First post, by TryAgain
I am interested in this question. Google search gives me a lot of ram disk stuff
which is close but not exactly what I need.
Old rigs are good for old stuff software. But everyone have had desire to
teach the old dog new tricks.
There is ancient solution to this eternal question- in essence- use virtual
memory, more and more, the better.
Back then, in old days and even now this spell lacks magic. It doesn't work.
It is SLOW at least...
But desire remains...feeding on this clue.
Most 486 class processors can use 4GB physical and 1TB virtual memory!
I found one guy, may be vogon member, which achieved this feat.
Basically, he run new program, x264 benchmark on i486-100 the latest and
greatest and won the last but in vogons point of view the first place in the
score list.
The benchmark took 8 months in waiting for the end year... He used server
class motherboard with 128MB of ram memory and... late SCSI disk with rare
controller with lot of cache.
He over ran and discussed a lot of problems to do that, but one of them he
hits and remains unanswered is this, Did use of paging hurt performance? He
claims it doesn't.
In fact, the performance wasn't goal at all, ... just the finish. The nature
of benchmark requires around 180 MB of ram.
So how to increase memory to around this amount easily?
The question is not new, nor original. For example it was discussed in vague here.
https://hardforum.com/threads/hard-drive-base … al-ssd.1114844/
Using SSD drive can show the real performance of such sort of solution, but
not quite so, they wear out.
In fact, good SSDs contain a lot of cache- 256MB are common, if only
companies producing them include in firmware of the drive a kind of jumper
which will resurect the drive as operated as dynamic ram disk,voila, that
will solve the problem that concern me.
I found this site
http://www.s100computers.com
It's very techy. But its encourage me to think that is possible.
The Micron Xpoint memory is advertising as game changer. In my opinion, it
will work as union of cache which is DRAM and Flash ram. So in this case,
cache match exactly flash. it's from now on never smaller. The long life come
from that that this cache is flushed on power off, so in server life this
kind of SSD is even more everlasting. In essence, one chip-two faces. Such
SSD will again solve the problem.
Another simple, but cost solution, buy 1TB SSD keep eye on cache in it. Then
create partition which match the cache. Firmware on disk for sure will never make
write operation, because cache is enough for maintain the data.
There exist 72 pins memory sticks which are of high capacity 64MB and 128 MB,
but they themselves have problems - other voltage operation, bios doesn't like
them and are not cacheable. Maximum extension on two slot memory 256 MB.
There are other solutions, multiplexing memory module on existing slot
through utilizing additional address pins of processor, but this broke
cache-ability of any size memory. This must be done for cache but remain
complex.
This can be done with adding an expansion card, but will require driver
similar to what disk controller needed.
So in view of last link to site given, It is more feasible if there is some
interface chip to ide port which address and read-write memory slot, for
example sdram stick which has 168 contact pins only. In such solution 512MB
will be attainable as one swap file. Such disk must register in
initialization with some letter and than maintain just one file in FAT32.
The essence is that cpu works serially and only with few memory
cells in a time. So paging mechanism take care of keeping the illusion that
there is contiguous line of ram available.
Please, someone with experience in hardware or low system software gives
value on this idea. For example, win98 operating system which can handle
no more than 512 MB. One 128MB sdram stick will cost a dollar.
The effect will be more pronounced on 486, because from any size memory
present on system the great leap will be four- five times.
Your comments follows.
