PDA

View Full Version : Understanding DDR Memory (dd2,ddr3) etc


Jimingle10
08-30-09, 01:32 AM
I enjoy building computers and tinkering with clocks and all, but have always avoided the memory other than straight adjusting the clock speed.

Looking for a good link to explain :

Memory Latency
Memory Timings


Also, can you mix different kinds of ram within the same type (ex: pc6400)?

I have 2x2gb sticks of some kingston pc6400 value ram, i do not know the timings or latency off hand right now. Can i mix other pc6400 modules with what i have? I found a nice deal on ocz memory with the big heat sinks on them locally on craigslist.

this is for my sig rig, obviously running 64bit windows.

Im in the mood to learn :)

-Jim

adelphia83
08-30-09, 11:53 AM
Yes you can mix different types of PC6400 modules together, even different speeds, like PC6400 and PC5300 together, but you'll be limited to the speed of the slowest module.

Mixing modules generally isn't recommended for overclocking as it throws some more variables in there, but it's okay for stock settings.

Memory latency and timings are two of the same. The lower the latencies (timings), the better the performance. It is possible in most cases to increase timings to allow greater frequencies, which typically equates to increased performance. It's a balancing act and is completely dependent on not just the memory, but the other components in your system.

What I generally do is find the max CPU speed and max RAM speed independently. This is done by decreasing the CPU multiplier such that the CPU can never hinder stability while I search for the highest possible RAM speed. To find the highest CPU speed, it's the opposite; I change the memory divider such that the RAM speed is so low that it cannot contribute to any instability.

Once I know the max speed of both, I find a happy medium so that both have a decent overclock while maintaining stability. They you can whittle away at the timings to see what maintains stability and what doesn't.

18 is # 1
08-30-09, 02:03 PM
I found a nice deal on ocz memory with the big heat sinks on them locally on craigslist.

I suggest posting the model and version number. We can tell you if it's a good deal.

Once I know the max speed of both, I find a happy medium so that both have a decent overclock while maintaining stability.
Usually the CPU speed makes a MUCH bigger difference in performance. There would have to be a huge gap in between straps to justify even a small change in CPU speed for more RAM speed.

Bobnova
08-30-09, 02:23 PM
The memory timings are the number of memory clock cycles required for the memory to do a certain action.
If you have your ram running at 1ghz (one billionth of a second per tick) and you have ram that requires cas9, to do the cas operation takes nine ticks, or nine billionths of a second.
Now if your ram is running at 2ghz (possible for ddr3) and cas9, it still takes nine ticks, but those nine ticks are each only half a billionth of a second, so cas9 takes 4.5 billionths instead of nine billionths.
Hence, cas8 ddr3-1333 is not very fast, while cas8 or even cas9 ddr3-2000 is quite fast.
Generally, lower timings and lower ram speed are better for small data packets, while longer timings and higher ram speed are better for larger data packets.

In reality, the difference between say ddr2-800-4-4-4-12 and ddr2-1066-5-5-5-15 is minimal and hard to prove in benchmarks.
Now the difference between 800-4-4-4-12 and 1066-4-4-4-12 is larger and testable. Not that you're likely to find 1066-4-4-4-12 anywhere, but it's just an example.
Also because of this, DDR3 at lower then 1600 is often not actually faster then ddr2, as ddr2-1200-5-5-5-15 will equal or beat ddr3-1333-7-7-7-21 does to it's faster timings.



The bottom line is that it's a mess, and if you have ddr2 800 or better or ddr3 1600 or better you're doing pretty good, and it's not worth worrying about unless you want that last little .5%.
(As you may have guessed from my sig, i do want it)

adelphia83
08-30-09, 02:51 PM
The memory timings are the number of memory clock cycles required for the memory to do a certain action.
If you have your ram running at 1ghz (one billionth of a second per tick) and you have ram that requires cas9, to do the cas operation takes nine ticks, or nine billionths of a second.
Now if your ram is running at 2ghz (possible for ddr3) and cas9, it still takes nine ticks, but those nine ticks are each only half a billionth of a second, so cas9 takes 4.5 billionths instead of nine billionths.
Hence, cas8 ddr3-1333 is not very fast, while cas8 or even cas9 ddr3-2000 is quite fast.
Generally, lower timings and lower ram speed are better for small data packets, while longer timings and higher ram speed are better for larger data packets.

In reality, the difference between say ddr2-800-4-4-4-12 and ddr2-1066-5-5-5-15 is minimal and hard to prove in benchmarks.
Now the difference between 800-4-4-4-12 and 1066-4-4-4-12 is larger and testable. Not that you're likely to find 1066-4-4-4-12 anywhere, but it's just an example.
Also because of this, DDR3 at lower then 1600 is often not actually faster then ddr2, as ddr2-1200-5-5-5-15 will equal or beat ddr3-1333-7-7-7-21 does to it's faster timings.



The bottom line is that it's a mess, and if you have ddr2 800 or better or ddr3 1600 or better you're doing pretty good, and it's not worth worrying about unless you want that last little .5%.
(As you may have guessed from my sig, i do want it)

Good explanation, I've been wondering about the difference between DDR2 vs DDR3. I've kind of been out of the loop, my last real venture into overclocking was back in the Athlon 64 / Samsung TCCD days. I guess I have to get up to speed.

hansen
08-30-09, 05:58 PM
Now if your ram is running at 2ghz (possible for ddr3) and cas9, it still takes nine ticks, but those nine ticks are each only half a billionth of a second, so cas9 takes 4.5 billionths instead of nine billionths.

Remember that the latencies are given in whole cycles. 2000 Mhz (effective) is 2000 half cycles and 1000 whole cycles.

Daddyjaxx
08-30-09, 06:42 PM
DDR3 is faster than DDR2, but the latencies are much higher.

Spyguy
08-30-09, 08:05 PM
DDR3 is more efficient and supports lower voltages; thus reducing power requirements by a tad bit. DDR3 also has a higher bandwidth, which is what pcXXXXX stands for. I would not recommend mixing modules of the same speeds because they are all built somewhat differently; which can create a small unnoticeable wait time when sending and receiving data between modules.

adelphia83
08-30-09, 09:34 PM
which can create a small unnoticeable wait time when sending and receiving data between modules.

Timings are timings and the RAM chips are held to that. There is no "unnoticeble wait time". Either the RAM chips can keep up with what they're paired with or they can't, which would manifest itself as instability.

There is nothing wrong with pairing modules of different types-- Some don't get along well with others, which will limit overclocks and often stock speeds. As always, the speed would be limited to that of the slowest module.

Jimingle10
08-30-09, 10:22 PM
Very good replys, however im still confused as to what the timings actually stand for

the 4-4-4-12 or what have you?

Spyguy
08-30-09, 10:33 PM
As always, the speed would be limited to that of the slowest module.

Yes, that is what I meant. Some brands don't get along so well, and the small wait time is the limit by the slower module. Even thought the modules are at the same specs, one brand might uniquely be slower in a way. Sorry I worded it in an unclear way, but that is basically a deeper way of saying it.

Jimingle10
08-31-09, 01:34 PM
Also, besides the timings....i noticed that different modules have different voltage specs? Does this matter when mixing?

+sYnc3+
08-31-09, 04:36 PM
Also, besides the timings....i noticed that different modules have different voltage specs? Does this matter when mixing?

Voltages are key..without proper voltages on ram..you can run into stability issues..same goes for cpu..ie...say you have corsair 555-15@ 1.9 v and kingston 555-15@ 2.2v...if say you run@ 1.9 across the board..you may run into stability hiccups..or if you run 2.2 across..same thing..Hope this helps..

18 is # 1
08-31-09, 08:18 PM
Voltages are key. Hope this helps..

Maybe...depends on the sticks.

Still waiting Jimingle10 for you to tell us about your amazing find.....

Bobnova
08-31-09, 08:49 PM
Or in the case of running 1.9v memory at 2.2v, it may simply fry.

adelphia83
08-31-09, 09:02 PM
Mixing modules takes some common sense.. The above post is spot on, you wouldn't want to mix 2.2v performance RAM with 1.8v value memory.. Unless you want to roll the dice and see if the performance RAM plays nice at 1.8v. It probably would at reduced speeds and timings.

If you're looking to increase capacity at the cost of overclocking headroom and speed, mixing modules can be a safe bet. But overclocking generally necessitates matched pairs since timings, voltages and frequencies all need to be fine tuned, which is almost impossible when mixed.

18 is # 1
08-31-09, 10:34 PM
Or in the case of running 1.9v memory at 2.2v, it may simply fry.

+1
The new Low Voltage OCZ will fry at 2.0v:mad:

Bobnova
08-31-09, 11:35 PM
That, is why i flat refuse to give it even +.01 :P
Well that and the fact that it already runs at 1200 and far, far out paces my cpu's FSB cap in data transfer :(

18 is # 1
09-01-09, 06:53 AM
That, is why i flat refuse to give it even +.01 :P
Well that and the fact that it already runs at 1200 and far, far out paces my cpu's FSB cap in data transfer :(

You know some of those guys at XS are running over 2v with spot fans.:screwy:

Bobnova
09-01-09, 10:44 AM
Some of the guys at XS are crazy, we already knew that :P

SkiBum1207
09-04-09, 01:38 AM
Very good replys, however im still confused as to what the timings actually stand for

the 4-4-4-12 or what have you?

They stand for the latencies. I cant remember off the top of my head what each stand for but honestly the actual definition is beyond 99% of caring. Simply they dictate how many "cycles" pass for each ram refresh. Lower number signifies lower latencies aka. lower time to access data. The first number has the greatest affect on latencies and stability, the final number has the least. Usually as a general idea the first 3 added together equeal the last number but thats just a guideline. So try to get them as low as possible whie maintaining stability. However a higher latency (555-12) will give a higher MHZ OC than a (444-12) so just bench to find the perfect mix!

Hmmgotmilk
09-04-09, 02:02 AM
this link may help to explain some of the memory timings you see:

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