Intel’s Xeon Phi Coprocessor has made quite a name for itself in the High Performance Computing sector. This is because while many supercomputers use GPGPUS (General Purpose GPUs) such as NVIDIA Tesla which rely on a CPU and special code to operate the Xeon Phi requires neither. We have just received details of the next generation of the Xeon Phi, which is codenamed “Knights Landing”.
The new generation of the Xeon Phi will roughly triple the performance of the previous generation. The Knights Landing Xeon Phi will be made using the Broadwell Architecture and 14nm fab process. The card will appear in two form factors this time. You will have the traditional PCI-Express form factor, which the current Xeon Phi is based off of and you will have a LGA socket form factor, which will apparently cut costs significantly.
The new Xeon Phi will churn out roughly 3 Tera Flops of double precision. Don’t think that’s a lot? Well NVIDIA’s latest Tesla K40 only manages 1.33 TF and AMD’s offering only does 1.48 TF. And of course your have to remember that the Xeon Phi does not require a CPU for operation. The current generation Xeon Phi features 61 cores, this will be moved up to 72 cores with Knights Landing. It will also feature AVX 512 and 384 GB of bandwidth.