"James" said:
Need to upgrade the memory on my p4p800se. This is what I have now ..
Memory Module(s)
Memory Module 1 : Micron 256MB 16x(16Mx8) DDR-SDRAM PC2100U-2533-750 (CL2.5
up to 133MHz) (CL2 up to 100MHz)
Memory Module 3 : Nanya NT128D64S88A0G-7K 128MB 8x(16Mx8) DDR-SDRAM
PC2100U-233-750 (CL2.5 up to 143MHz) (CL2 up to 133MHz)
I'm thinking : 3 sticks of 256 pc2100 to obtain "dual channel" operation.
Would this work or would I be better off just starting fresh?
I would like to < $100 - $150 if possible.
thanks , james
Well, what you buy depends on what problem you are fixing.
The first priority is having enough memory to prevent the
OS from swapping out some programs, to make room for the
others that the user runs. Swapping to disk will make a
machine perform like a slug, if you try to run too many
programs simultaneously on a machine with too little RAM.
The processor to memory subsystem path is decoupled, so if
you use a slow memory configuration, the processor will still
run and compute correctly. If the programs you run are compute
bound (do a lot of calculating, but don't make a lot of random
references to different areas of memory), then speed wouldn't
be an issue. If you are running a program like Photoshop, it
likes both quantities of memory and speed in that memory.
In terms of speed, my measurement using memtest86 on a 865PE
based board, showed a 30% memory bandwidth improvement by going
dual channel. So, buying a 256MB 16x(16Mx8) to match your Micron
and buying a 128MB 8x(16Mx8) to match your Nanya, gets you the
30% more memory bandwidth, and a total of 768MB of ram (enough
for anything, except Photoshop with its "infinite" appetite).
The other dimension of the problem is basic clock speed on the
sticks of RAM. If you were to buy a pair of PC3200 DDR400 sticks,
the clock rate on the ram is 400/266=1.5 , for a 50% improvement
in clock rate (assuming a FSB800 processor). A lot of the accesses
to RAM are done in bursts, and the burst rate improves with the
higher clock speed.
With the budget you have listed, for the first option, of just
"dualling up" the memory, I can see a 256MB module costing
$25-38 US, and maybe slightly less than that for a 128MB module.
So, maybe $65 will buy you 30% more bandwidth.
If you throw away your current memory, and buy 2x256MB matching
PC3200 sticks, I can see some branded RAM at $40 for 256MB.
You could get two of those for $80. You could get four of those
for $160, or buy 2x512MB for $160. Buying enhanced CAS at this
point, would be a waste of money and put you way above budget.
CAS3 gets the job done just fine.
It is nice to have memory with a warranty on it, if you see
yourself using this computer for a few years. If you expect to
have more money to throw at computers in the near future, then
buying a memory with less of a warranty is another option.
Don't buy memory with unbranded chips on it (i.e. no company
name printed on the chips), because they are likely rejected
chips of unknown quality.
And deal with a company that does allow bad memory to be returned.
When you get the memory, test with memtest86 from memtest.org.
The test program is floppy based, and will allow you to determine
if the memory is error free or not. Memtest86 tests _all_ the
memory, and moves the program code out of the way, and test
underneath itself. Don't accept memory that cannot pass this
test, no matter how much of a "bargain" it was.
In terms of "feeling faster", I think you'll find 2x256 PC3200
makes a difference you can detect, when it is run in dual channel
mode. I could tell the difference between running in single
channel mode and dual channel mode, because when I bought the
865PE based computer, I could tell when opening windows that
it was too slow. The local company who sold it to me, put unmatched
DIMMs in it - they were both 256MB modules, but one module had
x16 chips and the other had x8 chips. It took me two tries before
they got it right :-( The only reason I bought a pre-built computer,
was for the warranty - the machine was a gift, and I didn't want
to be accused of giving a "Frankenstein" computer
HTH,
Paul
