On 05/01/17 18:50, Luke Kenneth Casson Leighton wrote:
Why would they want to cripple their product by restricting themselves to the set of interfaces Luke has chosen.
?? peter!!
Intel operates under a totally different set of constraints from Luke. If Luke wants to make a successor to his compute cards he needs to find a new SoC that has the right set of interfaces. If Intel wants to make a successor to their compute cards they can ensure that one of their upcoming SoCs has the right set of interfaces.
which are, in your opinion, the "right set of interfaces"? serious question. if you're going to make such comments, you'd better be prepared to back them up and be prepared to justify them with a *REALLY* thorough analysis.
If you look through the history of this list you will find the evolution of EOMA68 is a battle to find a compromise between
1. Interfaces that are useful. 2. Interfaces that are ubiquitous on SoCs today 3. Interfaces that are likely to be ubiquitous on SoCs tomorrow. 4. Interfaces that fit within the pins of a pre-existing economical connector.
Intel doesn't have to worry nearly as much about 2 through 4 as you do. They have no reason to make it easy for competitors to make compatible products. They can ensure that their own future SoCs retain the Interfaces previous ones had. They think and work on a scale where custom connectors are an economical option.
Of course this also means they have a much higher threshold of success. A product line with hundreds of thousands of sales would be a big success for someone like you but would likely be considered a flop for them.
If I was in their place I would be including PCIe, SATA and Ethernet (likely in some kind of MII form so the card isn't burdened with the cost of a transceiver).