Memory timings

[wxrkny]

I have 1 GB of PC3200 DDR RAM. The timings are 3-8-3-3 which i know suck. But when i go to 2.5-8-3-3 or 3-7-3-3 my memory score in PCMark 2002 go down. Could you please recommend a better memory benchmark or an explanation as to why my score goes down.

Here are my system specs:
Abit IS7 motherboard
P4 2.8 Ghz 800 mhz fsb
1 GB of PC3200 DDR RAM
120 GB Western Digital Hard Drive w/ 8MB cache
Radeon 9700 pro
Audigy 2

Also my 3dmark 2003 score was 5034 is that normal for my system?

Thanks in advance for your help.

 

[dunkster]

http://www.mushkin.com/mushkin/pop-up/latencies.htm

The above article provides an excellent discussion exactly why lowest Tras may hurt memory bandpass.

http://www.corsairmicro.com/memory_basics/153707/index.html

The memory tutorial at the above site is an excellent all-around discussion of DDRAM.

The Tom'sHardware site has a 'guide' article on memory that's good. Perhaps best is a timing diagram illustrating the time domain relationship of each of the latency settings.

Hope this helps!

 

[Soulkeeper]

ok
many people recommend adding the first number to the second to get the tras
for example:

2-2-2 timings would use 6
3-3-3 timings would use 8
2.5-2-3 would use 6 or 7

i recommend running yur timings as low as you can while following this rule
the reason making this number lower doesn't always make yur system faster is because it has something to do with the timing you have to access info and when it stops sometimes info isn't accessed

let me give you a link that explains much better than a tired person like myself can
Mushkin's spin on the subject

 

[wxrkny]

Is PCMark 2002 a reliable source for memory benchmarking or should i use a different memory benchmarking utility? Please suggest better benchmarking software if you know of any. Thanks again.

 

[wxrkny]

bump...Please Help

 

[oldfart]

Those timings do indeed....:frown:

What kind of ram is it? Can you run more aggressive timings with stability? Are you overclocking?

Run some real application benches. Dont bother with PCMark/Sandra/Aida...etc. Run benches on games and apps that you actually use and see what works best.

 

[dunkster]

Here are four of the most-used benchmark programs to qualify system performance:
1) Sandra memory bandwidth benchmark will report memory bandwidth and also memory efficiency.
2) Memtest86 is excellent for determining if memory operation is error-free with given settings.
3) Prime95 is excellent for determining if CPU operation is stable at given settings.
4) 3DMark2001 is an excellent program to determine overall 3D performance (total system bandwidth).

Here's one method to test memory timing and frequency limits:
Use a low multiplier (board and CPU unstressed), such as 5X. To further reduce stress on board and CPU, you can set memory to run asynchronously at 120% of FSB.
Set tightest practical memory timings.
Increment FSB, adjusting Vcore and Vdimm as required, until errors are produced by Memtest. Note the memory frequency, since that will be the FSB limit for your memory with those timings when running synchronous memory at 100% of FSB. To further map memory characteristics, relax memory timings and repeat.

You can save a lot of testing time with Memtest86 by running only Test#5 (block moves), since it is usually the most critical test. Use 2 passes for certainty, since Test#5 finishes quickly.

Hope this helps!

 

[pallmall]

Agreed to dunkster 99%.
The 1 % is that I sometimes get errors on memtest step #5 in pass 4, 5 or 6...

I find the sandra memtest only useful when determining the max tRas.
When memtest score shows low points by 7,2,2,2.0 compared to 11,2,2,2.0 without any errors,
it is saying that the tRas is still too tight.

SuperPi is useful for the comparison of combined test for cpu and memory to avoid the hassle of specially tweaked vga drivers.

3dMark2001 is useful for the comparison of combined test for cpu, memory and vga upto dx8.

My 1 cent.

 

[newindy4130]

what do you think of 2-5-3-3 ?

 

[pallmall]

Originally posted by: newindy4130
what do you think of 2-5-3-3 ?

About what?
Rearranging the timings to my preferred order, it would be 5,3,3,2.0.
I think the last 3 numbers are mostly restricted by memory spec.
The first tRas need to be adjusted by fsb changes mostly.
Overall, the timings ontimazation is done by going high fsb with relaxed timings or
preferred timings with highest possible fsb.

 

[maluckey]

I agree with Oldfart (mainly because I am an old Fart). I run 3D games like Unreal Tournament 2003, and Serious Sam, for Real World tests. It has been my experience that benchmarks are a quick view, not the whole picture.

When I run ultra tight settings, in 3D games, the absolute lowest framerate is higher in almost all cases, than with looser timings, even though some benchmarks say the setup is slower. The thing is, the absolute highest framerate may drop in some cases, but it's irrelevant because you cannot see the difference of 10 FPS above 100 FPS anyways, while 2-3 FPS below 45 FPS IS noticible.

Running benchmarks to test performance is like drag racing your car on the computer, using a simulation, to see how fast it is. I's just an educated guess, nothing more, and only once you run it for real do the real results show up.