We had some minor collaboration with Intel, went quite deep into SYCL/dpcpp a few years ago, even bought some fpga hardware.
Constantly ran into showstopping bugs in the compiler chain, that led us to think not many people were actually using in anger. Some of the bugs weren’t present on their devcloud, we assume they had livepatched some of them out.
And every few months they decide to reorganise or rename oneAPI, and every time they do this all the documentation URL change so it’s impossible to find old docs.
Then they deprecated the card we bought the month after, promised ongoing support, but the only available oneapi required RHEL8, whereas the drivers for the card only supported RHEL7. They removed the docs saying how to resolve this, and they didn’t work any more anyway if you had saved. And you couldn’t get old oneAPI without a paid support contract (and that had the bugs that we had proven to them anyway).
Long story short, will never, ever, ever touch anything INTEL software ever again.
We had some minor collaboration with Intel, went quite deep into SYCL/dpcpp a few years ago, even bought some fpga hardware.
Constantly ran into showstopping bugs in the compiler chain, that led us to think not many people were actually using in anger. Some of the bugs weren’t present on their devcloud, we assume they had livepatched some of them out.
And every few months they decide to reorganise or rename oneAPI, and every time they do this all the documentation URL change so it’s impossible to find old docs.
Then they deprecated the card we bought the month after, promised ongoing support, but the only available oneapi required RHEL8, whereas the drivers for the card only supported RHEL7. They removed the docs saying how to resolve this, and they didn’t work any more anyway if you had saved. And you couldn’t get old oneAPI without a paid support contract (and that had the bugs that we had proven to them anyway).
Long story short, will never, ever, ever touch anything INTEL software ever again.