DDR2 versus Rambus

  • Thread starter Thread starter JohnMullins
  • Start date Start date
JohnMullins said:
What memory gives optimal performance with Pentium4 800 FSB?
Click to expand...

I believe that, if you knew what Rambus actually was, you wouldn't be asking
the question. RAMBUS, as it stands, is a dead technology and is also
horrendously expensive - 256MB Rambus DRAM will set you back around £100,
depending on where you buy it. The same amount of DDR2-PC4200 will set you
back around between £25 and £30. IOW, you could buy 1GB DDR2 for the price
of 256MB RDRAM. Now, which would you rather have: - more for less - or less
for more...?

I don't think I need to say any more, do I?
 
JohnMullins said:
What memory gives optimal performance with Pentium4 800 FSB?
Click to expand...
RAMBUS was, in my opinion, a flash in the pan. It was dead before it
even got off the line.

Wheaty
 
JohnMullins said:
What memory gives optimal performance with Pentium4 800 FSB?
Click to expand...

i actually found a new P-4 mobo with cpu
in a new case...out by someone trash.
at first i thought the machine was stripped down...then realized someone
was building one...
then threw it out...
it took RDRAM and once i priced it out, saw why they threw the machine
out...
however i've heard that machines that use it perform very well...
so i found some cheap RDRAM on ebay that no one had even bothered
to bid on...and in a few days will at least be able to try it out...

but if i was going to have to go out and buy components...
i'm sure i'd get a mobo that used DDR
 
philo said:
i actually found a new P-4 mobo with cpu
in a new case...out by someone trash.
at first i thought the machine was stripped down...then realized someone
was building one...
then threw it out...
it took RDRAM and once i priced it out, saw why they threw the machine
out...
however i've heard that machines that use it perform very well...
so i found some cheap RDRAM on ebay that no one had even bothered
to bid on...and in a few days will at least be able to try it out...

but if i was going to have to go out and buy components...
i'm sure i'd get a mobo that used DDR
Click to expand...

I have one machine that takes it that I bought used, and the reason that
I got this one versus the one that had 2 faster CPU's in it was that
this one had 2GB of RIMMs in it and the other one didn't. It was more
expensive to buy the RIMMS than it was to buy 2 Xeons to upgrade the
processors. That was a few years ago, but I can't imagine the prices
have gotten that much better. The memory was so expensive then that
they could have made more money parting this system out than selling it
complete. The ram alone would have fetched more than I paid for it.
When the first component dies in this system I will part it out and
hopefully someone will still need the parts.

In the case of the system I have which uses PC800, even a dual channel
DDR333/DDR400 box has more memory bandwidth than mine.(This machine was
spiff when it was new, but 5 months later they came out with faster
chips it couldn't support). Of course, if you have PC1200 in there,
maybe you can make a case for RDRAM, but I don't even want to think
about what those modules cost. It's pretty much moot though. I don't
think anyone is going to be selling stuff like that except to support
existing systems. RDRAM is dead.


From here on out it's pretty much going to be DDR and DDR-2, and
eventually they will get the latencies on DDR-2 worked out so. It took
them awhile to get it right for DDR as well.
 
While this may be true for WinRar (and i even happen to use
winrar quite a bit), it may not correspond to more common
uses of a system? I saw it was not the actual compression
but still wouldn't it be using similar access to what Winrar
uses?
Click to expand...

Well, data traveling thru mem.controller & RAM both ways is mostly
more or less compacted (OS & programs files & also user data mostly) &
WinRAR´s built_in benchmark emulates this even too good. This types of
data IMHO are more than half of all data used (more than 50% for sure)
when doing jobs on PC & chunks of data are much bigger than few years
ago (getting bigger & bigger!)

For example for bandwith test Sandra uses very short bursts of small
ammount of data & than calculates theoretical bandwith. This situation
in real life takes maybe 1% of data processed (almost never) & thats
is why is not suitable to test real life performance of ram with it!

It is similar like for CPU testing: you do not for example test a CPU
power with decoding mp3´s 2 wav (and play them) but ENCoding wav 2
mp3, which heavily stresses CPU & than measure time to do it.
(similar way does that WinRAR´s built_in test, but only for
mem.controller & ram!) ... very simple ... :-)

Similar situation is with AGP transfers: 8x against 2x speed
(bandwith-theoretical!) doesn´t mean 4x improvement in speed; in real
life is more about 10% in total! & thats because here we have also
more or less compacted data traveling thru interface!

it is like Ferrari versus a Mack Truck (bandwith against transport a
load or something...) stuff ... :-)
 
Well, data traveling thru mem.controller & RAM both ways is mostly
more or less compacted (OS & programs files & also user data mostly) &
WinRAR´s built_in benchmark emulates this even too good. This types of
data IMHO are more than half of all data used (more than 50% for sure)
when doing jobs on PC & chunks of data are much bigger than few years
ago (getting bigger & bigger!)

For example for bandwith test Sandra uses very short bursts of small
ammount of data & than calculates theoretical bandwith. This situation
in real life takes maybe 1% of data processed (almost never) & thats
is why is not suitable to test real life performance of ram with it!

It is similar like for CPU testing: you do not for example test a CPU
power with decoding mp3´s 2 wav (and play them) but ENCoding wav 2
mp3, which heavily stresses CPU & than measure time to do it.
(similar way does that WinRAR´s built_in test, but only for
mem.controller & ram!) ... very simple ... :-)

Similar situation is with AGP transfers: 8x against 2x speed
(bandwith-theoretical!) doesn´t mean 4x improvement in speed; in real
life is more about 10% in total! & thats because here we have also
more or less compacted data traveling thru interface!

it is like Ferrari versus a Mack Truck (bandwith against transport a
load or something...) stuff ... :-)
Click to expand...

Except I still suspect it isn't stressing latency in highly
random accesses rather than a more serialized data flow of
such large chunks. Linear Processing of a data file by an
application may be different than typical use of a system.
 
Except I still suspect it isn't stressing latency in highly
random accesses rather than a more serialized data flow of
such large chunks. Linear Processing of a data file by an
application may be different than typical use of a system.
Click to expand...

.... may be .., but IMHO not too much ..
 
I'd hesitate to go with DDR2. Get something that's tried and true. DDR

I never owned Rambus, too expensive!
 
Back
Top