12 ms·
The value of adding vastly more memory in a server is obvious. However, could someone explain the problem that required solving by a team of 20 people?
by xirium 19y ago
The value of adding vastly more memory in a server is obvious. However, could someone explain the problem that required solving by a team of 20 people?
- edw519 19y ago19 to get Red Bull.
- wmf 19y agoChip design is hard. Measure twice, cut once.
- xirium 19y agoWe've got one article ( http://news.ycombinator.com/item?id=123263 http://news.ycombinator.com/item?id=123263 ) in which memory controllers are being brought on-die to increase speed. We got another article ( http://news.ycombinator.com/item?id=123158 http://news.ycombinator.com/item?id=123158 ) which describes adding another tier of memory controllers to increase throughput. The latter could be hard to design if you were retro-fitting a tree of pipelined memory controllers and you were concerned about interoperability. However article makes it look like an attempt to bodge extra address lines.
- wmf 19y agoI think it's more about extra chip selects and reducing bus loading. AFAIK, standard DDR2 memory controllers only support so many ranks on the bus, so MetaRAM introduced some muxing to make four or eight ranks look like two. It's a clever hack.