the effects of RAM timings

[Statross]

i'm trying to find out all i can about memory, its one thing i'm low-in-knowledge on.

so with that in mind, i wqant to know, now, what the timings in RAM actually do.

browsing memory prices today everything seems very inconsistent. there's some OCZ with 4-5-4-18 thats alot more expensive than some with 4-4-4-12 and some corsair with 5-5-5-15 more costly than some 4-4-4-12 and all sortsd of others inbetween.

so i'm sure timings arent everything (well, who didnt know that anyway) but how much do they actually affect it?

so far i think lower numbers = faster turn-around. why? because its described as lower latency. but thats about as far as i've got. so why ar ethe timings important? and why do they make RAM with 5-5-5-15 or 9-9-9-99 at all if its worse than 4-4-4-12?

and if the timings affect how fast data can be accessed or changed then then how much does it affect the speed on RAM rated at 533 or 667 or 800 or whatever.

also: as i'm gonna OC whatever possible in my self-built later this year...can i change my RAM timings?

memory frequency i pretty much understand on a basic level, voltages i do aswell but timings i really dont and wikipedia has been of very little help, lol.

 

[NewbieTechGodII]

Very good questions, indeed.

 

[Statross]

well, thats what i thought...but in the last month or so i've seen people recommending loosening timings...which indicates to me that timings can be changed (as i thought they probably could)...but why would keeping them tight help OCing?

or why would loosening them up help OCing?

 

[theworminator]

Basically, it's like a delay for the ram to get info. We'll just use arbitrarily chosen numbers, and say it takes 4 cycles of the clock to get the info. This is slower than if it took only 3 clocks, assuming both clocks are at the same speed. But if the 4-clock is twice as fast, the memory with the timing of 4 is faster. Obviously, it's more complicated than this, with more than only one number, but that's the gist of it.

Ram #1
Timing: 3
Clock speed: 400 mhz

Ram #2
Timing: 6
Clock speed: 800 mhz

They'd essentially be the same speed. I'm hoping I'm right here Look on wikipedia for the explanation of each number.

 

[Statross]

well, thats what i thought...but in the last month or so i've seen people recommending loosening timings...which indicates to me that timings can be changed (as i thought they probably could)...but why would keeping them tight help OCing?

or why would loosening them up help OCing?

The idea is to keep your timings as tight as your overlock will possibly allow. for example ddr2@1000mhz 5 5 5 15 will be significantly slower than ddr2@1000 4 4 4 5.

but why? and what does "as tight as the OC will allow" mean? does it mean if your OC goes too far then you need to losen timings? what do these timings do?

 

[theworminator]

You can manually lower timings in the BIOS, but you chance causing instability if you're trying to force it to run too fast.

 

[wimcle]

On the other hand , outside of superpi it has little to no effect on performance

with pc2-6400 (800Mhz) you will see 1-2 fps increase going from cas3- > 4 -> 5.

http://www.anandtech.com/memory/showdoc.aspx?i=2900&p=6


In my opinion: cheap, can do 800Mhz at 1.8v, and a lifetime warranty, are the specs that matter.

 

[Statross]

is it better to be able to do 800mhz at 1.8 than 2.1?

 

[Statross]

okie dokie, well doing 800mhz is good for OCing, having voltage headroom is also useful and warrenties for anything are never a bad thing.

but for OCing, why would you need to adjust the timings? i mean, adjusting the timnings may get a tiny bit of extra performance with everything else left the same.

but OCing people always seem to recommend tightening timings or getting RAM with tight timings...

so...why is this important for OCing? or...is it important for OCing? is it just poeple who dont know much advising tighter timings becuase of "lower equals better" mentality?

 

[funnyman06]

The reason they recommend the tight timing ram is because the tighter the timings the more headroom you have with the ram, so ram that like DDR400 at 2225 will normally OC farther than like DDR400 at 3338, but then the voltage also needs to be taken into consideration because voltage = heat. So you normally want the best timings at the lowest voltage so like i have DDR thats DDR400 and 2336 @ 1.75 volts, so it runs fairly hot. But then i feel i could safely run it at 2 volts with the cooling i have so the timings are alright for OCing and the volts are a tad bit high so it wont OC as well as something with like 2225 @ 1.6 volts

 

[Statross]

hmmmm...i see

but when you want to increased frequency you have to up voltage to keep it stable? so tighter timings at lower voltage means you can OC to a higher frequency at a comparatively lower voltage?

but still, why does the timing have a positive or negative effect on OCing with regards to frequency or voltage?

lower = more headroom....but more headroom for what, exactly?

 

[KTev]

If you take two chips with the same speed, the one with the tighter timings is going to be of a higher quality. With the option of relaxing the timings there is more potential for a better OC.

For normal performace, most won't see a big difference from slight timing or hz changes.

The wikipedia if fairly good. If you don't understand it I you might just want to give up. If divide the speed by the timing you can get a low tec relative compare.

 

[funnyman06]

as the guy said before ram at 800 Mhz and timing of 6 will run the same as 400Mhz and a timing of 3, so the point of the low timings are that when you are OCing you dont want to be loosening performance, there is a balance. Im trying to think of a good metaphor...

okay lets try this on for size..

So say your going to build a car engine, and you have only so much money and you need to build the engine and the exhaust system. We will say the engine is the clock and the exhaust is the timings, its really all arbitrary, but whatever.

So you dont want a huge engine with tiny exhaust you will be choking it, and if you have a huge exhaust why not build up the engine so meet the demand. So if you have a huge clock like DDR2 has, then its really hard to get an exhaust big enough to freely move the air, so you make due with a standard exhaust. And for most people that works fine. But then you have the OCer, who has this engine with massive pipes but the engine is lacking, so he/she can now increase the clock a lot more because he has more headroom before he gets an increase in back pressure. But like i said, you have a set budget kinda, so the big pipes are expensive so you have to decrease them to get the bigger engine, and that middle ground is the place where most OCers want to be. Where you have the best balance between clock and timings.

Maybe this helped maybe it just confused, if not ill try something else, but this headroom is how high the clock can go before you start undoing the benefit of the OC. this is where multipliers come into play. so your FSB can be at like 300 but ram run at 200 or whatever

 

[Statross]

its not that i cant understand it (i's not exactly technical) it doesnt say anything about timings...unless theres an article on RAM timings i cant find?

i'm looking at one called DDR2 SDRAM

 

[Statross]

as the guy said before ram at 800 Mhz and timing of 6 will run the same as 400Mhz and a timing of 3, so the point of the low timings are that when you are OCing you dont want to be loosening performance

lol! i dont know why that had to spelled out to me for me to get that...i think i'm still piecing everything together...

so yes, now, of course, i see why low timings are even better for OCing.


but one thing i dont get is how it is that you can reach a point where turning 4s into 5s helps an OC? at what point does 4-4-4-12 need to be turned ito 5-5-5-15 and why?


as for multipliers...i know what one is, but with multipliers and OCing people quote all these ratios to have RAM and CPU running at...1:1 is the only one that makes sense to me (i havent looked into that one yet, it's probbaly next up for investigation)

 

[theworminator]

They're known as CAS latencies.

http://en.wikipedia.org/wiki/CAS_Latency

 

[funnyman06]

o that parts simple, you will do that at a point of instability, either your pushing it to much and it wont post, or itll BSOD, or something. So you loosen them up so it can breath... or you increase voltage to increase stability. Thats what its all about, get the highest stable clock, so like my ram will probably have to get loosened up before i can increase the clock or it just wont post, and also the fact that im running 4 sticks will just add to the complexity because not all ram is created equal... So its basically a guessing game, trial and error till it works how you want it.

 

[Statross]

ahhh..wait, i think i have it...

if you're pushing and pushing the frequency then theres alot of information flying through the chips and more and more as you increase the frequency.

so with tight timings, theres low latency so lots going on and, i suppose, with the increased frequency, the RAM cannot keep up with requests. sooo unless you increase the voltage (so it has more power to keep up) which = heat (so you cant keep doing that forever) you can relax the timings so that not quite as much information is going through but at least that way its not collapsing ontop of itself from heat or information overload?

i think thats it? lol

 

[Statross]

They're known as CAS latencies.

http://en.wikipedia.org/wiki/CAS_Latency

ah, thanks, i'll check that out in a minute

 

[theworminator]

It's a bit convoluted...the Ram Latency article might also be of help.
http://en.wikipedia.org/wiki/RAM_latency

 

[CompuTronix]

This might help you to put timings into perspective.

Benchmarks will show that 4-4-4-12 vs 5-5-5-12 DDR2 timings increase memory performance by less than 2%, which has little impact on overall system performance. Check out the following THG article regarding the relationship between memory frequency and timings:

http://www.tomshardware.com/2006/03/31/tight_timings_vs_high_clock_frequencies/

I ran memory benchmarks on my rig with PCMark04, 3 tests at 4-4-4-12, and 3 at 5-5-5-12. The average of the first 3 results divided by the average of the second 3 results are 12695 / 12550 = 1.16% It's still better than 0%, so we'll take whatever we can get.

 

[BMFM]

Allow me to try explaining it. No offence, but you guys seem to be dancing around the point without actually coming to a conclusion .

This is going to be a little 101. So please forgive me if it sounds a bit patronizing…

Ram latency is defined, in a simple way, as the time between a request for data and its retrieval.

Now, latency is measured in nanoseconds; so it will correspond to a relative amount of cycles at any given frequency.

For example: DDR2 800 CL3 – it means it will spend 3 cycles every time a request is placed and the data is actually retrieved. In this case (and in theory), the same chip clocked at 533MHz (DDR2 1067) would have a CL4.

So whenever you think: "latency" in cycles, you should think in relative terms. Objectively it would be measured in nanoseconds, as I've said before, but it's not a very practical measurement you can work with when over-clocking, or just properly setting you RAM.

And for a successful over-clocking; your latency settings (in cycles) should match the module’s latency measurement, in nanoseconds.

And that's why memory modules with tighter timings should allow for a higher over-clocking. By loosening them you will be able to clock your modules at a higher frequency within the measured module's latency.


And just so we're clear, about my example above: the DDR2 1067 CL4 will have the same latency as the DDR2 800 CL3, but will be considerably faster.

 

[Naw-yi]

plz all of u click here and learn:

http://www.hardwaresecrets.com/article/26

hope it helps all of you.

 

[plankmeister]

I don't suppose you read the sticky about RAM in the Memory section, did you? I had similar questions which were mainly answered by this helpful guide.

EDIT: Memory FAQ

 

[plankmeister]

Now, latency is measured in milliseconds; so it will correspond to a relative amount of cycles at any given frequency.

Milliseconds? DAMN! That's some SLOW memory... Even 10 year old EDO memory had 60ns latency...