So, I work in a university research center and we have a small cluster used for parallel processing. It consists of ~50 supermicro blades running a combination of Xeon 5600, 5400, and E5 processors. Our current setup uses an entire rack and a half, and a ton of power and creates a lot of heat. With all the talk of the Xeon Phi lately, I am thinking that it might be a way to save costs or at least be able to scale much better than we are.
Our cluster uses MPI for spreading the load and my understanding is that Phi uses the same thing. A lot of our researchers use older code so GPU accelerated processing is out because it would all have to be re-written. From what I've read if we were to migrate from our blades to the Phi, it looks like our code should work fine. Is this correct?
I'd love to hear someone that uses the Phi in real-world scenarios. Thanks!
Our cluster uses MPI for spreading the load and my understanding is that Phi uses the same thing. A lot of our researchers use older code so GPU accelerated processing is out because it would all have to be re-written. From what I've read if we were to migrate from our blades to the Phi, it looks like our code should work fine. Is this correct?
I'd love to hear someone that uses the Phi in real-world scenarios. Thanks!