I see tidbits about Intel NPUs, including NPU-4 48 TOPS claim:
https://www.anandtech.com/show/21425/intel-lunar-lake-architecture-deep-dive-lion-cove-xe2-and-npu4/4
I find the Intel website poorly organized (yes, I think I could do a better job orienting it for engineers than a website full of pictures, sound and fury, signifying nothing!).
I can find this $686 processor with 48 NPU TOPS (INT8), but it doesn't specify if it supports ECC so I assume not (if it doesn't support ECC, "Fogettaboutit!")
https://www.intel.com/content/www/us/en/products/sku/240961/intel-core-ultra-9-processor-288v-12m-cache-up-to-5-10-ghz/specifications.html
Note that Intel does support ECC on some cheaper processors. Yet the retail price is not listed, confirming that the website is disorganized ... what impression does that give customers, run away?:
https://www.intel.com/content/www/us/en/products/sku/242804/intel-core-3-processor-201e-12m-cache-up-to-4-80-ghz/specifications.html
After the sad shakedown booting out Gelsinger (who was at least trying to fix the problems inside Intel), I find it difficult to understand if Intel is giving up on AI?
Look at this page on the website that shows a graphic where tools support CPU and GPU but not NPU:
https://www.intel.com/content/www/us/en/developer/tools/oneapi/optimization-for-tensorflow.html#gs.kaieij
I would like to use Keras => Tensorflow-cpu? or something, and get good performance for both training as well as inference on a NPU. Should I forget about Intel support and move on?
