU function
GPU display card brain determines that the performance of the graphics card file is different from that of 2D display card 3D display card, according to the 2D display chip processing 3D image special effects In order to rely on CPU processing power, the software accelerated 3D display chip is called 3D image special effect processing function set. The so-called hardware accelerated function display chip is in the display chip. The display card chip (PIN) in the current market adopts NVIDIA ATI's graphics processing chip
in 1999, NVIDIA company issued geforce 256 graphics processing chip. Firstly, the concept of GPU is proposed to rece the CPU dependence of the display card and carry out the original processing In particular, the core technology of GPU is hardware T & L, vertical environment material mapping vertex blending, texture compression bump mapping Dual texture four pixel 256 bit rendering engine and other hardware T & L technology, GPU logo
simply speaking, GPU can support T & L (transform and lighting edge conversion and light source processing) display chip T & L3D rendering important part, which is used to calculate the edge 3D position processing state light effect. T & L unit provides detailed 3D object advanced light effect; The operation of PCT & L is handled by CPU (so-called software T & L) Input response and other non 3D graphics processing work, the actual computing performance is discounted, and the computing speed of the graphics card waiting for CPU data is far behind the requirements of today's complex 3D games, even if the CPU working frequency exceeds 1GHz or higher, because the PC's own design problems have too much to do with CPU speed
about cpugpu related problems
question:
GPU competition is far more fierce than CPU competition, general-purpose pccpu Intel amd two times The two manufacturers can proce low-end procts Intel Although the procts of the first two manufacturers, such as 3S, can meet the needs of users, CPU manufacturers have not adopted the advanced GPU technology. CPU manufacturers have invested in their own proction lines, eliminated their original proction lines, and even failed to collect the initial investment. GPU manufacturers have designed their own procts, so they can only save their own equipment when they stop and upgrade their own proction lines The original
second problem
the CPU is used to process the data of game AI plot and other aspects, and each new DX released by Microsoft and each GPU can support the new features of DX. Some image tasks are performed by CPU, and some tasks are performed by GPU, which is more efficient than those performed by CPU before gravity
third problem
GPU is more efficient than CPU dedicated to image processing It's much better than cpucpu's general data processor in processing numerical calculation. It can complete tasks. GPU can replace CPU.
in addition,
amd has acquired a graphics card chip designer, AMD. Now cpugpu will take a fusion road to compete. How can cpugpu cooperate with each other to improve work efficiency? AMD is considering Intel's problem
the fourth problem
Microsoft releases Windows 7 Combining with gpucpu to enhance the use value of GPU hardware Windows 7 CPU and GPU group collaborative processing environment CPU computing non complex sequence code GPU runs scale parallel applications Microsoft uses DirectX compute GPU as the core of operating system DirectX compute enables developers to use the scale parallel computing power of GPU to create and introce consumer level professional computing applications Microsoft issued GPGPU universal computing interface to unify GPU universal computing standard, saying that Windows 7 GPU hardware only plays a more effective role in CPU
