Mageia forum

Code: Select all

[root@localhost opencl_runtime_14.1_x64_4.4.0.117]# ./install-cpu.sh
In case of failure, refer to release notes at https://software.intel.com/en-us/vcsource/tools/opencl-sdk
warning: opencl-1.2-base-4.4.0.117-1.x86_64.rpm: Header V3 RSA/SHA1 Signature, key ID f789186f: NOKEY
Preparing...                          ################################# [100%]
Updating / installing...
   1:opencl-1.2-base-1:4.4.0.117-1    ################################# [ 50%]
   2:opencl-1.2-intel-cpu-1:4.4.0.117-################################# [100%]
/var/tmp/rpm-tmp.CDdUim: line 4: /intel/opencl-1.2-4.4.0.117/etc/intel64.icd.rpmnew: No such file or directory
warning: %post(opencl-1.2-intel-cpu-1:4.4.0.117-1.x86_64) scriptlet failed, exit status 1
Done.

Intel will *likely never* support linux OpenCL on its Core processors. You just to have to follow the gigaflops to understand why...

Intel sell a *lot* of Xeons into supercomputing centers, where Xeons are bought hundreds, sometimes thousands at a time. These centers tend to run OpenCL on linux. Which is great...only not. If you look at a Xeon that was released in any given quarter, and then look at a Core released the same quarter, the Xeon tends to be $100-200 more expensive. Which is fine - it can pull more gflops, right? No, not on OpenCL (with a driver). Go look at that Xeon again, its gigaflops. Now look at the Core...but this time, look at the glops on the GPU. Add that to the gflops from the CPU. The Core has more total gigaflops, right? OpenCL doesn't care if the CPU or the GPU is the thread - that's the point of heterogenous computing. And the point of why Intel won't support it on linux.

So, if Intel were to support Core CPU/iGPU's on linux, somebody would eventually figure out that they can save thousands of dollars, and get more performance for their 1000-node supercomputer if they built it on Core. So, don't expect to see a linux Core OpenCL driver any time soon...unfortunately Intel will protect their interests on this one.