Rumors have it that the Intel Xeon Phi is finally walking out on the PCI-E and is soon going to be available with the usual LGA Socket Packaging. The coprocessor has been built around the MIC (Many Integrated Core) Architecture and can house 60 cores in a single package. These processors which have been used in the HPC and Supercomputing clusters in the GPU form factor, which had plenty of benefits to offer, will now be available with the LGA Socket.
Intel Xeon Phi is currently housing a GDDR5 memory spread over a 512 bit interface and generates a suitable bandwidth of 352GB per second. Where the GPU form factor brings along a lot of advantages, the movement towards the LGA socket will count for the loss of the GDDR5 memory and it is most likely that the GDDR4 memory will take its place. And as compensation to the loss of bandwidth which will eventually drop to 150GB per second, the memory will boost up to 256 GB which surely appears to be a good bargain.
Moving on, Knights Landing is the codename which has been given to the Intel MIC which is based on Haswell and Skylake. It was previously said that the Skylake refresh of Knights Landing would be equipped with PCI-E 4.0 and this would have given a noticeable boost to the performance of HPC but now Xeon Phi is being used in the HPC Clusters. To get a clear understanding of the coprocessor phenomenon, it is easy to consider a GPU as something given in the hands of thousands of monkeys and they are taught to perform the task at hand. The CPU can rightly be thought of as a project being dealt by ten intelligent scientists. This clearly states that in the HPC Clusters the GPU cannot be left to its own devices rather have a CPU guiding it hence showing that coprocessors are the better option for HPC. Unlike the GPUs, the Xeon Phi Coprocessors can give out performance without any guidance from the CPU and this becomes the greatest advantage.
Jul 31, 2014 0