tag:blogger.com,1999:blog-8891277981056105369.post3204696560547252164..comments2023-12-24T18:24:10.671-05:00Comments on Dr. Codd Was Right: Game OverRobert Younghttp://www.blogger.com/profile/09056808374481236610noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-8891277981056105369.post-18201643192332338672016-09-06T13:26:48.635-04:002016-09-06T13:26:48.635-04:00hybrid memory has been talked about for some time....hybrid memory has been talked about for some time. Intel/Micron 3DXPoint (here: http://www.simmtester.com/page/news/shownews.asp?num=17984) was touted as 'universal memory', but it's still too much slower than DRAM. time will tell what happens.Robert Younghttps://www.blogger.com/profile/09056808374481236610noreply@blogger.comtag:blogger.com,1999:blog-8891277981056105369.post-91976593054840721612016-09-06T12:57:08.221-04:002016-09-06T12:57:08.221-04:00I suppose it's also possible that memory chip ...I suppose it's also possible that memory chip makers will simply start grafting flash drive type chips onto the little cards they currently put DRAM (and some cache) on. At that point, it will get interesting seeing what ratios of DRAM and flash they start throwing together for various uses.Roboproghttps://www.blogger.com/profile/11304202067539670383noreply@blogger.comtag:blogger.com,1999:blog-8891277981056105369.post-24358273332522815712016-08-27T07:54:05.713-04:002016-08-27T07:54:05.713-04:00still not entirely clear, but my take on it is tha...still not entirely clear, but my take on it is that mmap() will use all 'memory' as one level. likely there'll be some driver in the OS to put most-recently-used in DRAM. we'll see. there seemed to be consensus that XPoint isn't consumer desktop driven. yet.Robert Younghttps://www.blogger.com/profile/09056808374481236610noreply@blogger.comtag:blogger.com,1999:blog-8891277981056105369.post-37444468369163044282016-08-26T23:24:42.545-04:002016-08-26T23:24:42.545-04:00So, it is a safe summary that this is largely like...So, it is a safe summary that this is largely like a traditional swap partition (or whatever OS-de-jure calls the virtual memory overflow area), but, the expectation is that the OS/driver code will know that random access works VERY well and that this is NOT spinning rust and try to schedule as if extra space had those latency restrictions???<br /><br />So if RAM (DRAM) is "L4", this becomes "L5" - random access with some kind of address bus hookup as opposed to a hard disk interface?Roboproghttps://www.blogger.com/profile/11304202067539670383noreply@blogger.com