Sign in with
Sign up | Sign in
Your question

Memory Latency Timing Question

Last response: in Memory
Share
September 10, 2009 1:17:59 PM

Can someone help a n00b like me. I am building myself a new Performance PC (DDR2 based, can't justify DDR3 at the moment). And it has been a very long time since I built a PC, long before you had to worry about your processor being compatible with various other aspects of PC Design/build.

My question is this are LOWER latency timing better than higher ones. I am currently looking at some memory with a latency timing of 5-5-5-18. But logic would dictate that this is pretty poor, and I should be looking for something along the lines of 4-4-4-16.

As this is a DDR2 performance PC I am building, can someone explain latency timing, as it would seem the more memory I want to us (4GB in this case) the higher the timing goes, but I also want the memory to be quick.

Would anyone be able to recommend some DDR2 too. I am currently looking at OCZ Reapers, and Mushkin Redline XPs?

Cheers
Darren
September 10, 2009 1:49:28 PM

Lower timings are better. You can overclock the timings. Test with memtest86+ for errors.
Higher memory speeds are preferable to tighter timings.
September 10, 2009 2:09:27 PM

So: GeIL 4GB (2x2048MB) PC2-8500 1066Mhz Black Dragon Dual Channel Kit;
- 240PIN DIMM Module for DDR2 motherboards
- 2GB Capacity (2x 2GB Modules) Dual Channel Kit
- DDR2 Module
- 1066MHz PC8500 speed
- CAS Latency 5.0 (5-5-5-15)

or: OCZ Reaper HPC 4GB (2x2GB) DDR2 PC-8500C5 1066MHz Dual Channel Kit;
- 1066MHz DDR2
- EPP 5-5-5-15 (CAS-TRCD-TRP-TRAS)
- Unbuffered
- OCZ Lifetime Warranty
- 2.3 Volts
- 240 Pin DIMM
- Reaper HPC Heatsink
- EPP-Ready
- 2.35V EVP

Would be perfectly acceptable, because they are faster as opposed to tighter timings. I'm prefering the Reapers, unless someone can suggest something better ;) 
Related resources
September 10, 2009 2:21:53 PM

2.3v is an extreme voltage. Make sure your motherboard supports it.
I prefer memory with lower voltages. I can always turn up the voltage if necessary.
September 11, 2009 10:08:08 AM

evongugg said:
2.3v is an extreme voltage. Make sure your motherboard supports it.
I prefer memory with lower voltages. I can always turn up the voltage if necessary.


What memory would you recommend?
a b } Memory
September 11, 2009 10:27:20 AM

evongugg said:
Lower timings are better. You can overclock the timings. Test with memtest86+ for errors.
Higher memory speeds are preferable to tighter timings.


I would suggest that most of the time, just the opposite is true, unless I am misunderstanding what you are saying.
800mhz memory with fast timings will generally perform better than 1066 memory with looser timings.
September 11, 2009 11:19:08 AM

And most 1066 memory is overclocked 800 memory. I also prefer memory that is closer to 1.8v - 2.3v is quite high for DDR2.
a b } Memory
September 11, 2009 1:59:59 PM

Higher MHz = better
Lower CL = better
Lower V's = usually better

And don't trust the specs on the sales sites - once you're down to a couple of choices, check everything out on the manufacturer's site for that exact model of RAM.

You need to examine the Motherboard to see what its base RAM voltage is and how far it can step up. My new M/B choice had a base of 1.8V with 3 steps of 0.1 to a total of 2.1V. So the OCZ 1066 RAM I originally purchased would only perform as 800 MHz because it needed 2.2V to run at 1066. So I returned it for 1.8V RAM - didn't have as good a CL either - and most would say I was better off with the lower CL and lower MHz. But I wanted that 1066!

I've now got it running at 1066 with forced lower latency numbers and am very happy. But you don't want to have to pump your max Voltage into the RAM just to meet its minimum requirement.

Since jit and others have commented here, maybe they can judge my formula for RAM speed - can you really make a generalization of throughput using percentages like this?

=======================================

Let's say you're comparing two RAM choices:
1066MHz with CL of 7
800MHz with CL of 5
Which is faster?

Moving from 1066 to 800 is a loss of 267 MHz or a loss of 25%
Moving from 7 to 5 is a gain of 2 or a gain of 28%
So the net of going from 1006/7 to 800/5 is a gain of 3%

Moving from 800 to 1066 is a gain of 267 or a gain of 33%
Moving from 5 to 7 is a loss of 2 or a loss of 40%
So the net of going from 800/5 to 1066/7 is a loss of 7%

The numbers don't match because % is based upon either 800 or 1066 - 5 or 7.

So if you can get 1066 with a CL of 6 vs 800 with a CL of 5, it's a no-brainer, go for the 1066
Moving from 800 to 1066 is a gain of 267 or a gain of 33%
Moving from 5 to 6 is a loss of 1 or a loss of 20%
Net 13% gain in performance!

Can the numbers be estimated so simply?
September 11, 2009 2:06:11 PM

I checked out the spec for my motherboard (An Asus P5Q SE)

Overclocking Features
Intelligent overclocking tools:
- ASUS AI Booster utility

Precision Tweaker:
- vCore: Adjustable CPU voltage at 6.25mV increment
- vDIMM: 5-step DRAM voltage control

SFS (Stepless Frequency Selection)
- FSB tuning from 200MHz up to 800MHz at 1MHz increment
- Memory tuning from 667MHz up to 1333MHz for DDR2
- PCI Express frequency tuning from 100MHz up to180MHz at 1MHz increment

Overclocking Protection:
- ASUS C.P.R.(CPU Parameter Recall)

So theoretically I can support ANY memory up to 1366Mhz???
a b } Memory
September 11, 2009 2:32:44 PM

(might read above, you likely missed my post while posting yourself.)

The manual says you can set RAM Voltage from 1.80V to 3.08V with 0.02 increments. WOW! I'd say you can buy any RAM you want to. 2.3Vs would be nothing to worry about - of course, draws a tiny bit more Watts!

I think your memory choices are solved - pick what you want! One thing to look for is to make sure that your processor choice doesn't affect mem choices - that 1366 is supported with your CPU. This is sometimes an issue with AMDs, I'm not familar with Intels. Read over what I said above about MHz vs CL.

Go to the Asus forum here and post everything about choices you know. Including this page for fast ref: http://usa.asus.com/product.aspx?P_ID=T4T7B10FfR6zJ4Ns

Look into the size of any optional CPU cooler you're considering, make sure it doesn't interfere with the large heatsinks on those RAM you're considering.

Get a big enough Power supply, etc..

Good luck!

September 11, 2009 2:34:55 PM

Lower timings are generally more important than having a higher frequency if the speeds aren't too far apart (obviously you can't compare 400MHz versus 1600MHz).
September 11, 2009 3:01:29 PM

There are two trains of thought on this, the first is that high speed DIMM's can make up for running slower timings by the amount of bandwidth provide the processor. Specifically, bandwidth is the amount of data that can be moved from one given device to another.
The other point of view is that memory with low latency can make up for the lack of bandwidth because the memory has a lower latency that in effect moves data between the CPU and memory faster. Programs that do not require a large amount of bandwidth tend to benefit more from quicker data transfers between the memory and the rest of the computer; such as games or 3D applications.

a b } Memory
September 12, 2009 3:32:48 AM

Thanks for the reference RemdA. That should give me the info needed to make my formula above be accurate. So... when do I have a weekend free?
a b } Memory
September 12, 2009 3:43:57 AM

Mongox said:
Higher MHz = better
Lower CL = better
Lower V's = usually better

And don't trust the specs on the sales sites - once you're down to a couple of choices, check everything out on the manufacturer's site for that exact model of RAM.

You need to examine the Motherboard to see what its base RAM voltage is and how far it can step up. My new M/B choice had a base of 1.8V with 3 steps of 0.1 to a total of 2.1V. So the OCZ 1066 RAM I originally purchased would only perform as 800 MHz because it needed 2.2V to run at 1066. So I returned it for 1.8V RAM - didn't have as good a CL either - and most would say I was better off with the lower CL and lower MHz. But I wanted that 1066!

I've now got it running at 1066 with forced lower latency numbers and am very happy. But you don't want to have to pump your max Voltage into the RAM just to meet its minimum requirement.

Since jit and others have commented here, maybe they can judge my formula for RAM speed - can you really make a generalization of throughput using percentages like this?

=======================================

Let's say you're comparing two RAM choices:
1066MHz with CL of 7
800MHz with CL of 5
Which is faster?

Moving from 1066 to 800 is a loss of 267 MHz or a loss of 25%
Moving from 7 to 5 is a gain of 2 or a gain of 28%
So the net of going from 1006/7 to 800/5 is a gain of 3%

Moving from 800 to 1066 is a gain of 267 or a gain of 33%
Moving from 5 to 7 is a loss of 2 or a loss of 40%
So the net of going from 800/5 to 1066/7 is a loss of 7%

The numbers don't match because % is based upon either 800 or 1066 - 5 or 7.

So if you can get 1066 with a CL of 6 vs 800 with a CL of 5, it's a no-brainer, go for the 1066
Moving from 800 to 1066 is a gain of 267 or a gain of 33%
Moving from 5 to 6 is a loss of 1 or a loss of 20%
Net 13% gain in performance!

Can the numbers be estimated so simply?

No, they cannot, this is not accurate. Percentages of 800 to 1066 vs 5 to 7 has no relelvance in this arguement, you are simply comparing a "number" to a "number" there is no basis or arguement here on how the parts actually perform.
For instance, everything being equal, 1066 memory actually performs about 2-5% better than the exact same memory at 800mhz, at the exact same timings.
800 mhz memory with timings of 5 will PERFORM better than 1066 with timings of 7. It is a fact.
a b } Memory
September 12, 2009 5:42:32 AM

OK jit - I've gonna be looking over the CAS Latency chart at Wiki and see if a formula that's more accurate can be made.

What about my last point? 1066 w/6 vs 800 w/5?
a b } Memory
September 12, 2009 10:13:56 AM

One thing you're not accounting for is that latency is a combination of clockspeed and timings.

You can do calculations, but they aren't quite as simple as the ones you posted. Here are the right ones:

DDR2-1066 is 33% higher clockspeed than DDR2-800. This means that in an ideal theoretical case, DDR2-1066 will have 33% more bandwidth.

CAS 7 RAM takes 40% more clock cycles than CAS 5 to access data. However, the clock cycles only take 75% as much time (since the memory is at a higher clock speed). This means that the memory latency in theory will be (1.40*0.75)=1.05 times longer on the 1066 C7 than on the 800 C5.

Basically, 1066 C7 should (in the theoretical case) have 33% more bandwidth, but 5% more latency than the 800 C5. The memory controller usually can't take advantage of the extra bandwidth fully though, so in most cases, the gain won't be quite this high. Also, very few programs need the extra bandwidth, so the actual, tangible gain will be quite small.
a b } Memory
September 12, 2009 12:43:38 PM

cjl - glad to have the input. This is exactly what I was looking for when I posted my simple formula. Thanks!
a b } Memory
September 12, 2009 3:07:31 PM

That is just it, the timings are "hard" numbers, you change them and it has a positive or negative effect on the speed of the memory, instantly. Bandwidth and bus speeds though are very relative, the only time a user will experience the gain in extra bandwidth or bus speed is if they really need it. Most gamers, and home users will never ever need the extra bandwidth, so even though it is there, you don't realize much gain. Its like the difference in running dual channel or single channel, everyone knows and wants to run dual channel mode. The data bus width is exactly 2 times wider(64 bit to 128 bit) but in actual real life performance, again the difference is very, very small, we are talking like 3% for most people, you simply cannot tell the difference.
You would need to jump 2 or 3 tiers in bus speed to match the performance on a gaming or daily use machine, to equal the performance of a 2 number drop in CAS.

And to go even a little further, and be completely fair with my assertations, an AMD platform can respond better to the higher 1066 bus speed, because of the hypertransport bus and the memory controller on the processor. IF the programs and applications can use the extra bus speed.

An Intel machine, (not talking about i5 or i7 here) does not repond to higher bus speeds as well. The memory on an Intel platform must communicate with the processor over the front side buss, which is usually 333mhz or less. The memory controller is on the board, not the processor. So using memory that runs on a faster bus is kind of useless. Not saying that AMD is better or Intel is better, just stating how they commuicate with the memory, that is all. The i5 and i7 platforms finally moved the memory controller onto the processor and started using a hypertransport type of bus setup as well.

Now, having said all that, yes I do run 1066 memory. My advice, AMD? If you have fast 800mhz memory, it is not worth spending money to upgrade, at all. If you are buying new, yes buy 1066, but hunt for 1066 with very tightest timings and low voltage combination you can find. CAS latency of 5 is very good on 1066 memory.
If you are looking at latency of 6 or 7, you are better off looking some more, or just getting fast 800mhz memory.
Intel rig that predates the i7 paltform? Buy the fastest 800mhz memory you can find, again with low voltage requirements.
December 25, 2009 9:54:52 PM

:o  no your all wrong because the fact is the limit of the memory banks is different for every chip type on the ram just like buying a Graphics card from a different company because each brand company makes there boards up of different compoents. For exmple: exlir is different from samsung chips but both may be used on the same board by a different manufacturer!!

Another way of putting it is team and OCz modules may have the same chips on them so they will perform very similarly

1066MHz is 800MHz chips hard modeded buy the company but the model of the chip determines the extent on OC capability....

Plus a MB which can adjust performance level adjust will really lower the latancies even if the timings are high!
say from performance level 11 down to 7 will reduce latancies plus each level lower is equal to 300MHz read speed.
if on everest the read speed was 6200 at perf 11 then at 10 it would be 6500 read speed like the FSB clock it has a big impact on latencies and access time coooooll!!!
March 6, 2010 4:40:16 AM

Can you plz help me as em going to purchase memory for my new pc i have the choices of following plz suggest which one is better:

-2GB 800MHz DDR2 Non-ECC Low-Latency CL4 (1GB x 2) 4-4-4-12
or
-2GB kit DDR2 1066MHz Non-ECC CL5 (1GB x 2) 5-5-5-15

and plz clarify that memory with higher MHz with lower latency timing is better or higher MHz with higher latency timing is better.
according to me

Higher latency = 5-5-5-15
Lower latency = 4-4-4-12

and also suggest higher MHz with lower latency memory such as 1066 MHz 4-4-4-12
or
2GB 800MHz DDR2 Non-ECC Low-Latency CL4 (1GB x 2) 4-4-4-12
which one is faster.....Thnx
March 7, 2010 11:31:43 AM

Truth be told, 1066 memory is often overclocked 800 memory. They really could be nearly identical, just that the one is OCed at the factory.
!