Timings vs Bandwidth (mhz)

[digitalfood]

Is trying to have the lowest and tightest timings better or is having a bigger bandwidth better? I know having lower timings allow the system to perform faster. I read someone's post of loosening the timing which allowed him to raise the speed of the memory to 270 mhz from the stock 200 mhz (or some number like that) which in turned gave him a bigger bandwidth from 4.6 gb/sec to 6.4 gb/sec (or some similar numbers I can't remember now).

That makes sense to me as there will be no bottle necking with the bigger bandwidth. The lower timings makes sense to me as well since it will take less clock cycles to perform a task. I'm sure there's a happy medium somewhere for a setting for the timing and speed to be working at optimum performance. Now comes the thought that the flow of data never uses the full available bandwidth, so the lower timings is in fact more important by cutting down the time it takes to complete a task. Is this statement correct?

If given the choice when overclocking, should you lean more towards trying to raise the speed of the memory as high as possible, or lower and tighten the timings as low as possible?



>:^/

 

[MadModMike]

I was the one who stated the increase in bandwidth with relaxed timings, the numbers are 5.7GB/s at 2-2-2-5 200MHz to 6.4GB/s 3-3-3-8 at 225MHz. Changing from 2-2-2-5 200MHz to 3-3-3-8 200MHz results in a loss of 100MB/s on RAM bandwidth (from the Modules I tested) and testing Game performance, I found a negligable difference in performance switching timings, I didn't really see the need for those tight timings. But, also the fact that I had higher bandwidth, did not result in the higher frames that I was expecting, even going to 250MHz and getting 7GB/s bandwidth at 3-4-4-8 timings.

I would suggest running 2-3-3-6 and going 225MHz 1:1 w/ CPU and that will give you the happy medium you inquired about. I did not, however, test application performance from timings or bandwidth, but I can say with great certainty, that the higher bandwidth will help cut down render times in 3D Modeling programs (Maya, LightWave, 3DS Max, etc.) as well as in Server performance. If you're running a Desktop computer geared for Gaming, I recommend the 2-3-3-6 if you want a medium point, otherwise drop timings to 3-4-4-8 and hope for 300MHz DDR600 if you love to overclock.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[digitalfood]

Ha! How do you like that? Someone actually used your info to help them!

Since you had higher bandwidth, but didn't result in higher framerates, I wouldn't be too surprised as the graphic card probably handled the majority of it.

I can live with the loss of 100MB/s to relax the timing for stability. That's not that bad of a trade off.

I think it was Xbit that had an article of value ram with high timings and comparing it to performance ram with low timings yielded an increase of about 5-10% increase in performance.

However, comparing the 5-10% increase to what you stated of only losing 100MB/s from 2-2-2-5 to 3-3-3-8, is it simply because of the tight times of 3-3-3-8? Instead of 3-4-3-7 (or some other similiar numbers)?

btw, it's for a desktop used mostly for gaming.

 

[MadModMike]

You lose performance because you increase the amount of time it takes to perform the specific tasks, that all accumulates to the 100MB/s loss in performance.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[dvdpiddy]

timings all the way man!

 

[chuckshissle]

I'd say timings is most important.

 

[mpjesse]

The key question here is: are you going to overclock the FSB?

Regardless of how fast (in Mhz) you memory is, if it's running at DDR400 w/ an Intel or AMD chip it ain't gonna make a frigin difference. If you're not overclocking the memory runs at 200mhz on both an AMD system and Intel system (both were designed to use DDR400/DDR2-400). (of course, the P4 EE uses a 266mhz FSB).

Anyways, if you don't plan to overclock then the most important factor is timings. If you do plan to overclock then the most important factor is Mhz- not timings. (though you should chose carefully... I.E. get decently timed high Mhz memory)

-mpjesse

 

[digitalfood]

Yes, I had planned on overclocking. What I found to be interesting and maybe my answer I was looking for is to go with bandwidth.

As wusy mentioned above, it only takes 4mhz to overcome the difference between 2 and 3 timings. I think, I can save a few dollars and not try to get the lowest timings, and put it into another component instead.

http://www.xtremesystems.org/forums/showthread.php?t=73830&highlight=ocz+gold

According to the above link the poster RyderOCZ states, "just the comparison between the 2 kits.....no difference really from 200 2-3-2-5 to 250 3-4-4-8 when CPU is constant."

Also, since I'll be using AMD, the memory controller is on die, so there is minimal effect the timings have. In fact, running 1t or 2t according to Angry's sticky at dfi-street had very little effect, maybe a 1% difference.

btw, I wasn't trying to answer my own questions. I just happen to find this as I was trying to find out more info.

 

[digitalfood]

I'm definitely going with 2x1gb. If I have to drop to 3, it's not a big deal as I wouldn't notice it unless I run benchmarks and compare the numbers.

 

[mpjesse]

Running 1T makes a decent difference according the benchs I've read. At any rate, you can get 1T w/ 2xwhatever in almost all cases. (if the memory is quality)

-mpjesse

 

[mpjesse]

Exactly. Intel processor seems to benefit the most from lower timed memory. This is because (as you said)of the Athlon's on die memory controller which tends negate any latency.

-mpjesse

 

[michaelahess]

Keep in mind, the more synchronise (don't comment on spelling your system runs, the better performance will be, if you run with an async bus fsb or ram timings, the system could take two cycles to process the data instead of one. This may be offset with enough overclocking, but that would take more experimentation than I have time to try.