3

This maybe only theoretical, but are there any advantages of using one or another in terms of:

  1. ISA
  2. energy efficency

Solution with using QPI looks better for me. Are there any advantages of using PCIe-P2P Tesla against QPI Xeon Phi?

1

Many comparison tests and benchmarks published in internet say that Tesla is faster in scientific applications, but I'm not an expert in HPC and I can't tell if test procedures used in these comparisons are not favoring Nvidia some way. Especially comparison published by Nvidia :)

I think Nvidia CUDA is mature technology, and I think there is bigger community of CUDA developers, so using Tesla may speed up your project development if you are working on some software.

I don't know if Intel MKL is mature technology, but for me - this is something very exotic and I guess finding some help/support may be harder than for CUDA development. On the other hand - I believe you can get top quality customer support from Intel.

You have to also consider, that Nvidia is now in 28nm "age", Intel in 22nm (I mean products available on the market). So theoretically we can expect similar energy efficiency today, but Intel goes into 14nm technology soon, and that means next Phi series will be much more efficient.

| improve this answer | |
1

Tesla is faster for numeric parallel computing. However, you have to be aware that GPUs are limited to performing data-parallel calculations and are not suitable for processing a wide variety of different tasks due to a phenomenon called branch divergence. Consequently, comparing CUDA and Xeon Phi is like comparing apples to oranges. If you need data-parallel numeric computation, Tesla is still faster. If you need to run some parallel x86 program, you want Xeon Phi. (And yes, even though it's x86 you still have to recompile.)

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.