by piotr5 » Wed Nov 04, 2015 9:15 pm
afaik adapteva has no chip-factory of its own, it needs to use other companies for production. in order to cram 1024 cores into the same space that previously was occupied by 16 cores you need to create smaller electronics. it's well known, intel touches the 10nm technology, epiphany is using nano-sized electronics which has about 5 times that size. do the calculations: 4x4 cores times 5x5, 400 cores is easily possible if they'd get development costs paid and access to intel's chip-factories. how is 1024 cores even possible? of course the chip could get 2-3 layers or trice the surface-area, lots of research. so you can see how this goal is quite unattainable till parallella rises in popularity. for now adapteva is a 1man company. the low margin it makes with selling parallellas does not allow for more employees. no employees means no research. not to speak of testing and developing. if a loan would be taken for all that, or investors would borrow money, this means the next product will need to cost so much that buying an according number of current parallellas would end up with nearly the same price-tag. would it be different, sales of current parallella would drop. in other words, if/when there is a new version, I expect it to cost nearly 6400$ for the 1024 cores. or maybe half that price, but not lower than that. who'd even want to buy that? open hardware is nice and all that, but the machines to produce it are by far not open, yet! for a new company it is quite impossible to offer the low prices of intel and such.
a rant about kickstarter: on crowd-funding the usual procedure is to first give away the hardware for production-cost, then sell them for a slightly higher cost. if you take a loan from a bank, the usual procedure is to sell for extremely high price to maximize the sales of both, the new product and the products you are known for and which are the reason your bank did give you a loan in the first place. only when the loan is paid you can half that price. and only when the old product has lost its appeal you can set price to what the old thing did cost. why are there even those 2 procedures? true, crowd-funders do take a risk, even if they get what they paid for there is no guarantee it'll be bug-free. on the other hand same risk you take when you buy intel's newest invention at a time where the price is still high. before there were 2 kinds of customers, rich ones who did take risk and high price, basically serving as beta-testers, and the wide masses who buy stuff the rich ones did throw away. through kickstarter we have a huge price-dumping in that area, now rich ones buy and sell at a low price, it simply doesn't make sense anymore to buy new stuff, better buy used hardware. it's well-tested, bugs are fixed through software, thanks to death of moor's law it isn't much slower than the new stuff anyway. now think what this means for a company like adapteva...
to be more to the point: if 1024 cores is what you want, you need to implement software which simulates parallella on various hardware and convince programmers to write their software for this emulator instead for the individual native platforms! at least that's what amd does. this way you lock programmers to the parallella platform, users obtain a well-tested platform and can switch over from their old hardware to epiphany 1024 once it's out. therefore investors who foresee such an opportunity will give adapteva the money and social-networking required for making it real. alternatively wait till a big chip-company buys adapteva and implements all that software development on its own. sadly such companies are too self-indulgent to accept any other technology than their own...