memory latency vs clock?

Do you have a question? Post it now! No Registration Necessary.  Now with pictures!

Threaded View

Is the a way to estimate the impact of memory latency and clock rate on the
overall system performance? In other words, if you plan to overclock, do you
buy faster memory with higher latency, or slower memory with less latency?

Right now I have an intel E4300 on an asus P5PE-VM. The BIOS only allows
adjusting the multiplyer 6 to 9 and manual entry of memory latency. No other
clock or voltage adjustments are available. But it is doable using clockgen.
When I increase the memory speed from 200 to around 240Mhz, the system
crash. If I use a different bank of DIMM (I tried one at a time), it crashes
around 255 Mhz. Both of the DIMM are DDR400 with 3-3-3-8 timing, but
different brand.

If I want to make the PC run faster (for video encoding), I can buy DDR400
with lower latency (e.g. 2 3 2 6), or I can get DDR500 with higher latency
(2.5 4 4 8). How do I make the decision?

And, is it better to install 2x512MB than 1x1GB DIMM?

Re: memory latency vs clock?

Quoted text here. Click to load it

All else being equal it's better to use lower density memory.
So opt for 2x512MB.  As for latency vs. speed, for Intel
systems you should focus on speed.  If you have no voltage
adjustment for memory, pay close attention to what the usable
range is for the memory you're buying.  E.g. if your board
only goes to 2.8V, buy memory with a supported range of
2.6-2.8V and run it at 2.8V.  This should get you the max
overclock for that memory.  FSBs over 240 is a problem
for most DDR memory, I've had great luck with Rev C
Infineon -5's, sold by Corsair (at least as of last June) as:

These modules run rock solid up to 240MHz @ 2.8V.
I've run into stability issues beyond 240 but I believe they're
due to my CPU (Northwood P4) and not the memory.

Re: memory latency vs clock?

Quoted text here. Click to load it

It's better at this point in time to look at higher density modules. 1
GB dimms will start becoming relatively small in relation to what we
wish to run. 2GB is a nice figure to run because it's larget than most
apps and games will need, while not being too large to come accross
the windows memory mapping issues.

Latency vs throughput depends on what you're doing. For media
encoding, you need throughput. For running super-pi - that's latency.
Both help.
Chips react to each differently. Netburst cpu's reacted better with
bandwidth - My pentium D is coasting on 2GB of 1066MHz but such speed
is useless for core 2 duo. In fact, anything over 667MHz is barely
noticed with core 2. The advantage of higher clocked memory is it
directly translates into usable lower cas at lower speed. My memory is
actually running at 888MHz with cas 4-4-4. Core 2 simply cannot
utilize or doesn't need the kind of bandwidth netburst did. Netburst
never performed so well, but it showed steady improvement with memory

The problem i see is your board first of all, isnt going to give you
optimal performance for core 2 duo. 865 didnt even support dual core!
that's a seriously re-engineered chipset! Last time I saw 865, It was
powering my roommate's 2.4C northwood. You're not going to hit optimal
memory bandwidth either. optimal bandwidth for core 2 duo is hit
between 533MHz and 667MHz on dual channel memory. Given that your
board isnt meant for overclocking, I'm not even going to suggest ddr
500. I don't even know how well your board will react to low latency,
2-2-2 1t memory either. You dont have a high performance board - I'd
look at compatibility and reliability. If it were me, and I happened
to be in your position, I'd look at ddr 500 dimms, but that is risky,
as your board may not work properly with it.

Site Timeline