What's more important for RAM speed: Dram Frequency or Timings?

MrMatt

Banned
Mar 3, 2009
3,911
7
0
which is better:

Running RAM at a frequency of 1333, with timings of 7 7 7 18, or running at 1600 with timings of 8 8 8 22?
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
No question, 1600 @ 8-8-8-22

You'll have 20% higher bandwidth and 5% lower latencies over the other timings you listed.

Now if you had tighter latencies at 1333, say 6-6-6-14 then you'd have to make a choice between better latency versus higher bandwidth as a 6-6-6-14 latency @ 1333 is 11% better than an 8-8-8-22 latency @ 1600.

But 7-7-7-18 @ 1333 is actually 5% worse latency (in ns, which is what matters) than 8-8-8-22 @ 1600, so the choice is obvious.
 

Blain

Lifer
Oct 9, 1999
23,643
3
81
which is better:

Running RAM at a frequency of 1333, with timings of 7 7 7 18, or running at 1600 with timings of 8 8 8 22?
Better for what?
You need to give us as many details as possible, in order for us to give you a coherent reply.
 
Last edited:

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
Generally speaking timings are not noticed in real world apps. Benchmark only apps like Sandra, Everest, and SuperPi do benefit from lower timings. Having said that, I would go with more bandwidth first. More memory bandwidth can lead to some real world performance differences in tasks like Photoshop, Video editing etc. Where lots of memory is used at a time and you are pushing large amounts of data very quickly into memory.

For gaming neither one will have a sizable benefit. You won't even notice it and any difference will be within the margin of error across multiple timedemo runs.Your CPU and primarily GPU will be the determining factors. So if you are not overclocking and that is not a priority, and your choice of memory impacts the amount of money available for your video card, go with the slightly slower memory. DDR3-1600 memory is going to offer more overclocking room without needing to rely on the memory to run above spec.
 
Last edited:

lothar

Diamond Member
Jan 5, 2000
6,674
7
76
No question, 1600 @ 8-8-8-22

You'll have 20% higher bandwidth and 5% lower latencies over the other timings you listed.

Now if you had tighter latencies at 1333, say 6-6-6-14 then you'd have to make a choice between better latency versus higher bandwidth as a 6-6-6-14 latency @ 1333 is 11% better than an 8-8-8-22 latency @ 1600.

But 7-7-7-18 @ 1333 is actually 5% worse latency (in ns, which is what matters) than 8-8-8-22 @ 1600, so the choice is obvious.

Informative thread.
The only thing I've been going by is "the lower the CAS rating the better". DDR 2000, 5000, and DDR 1 million doesn't achieve much in the real world besides making a "splash" in Everest benchmarks and WinRAR.

As a general rule, I have pretty much been going by this.
"Go one step above the RAM speed your processor officially supports with as low CAS as you can afford"
Lynnfield: DDR3 1600
Bloomfield: DDR3 1333
That seems to give a lot of flexibility when it comes to overclocking(and you can choose to run DDR3 1600@1333 with lower timings if you want) without having to pay an arm and a leg for it.

How do you calculate latency?

*EDIT*
Price is certainly a big factor(if not the biggest) to consider as well.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Informative thread.
The only thing I've been going by is "the lower the CAS rating the better". DDR 2000, 5000, and DDR 1 million doesn't achieve much in the real world besides making a "splash" in Everest benchmarks and WinRAR.

As a general rule, I have pretty much been going by this.
"Go one step above the RAM speed your processor officially supports with as low CAS as you can afford"
Lynnfield: DDR3 1600
Bloomfield: DDR3 1333
That seems to give a lot of flexibility when it comes to overclocking(and you can choose to run DDR3 1600@1333 with lower timings if you want) without having to pay an arm and a leg for it.

How do you calculate latency?

*EDIT*
Price is certainly a big factor(if not the biggest) to consider as well.

This is old, but timeless as it still applies even in the day of DDR3:

http://forums.extremeoverclocking.com/showthread.php?t=135503

http://superuser.com/questions/36494/ram-access-speeds-latancy-vs-bandwith

An easy rule of thumb for comparing latencies is take the specified timings and divide by the rated "speed".

So 7/1333 = 0.00525 and 8/1600 = 0.005

So the absolute latency of a 7 clock delay at 1333 clock is worse (slower by 5%) than the absolute latency of an 8 clock delay at 1600 clock.

This is why I said the OP really did not ask us to make a choice, the answer is simple because in the case of 1600 8-8-8 both bandwidth and latency are superior over 1333 7-7-7 so there is simply no question as to which is "better".
 

lothar

Diamond Member
Jan 5, 2000
6,674
7
76
This is old, but timeless as it still applies even in the day of DDR3:

http://forums.extremeoverclocking.com/showthread.php?t=135503

http://superuser.com/questions/36494/ram-access-speeds-latancy-vs-bandwith

An easy rule of thumb for comparing latencies is take the specified timings and divide by the rated "speed".

So 7/1333 = 0.00525 and 8/1600 = 0.005

So the absolute latency of a 7 clock delay at 1333 clock is worse (slower by 5%) than the absolute latency of an 8 clock delay at 1600 clock.

This is why I said the OP really did not ask us to make a choice, the answer is simple because in the case of 1600 8-8-8 both bandwidth and latency are superior over 1333 7-7-7 so there is simply no question as to which is "better".

Pretty cool.
Never knew that.
 

hanspeter

Member
Nov 5, 2008
157
0
76
The latency numbers are given in whole cycles. So you should divide by the real clock, not the effective speed.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
The latency numbers are given in whole cycles. So you should divide by the real clock, not the effective speed.

Its called a "rule of thumb" not "the exact formula" for a reason...

At any rate doing either leads you to the same conclusion.

Whether you go to all the trouble of calculating latencies in nanoseconds or if you simply ratio the effective speed to the latency in clocks you are going to arrive at the same conclusion.

(This works because the effective speed itself shares a common multiplier too.)

Give it try.
 

autumn_suns3t

Junior Member
Oct 26, 2014
9
0
0
No question, 1600 @ 8-8-8-22

You'll have 20% higher bandwidth and 5% lower latencies over the other timings you listed.

Now if you had tighter latencies at 1333, say 6-6-6-14 then you'd have to make a choice between better latency versus higher bandwidth as a 6-6-6-14 latency @ 1333 is 11% better than an 8-8-8-22 latency @ 1600.

But 7-7-7-18 @ 1333 is actually 5% worse latency (in ns, which is what matters) than 8-8-8-22 @ 1600, so the choice is obvious.

What about 12GB (1×4 + 1×8 modules) of 1600 @ 8-8-8-22 versus 8GB (2×4) of 1866 @ 7-7-7-18 versus 8GB (1×8) of 1866 @ 7-7-7--18?

I can't afford 16GB, but I am tempted to make a 12GB set.
Also, I know 1×8 is < 2×4, but buying 1×8 allows me to reach 16GB by adding amother 8GB module in the future and still take advantage of dual channel.
 
Nov 26, 2005
15,092
312
126
What about 12GB (1×4 + 1×8 modules) of 1600 @ 8-8-8-22 versus 8GB (2×4) of 1866 @ 7-7-7-18 versus 8GB (1×8) of 1866 @ 7-7-7--18?

I can't afford 16GB, but I am tempted to make a 12GB set.
Also, I know 1×8 is < 2×4, but buying 1×8 allows me to reach 16GB by adding amother 8GB module in the future and still take advantage of dual channel.

Why are you bringing up a post/thread from 2009 ???
 

DrMrLordX

Lifer
Apr 27, 2000
21,602
10,793
136
It's just another first-poster doing a thread necro. Seems to be a lot of that going around right now.
 

autumn_suns3t

Junior Member
Oct 26, 2014
9
0
0
Why are you bringing up a post/thread from 2009 ???

Sorry but I can't see how it could be better to ask what I asked in a newly-started thread than reliving this one. Any reason?
I think all this thread is still relevant in 2014, no outdated topic, and good for people to see what was written before, this is all. (I admit I didn't read forum rules much carefully, how old can a thread be it to be not buried on this forum?)
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
21,602
10,793
136
While this thread may still be relevant, the answers today would be different from five years ago, especially for things like APUs. Thread necro is usually considered to be poor form. Best practice is to start a new thread and link back to the old one if you consider it to be important.