I think the problem lies with a speedup trick used on some boards. With 8 megs
you have filled two bank of memory up. Thus you can store alternate bytes in alternate locations. So to access consecutive bytes you half the access time, by
jumping between memory units. When you have 16megs you only fill 4 slots which
stops you using the speedup trick.
SOLUTION SPLASH OUT ON 32MEGS!!!
: I have a 486SX running a 40MHz with 16MB of 60ns RAM. I've been
: hearing about some people experiencing slowdowns when upgrading from
: 8->16MB, so I was wondering if I am experiencing the same problem
: but just not know about it. Since I have a 16MB SIMM, I can't
: "downgrade" to 8MB and measure the performance difference. I'm
: getting about 20 bogomips(flops?) at startup. Is anyone getting
: significantly faster speeds with 8MB? Or is there an easy way for
: me to benchmark my computer? xbench won't do any good, since I'm
: using a Diamond Stealth 32 and tweaking it to get X to work.
: I found out that memory > 16MB was not being cached on my
: machine was when I noticed that a certain application program I was
: working on was taking significantly different amounts of time to
: execute every time I ran it.
: This variability in execution times, it turned out, was caused by
: the fact that when the binary happened to be loaded into memory < 16MB
: the instructions it contained were being cached but when it happened
: to be loaded into memory > 16MB it was not being cached.
: These variations in execution times were as large as a factor of 10
: in some instances.
: When I recompiled the kernel with 16MB_LIMIT enabled, the variation
: disappeared with no increase in the average execution time that I was
: able to detect.
: As soon as I get around to buying a second 128MB SRAM chip, I will
: take a look at what happens if I am running with 20MB ram, 256K cache,
: and 16MB_LIMIT disabled again.
: Bill Hogan