Intel and NVIDIA present their offers for massively parallel computing
If there is anything that is giving a lot of talk in recent years is the massive parallel computing and large each year chipmakers have products with greater processing power. In the last days of the event GPU Technology Conference (GTC) have survived many commercials which highlighted the intel and Nvidia. From The Register meet the above by Intel and Nvidia.
Nvidia GPUs think the solutions will remain dominant in the massive parallel computing, but admits that the adoption of massively parallel computing has been hampered by the relatively complex APIs GPGPU CUDA and OpenCL, so bet it all on your new API OpenACC , which was developed in conjunction with CAPS, Cray and PGI, and promises to be much easier to learn than OpenCL or CUDA, and soon plan to further facilitate the work of programmers through the merger of OpenACC in OpenMP .
Intel has focused its efforts on bringing all segments x86, and so far its Xeon product line but remains a very successful product, in recent times has been overtaken in some situations by the GPU, but Intel now says its new micro-architecture MIC (Many Integrated Core) represents the future of massively parallel computing, with products like Knights Ferry and Knights Corner owning more than 50 x86 cores, which together with SIMD units combine the best of both worlds (CPU + GPU), offering over 1 Teraflop of computing power in double-precision calculations (Intel still does not reveal the exact figure).
Intel sends a not so subtle nod to the GPUs to mention that adapt the existing x86 code to mic is as simple as going to the last line of code and add a couple of lines of directives, which will take a few minutes to be performed; while migrating the code to the GPU is a long project.
Both Nvidia Tesla knights corner K20 and Intel battle for supremacy in massively parallel computing, but many analysts believe that putting aside the ease of programming, perhaps most importantly, which one offers the best power / W, data that Nvidia reported to be between 250 to 300W, while the TDP of Intel Knights Corner is still a mystery, although it is presumed that the manufacturing process to 20nm Tri-Gate you the advantage in this regard.
As for AMD, yet does not provide any information about future cGPUs FireStream architecture based Core Graphics Next (the one used in its current Radeon HD 7000 Series GPUs), so on the basis of 0,947 Teraflops of power calculations twice precision, could give competition to Intel and Nvidia products to be launched later this year.
ARM: A new challenge for x86
The cGPUs are not the only threat to the reign of x86 in the segment of massively parallel computing, since arm has also reached the server segment with its new architecture to 64 bits armv8 in combination with cGPUs, promises to revolutionize the segment HPC (High Performance Computing) offering a much better ratio of consumption per watt than the x86.
Nvidia is currently the only company that has announced a future product based on such combination, which for now is known by the codename Denver, which will be launched between 2014 to 2015, but is only a matter of time before other companies announce similar solutions.
Link: Core Wars: Inside Intel’s Power Struggle with NVIDIA (The Register)Tags: amd, arm, armv8, computing, double precision, FP64, GCN, HPC, intel, K20, kepler, knights corner, Larrabee, mass, mic, Next Core Graphics, Nvidia, parallel, Tesla