I want to upgrade my 1GB System (P4C800-E Deluxe) without obsoleting the
existing memory.
I currently have Kingston ValueRam 1GB (2x512MB) DDR400(PC3200) Dual
Channel Unbuffered CAS Latency 3 (p/n KVR400X64C3AK2/1G).
These 2 sticks take up 2 of the 4 slots on the motherboard
In the remaining 2 slots, I want to add Kingston ValueRam 2GB(2x1G)
DDR400 (PC3200) Dual Channel Unbuffered CAS Latency 3 (p/n
KVR400X64C3AK2/2G)
This would upgrade my system to 3GB (and hopefully improve my
video/photo editing response times - this is a WinXP Pro system)
Is there any reason this configuration shouldn't work?
Any gotchas I should watch out for?
Any other suggestions? (one of my concerns is diminishing returns - I
can add 1GB of ram for $77, but the 2GB will cost me ~$200. Is 3GB
worth $120 more than 2GB? I do a lot of Photoshop/Showbiz
DVD/RawShooterEssentials editing, as well as a lot of multi-tasking)
Regards,
Bill
Intel wrote a memory guide for the 875P. It is here.
http://developer.intel.com/design/chipsets/applnots/252730.htm
2x512 + 2x1024 , runs in normal mode. (see figure 6 in doc)
2x512 + 2x512 , runs in dynamic mode (see figure 5 in doc)
Table 4 at the end of the doc, lists configs from highest
to lowest performance. There is a slight advantage to
dynamic mode, as more pages can be kept open for future
overhead-free ops on the memory. When a page has to be opened
for a memory access, there is extra overhead. The normal mode
would have more page opening incidents than would the dynamic
mode. (This statement is access pattern sensitive, and
pathological access patterns can be constructed that would
defeat the advantage. The Intel optimization assumes someone
isn't trying to defeat it on purpose.)
Note that Intel doesn't condone overclocking, and a two DIMM
overclocked configuration could well be the actual #1 config
for that table. Two DIMM configurations of DDR memory overclock
better than four DIMM configurations. There is less of a
difference with DDR2, due to the improved bus termination
scheme with the new memory.
I haven't seen any articles, giving benchmarks for all of
the various modes possible. PAT gives a few percentage points,
CAS2 would be worth a few percentage points, and probably the
2GB config above would be a few percentage points better
than the 3GB configuration, due to dynamic mode. Perhaps a bit
of searching in the private forums can quantify all of that.
I own 4x512MB 2-2-2-6 memory, and these are the results I got
with memtest86, and the bandwidth monitor in the upper left of
the screen. DDR400 memory, FSB800 proc, P4 2.8GHz Northwood, stock.
BIOS was 1014.
2x512MB 2732MB/sec
4x512MB 2667MB/sec (PAT disabled)
4x512MB 2732MB/sec (PAT enabled - performance mode set to standard)
For some reason, the dynamic mode improvement, of interleaving
twice as many pages, didn't seem to improve the synthetic
measurement that memtest86 does. I never did bother investigating
further about that.
I also ran some single channel tests with two DIMMs. I'm quoting
these to show how much impact CAS2 versus CAS3 makes:
2x512MB single channel 2-2-2-6 1681MB/sec
2x512MB single cnannel 3-3-3-8 1584MB/sec
To work out the real world impact, I use the "1/3rd rule". In
the above example 1681/1584 = 1.06 or a 6% improvement. In the
real world, you would see the application run 2% faster. keep
that in mind, when viewing all these "synthetic" benchmarks.
I haven't heard of any probs with 2x512 + 2x1024, but if you
turn any up, post back and tell us
I'm considering playing
around with a RAM upgrade in the new year, once the "Christmas
price gouge" is over. DDR prices are still falling at wholesale,
so there should be savings to be had at retail, at some point.
Paul
