Marco Peereboom wrote: > I really have a hard time buying this. I can see how you ended up with > some crap in that memory upon reboot but I fail to see how that memory > could retain its contents. Not knowing the situation you might have > had some huge caps on that machine; or even battery backed up ram. This > combined with low power mode content can be stored for days (we do that > on RAID cards). But that is also ram that does not require a clock to > retain its contents.
Buy it, really. Twenty+ years ago, I'd noticed this, having completely powered down my computer, decided I had something more to do, flipped the power switch right back on, and I was sitting at a command prompt. I marveled, I did it again, it worked again. I called my roommate over, he and I marveled over this...every few times we tried it, it would end up rebooting (or Hung). We knew that the rated refresh time was 2ms for the RAM, and yet, here we were with the machine CLEARLY OFF (the DC fan on this thing was deafening, no question when it had power and when it didn't) for over a second, and coming back up right where we had powered it down. Keep in mind, it was most likely the reset circuit that was causing the reboot, not fading of data in the DRAM chips. And yes, the flash of "old" screens on video cards was also a give-away that this was happening. And yes, these were all DRAM machines, I was VERY familiar with the circuits and designs of this machine, being this machine was documented like none are today (I'm glancing up at the boot ROM source code for the machine -- it came in printed form with the system!), and I had rebuilt the RAM system on the these machines a few dozen times (I worked out an upgrade for the things to go from 64k chips to 256k chips, allowing the "maximum" of 768k on-board, it was a popular upgrade at our store). Is the effect real? ZERO question in my mind. I'm amazed that it goes as long as people are saying, but in thinking about it, I'm not so surprised. Keep in mind, the difference in basic design between an EPROM and a dynamic RAM chip is just one of retention time and how the state is changed. EPROMs are rated for ten year retention and routinely hold for twice that, so I'm not too surprised that the gate of a CMOS transistor can hold a charge for a few seconds...and if that, why not tens of seconds. Heck, almost 30 years ago, people were popping the cap off 4k and 16k DRAM chips, using an 8mm movie camera lens to focus an image on the chip, charging all the cells, wait a while and then read all the data...the light would cause the cells to discharge faster, and you could get a crude, 1 bit, digital picture. It took a while even then for the charge to drain off the gates enough to see the image. The effect is hidden by RC circuits that fire off hardware resets and energy saving monitors that don't have a picture on the screen until the machine has started booting (and now, LCDs which have to "sync" to the image) and ROMs that clear screens and start the boot process before we notice that the data isn't gone yet. It's also a matter of numbers -- If you say a DRAM has one second refresh times, but every few months one bit may fade too fast someplace, that would be completely unacceptable for a "good" system. HOWEVER, if 99% of your data is still intact after ten seconds...you can probably get SOME interesting data off the thing. So, you design for the worst possible environment, and refresh your data every 2ms or more often...but that most certainly doesn't mean ALL the data is gone after 20ms..or 20 seconds. Based on what I've seen, the only part I'm having trouble with is someone probably just got their doctorate on something that I considered a pointless curiosity twenty years ago. It's still a mostly pointless curiosity, and I'm still lame at "working the system". But yes, if someone has access to your system enough to flood your system with liquid hydrocarbons and liquid nitrogen...you got bigger security problems than your memory not forgetting. Nick.
