I'm all too wary of new industry hype. Anyone know the deal with these
Technical Director, *ia Center for Computer Music
Programmer / Analyst, Dean's Office (School of A&S)
[SNIP]Quote:>I'm all too wary of new industry hype. Anyone know the deal with these
1. "Parity" is not a type of RAM; it's an arrangement that can be implemented
in any type of RAM system. For every byte of memory, and extra bit is added
that is computed based on the data as it is stored. When the data byte is
read the parity is re-calculated and compared to the parity bit that is stored
at the matching location. A mis-match is a parity error and is assumed to mean
that one or more data bits of the addressed memory location have failed.
2. "EDO" is an enhanced version of standard "FastPage" mode DRAM chips.
These chips can continue to provide the data from one memory cycle while the
next cycle is beginning. Systems designed to take advantage of EDO RAM can
run a little faster.
3. "SDRAM" is Synchronous DRAM. This is a newer type of DRAM that requires
very special design considerations but the benefit is extreme speed. Standard
EDO DRAMs can be accessed in 60 or 70 nano seconds (billionths of a second).
SDRAMS can be accessed in as little as 10 nano seconds.
Hope that helps.
> I'm all too wary of new industry hype. Anyone know the deal with these
> memory types?
EDO RAM - Extended Data Out, is faster than normal FPM RAM
SDRAM - latest, very fast RAM.
Michael Meissner, Cygnus Solutions (Massachusetts office)
4th floor, 955 Massachusetts Avenue, Cambridge, MA 02139, USA
: >3. "SDRAM" is Synchronous DRAM. This is a newer type of DRAM that requires
: > very special design considerations but the benefit is extreme speed. Standard
: > EDO DRAMs can be accessed in 60 or 70 nano seconds (billionths of a second).
: > SDRAMS can be accessed in as little as 10 nano seconds.
: Hmm, well - if you know don't assume that SDRAM is 6-7 times faster as EDO RAM
: then this is fair enough.
: Nevertheless the timing specs refer to two different things with SDRAM and EDO
: ram, so that the real speed advantage of SDRAM is just about 10%.
all dram delivers data at approximately the same _latency_, around 75 ns.
fpm, edo and sdram differ mainly in how quickly they deliver data that
consecutively follow the initial word. fpm takes 3, edo takes 2, and sdram
takes one clock per trailing word in a burst: timings are normally quoted
as x333,x222,x111, where x is the initial-word latency, and ranges from
5-8 on a 66 Mhz clock. assuming x=5, they take 14,11,8 cycles to complete
the burst. this is clearly more than a 10% advantage for sdram.
of course, that's assuming you're only interested in latency or bandwidth
of cache misses. most apps see a cache hit rate of 50-90%, which means
that the advantage for sdram is negligable if the hit-rate is high.
In the future, we all should try to answer questions with SOME attention
to the questioner. Had he wanted to know more, he might have asked,
but for all we know, he wants to upgrade a particular motherboard that
won't even take advantage of EDO ram, to say nothing of SDRAM.
the flaming that's been going on in comp.sys.intel (and elsewhere,
most likely) wrt EDO vs FP ram and async vs burst-mode caches reminds
me of a similar debate that happened when sun released their SS10
systems with L2 caches -- there was some question whether or not
systems with the L2 cache actually ran _slower_ for memory-intensive
to resolve the issue, i crocked up a little piece of code ("cbash")
that can be used (with a bit of user effort) to measure miss latencies
in the memory hierarchy. cbash was originally targeted for sunos
platforms, but it also runs fine under linux.
is there anybody out there (perhaps a vendor) that can assemble a test
platform running linux into which they can swap various flavors of
memory and cache, then run a few tests? (if i remember correctly,
venkat did this recently with some windows apps?) if so and you've got
the urge, cbash can be snarfed via anonymous ftp from cag.lcs.mit.edu
in /pub/tuna/cbash-12aug93.tar.Z. give it a run, and let us all know
hopefully adding something constructive to the flaming,
http://cag-www.lcs.mit.edu/~tuna/ At the wheel of her Shark-de-Ville
10. EDO vs SDRAM
11. SDRAM vs EDO