Is Intel withdrawing compiler support for KNC?

[roger567]

The changes section in the coming C++ 18.0 compiler says

"Support for the Intel® Xeon Phi™ x100 product family coprocessor (formerly code name Knights Corner) is removed in this release"

(https://software.intel.com/en-us/articles/intel-c-compiler-180-for-windows-release-notes-for-intel-parallel-studio-xe-2018#180_KNC)

So does that mean the cards become unusable if the C++ compiler is upgraded?

Surely Intel would not abandon customers in this way. Please tell me I am misunderstanding the text.

[Loc_N_Intel]

Intel® C++ Compiler 18.0 Pre-Release (Beta) supports Intel® Xeon Phi™ x200 Processor but it doesn't support Intel® Xeon Phi™ x100 Coprocessor (for example, you cannot use the -mmic flag in Linux to compiler for the  Intel® Xeon Phi™ x100 Coprocessor).

Let me check with the compiler team to see what is the plan for the release version.

Thank you.

[Loc_N_Intel]

Hi Roger,

I just get the answer. Unfortunately, you can only use Intel® C++ Compiler 17 to compile for Intel® Xeon Phi™ x100 Coprocessor. However, you can install both versions 17 and 18 on your machine if you want to use Intel® C++ Compiler 18 for something else. Thank you.

[roger567]

Thanks for looking into this Loc,

That is shocking - that Intel would abandon the x100 and its users so quickly.
x100 cards are still being sold new and as far as I am aware the x200 coprocessor cards are not widely available yet.

The timing is fortuitous for me: I was just about to expand Xeon Phi use. Now I shall stop all future development.

Clearly I will also not now consider moving to the x200 in case the same thing happens with the next series in a couple of years.

This is the best argument I have heard in favour of Tesla.

[JJK]

This is indeed quite shocking:

• the x200 card has been pulled from the market
• the x100 card is no longer supported after Intel C++ 2017

thus Intel no longer has a coprocessor card that is capable or running true c++17 code - unless someone adds full KNC support to the entire GNU C++ suite.

It looks like Intel is pulling out of the accelerator card business (as well as some other business areas, BTW).

What a pity...

As a side note: NVidia is also rapidly dropping support for older GPU cards - I've got some old M2070's lying around and pretty soon the next version of CUDA will drop support for them, I fear.

[Jeffrey_H_Intel]

Roger567 wrote:

That is shocking - that Intel would abandon the x100 and its users so quickly.

x100 cards are still being sold new and as far as I am aware the x200 coprocessor cards are not widely available yet.

The timing is fortuitous for me: I was just about to expand Xeon Phi use. Now I shall stop all future development.

Clearly I will also not now consider moving to the x200 in case the same thing happens with the next series in a couple of years.

This is the best argument I have heard in favour of Tesla.

Please do not confuse deprecation of support in the latest software release with abandonment of users.  It's not like the Intel 17 compilers have stopped working on KNC.  Five years is a pretty standard period of time to maintain support of a hardware product in the latest release of software.  Have you evaluated whether all the features of CUDA 8.0 are supported on the K20 (compute capability 3.5) architecture, which was released around the same time as KNC?