835mhz cas 4 vs 1000mhz cas 5

Which memory settings should i use?

  • 835mhz cas 4 @1.9v

    Votes: 16 50.0%

  • 1000mhz cas 5@ 2.2v

    Votes: 16 50.0%
  • Total voters
    32
G

grimreeferx09

Limp Gawd
Joined
Jul 22, 2008
Messages
355
i just got the gskill pi ram from newegg, and it can do 835 mhz @ cas 4 4 4 12 with 1.9v and 1000 mhz @ 5 5 5 15 with 2.2v
is one explicitly better than the other? or should i just go with the lower voltage one?
 
What are the other system specs? In my case, I run the Q6600 @ 9 x 400, so I don't have a need to overclock the ram past 800 and would rather have the lower timings.
 
I'd say it would be preferable to run the settings that require the least voltage, which would be 835MHz at CAS4 in your case.
 
Run some benchmarks and fnd out which performs fastest, but the general rule of thumb is choose memory timings (primarily the MHz rating, not the actual CAS timing stuff) that match the FSB 1:1 or as close as you can get to that ratio for best performance. I had trouble believing that myself recently, but...

It really is true that matching or bringing the RAM to FSB ratio as close to 1:1 - with 1:1 being the perfect match, of course - will offer a considerable boost in system performance overall.

Technically the 4:5 ratio with the 835 would offer better performance (80% match, sorta) than the 2:3 (66% match, sortakinda) with the 1000. The additional speed of the 1000 over the 835 would balance out in the long run, but the 835 RAM can probably be dropped to lower actual CAS timings to add just a little bit more boost.
 
ok, i ran everest:
ddr2 1000 cas 5
read 7999 MB/s
write 6078 MB/s
copy 5890 MB/s
latency 60.3 ns

ddr2 835 cas 4
read 7714 MB/s
write 6075 MB/s
copy 5585 MB/s
latency 60.2 ns

it's pretty close. looks like 1000 is faster, but not sure if its worth the extra volts
 
due to the architecture, real world performance is believed to be better with closer to 1:1 with tighter timings.. 835 mhz cas4
 
Will the RAM do the 1000mhz at less volts? Most stuff these days will do it at about 2.0v or 2.1v. 2.2v should not be lethal but check your spex from the manufacturer to be sure.

And, your quad core CPU can use that extra bandwidth under certain circumstances so if you are comfortable with it, I'd suggest you run the 2:3 for DDR1000.
 
Those of you claiming 1:1 is the best option need to learn how to bench for yourself.

1:1 is the slowest possible ratio to run.

Always aim for the highest speed, as Intel systems thrive on bandwidth, not timings, & definitely not 1:1.

This can easily be shown with Everest if you want a simple quick example, but you can bench with Super Pi, WinRAR, or games to see this as well.

It's nothing new, it's been reality at least since the Core 2 Duos first came out.

Sadly, i still hear people spreading mis-information around that 1:1 is the best, etc, when in reality, it's the worst ratio you can run*

*nForce chipsets runnings synced 1T would be the exception.

Obviously, in many cases your RAM cannot run fast enough to be run higher than 1:1, & for those cases, by all means, run 1:1.
But suggesting 1:1 when higher ratios are available makes no sense at all.

There will be certain cases when let's say, you can do either DDR2-800 4-4-4 max vs. DDR2-889 5-5-5, & in cases like that, it'll be close, & will likely be faster to do DDR2-800 4-4-4.

But generally speaking, faster speed CAS 5 will beat slower speed CAS 4, etc.

It's just how it works with Intel.

Now for grimreeferx09, if you really cannot do DDR2-1000 at lower vDIMM, i'd prefer to run the lower speed with lower voltage myself.

But does it really need that much voltage to do DDR2-1000?

I would expect it to do that speed with similar voltage to DDR2-800?

If not, it's your call, but you'll find thru benching that generally, bandwidth > timings, unless the two option are really close together in speed (but different timings).
 
sardonic.n7 said:
Those of you claiming 1:1 is the best option need to learn how to bench for yourself.

1:1 is the slowest possible ratio to run.

Always aim for the highest speed, as Intel systems thrive on bandwidth, not timings, & definitely not 1:1.

This can easily be shown with Everest if you want a simple quick example, but you can bench with Super Pi, WinRAR, or games to see this as well.

It's nothing new, it's been reality at least since the Core 2 Duos first came out.

Sadly, i still hear people spreading mis-information around that 1:1 is the best, etc, when in reality, it's the worst ratio you can run*

*nForce chipsets runnings synced 1T would be the exception.

Obviously, in many cases your RAM cannot run fast enough to be run higher than 1:1, & for those cases, by all means, run 1:1.
But suggesting 1:1 when higher ratios are available makes no sense at all.

There will be certain cases when let's say, you can do either DDR2-800 4-4-4 max vs. DDR2-889 5-5-5, & in cases like that, it'll be close, & will likely be faster to do DDR2-800 4-4-4.

But generally speaking, faster speed CAS 5 will beat slower speed CAS 4, etc.

It's just how it works with Intel.

Now for grimreeferx09, if you really cannot do DDR2-1000 at lower vDIMM, i'd prefer to run the lower speed with lower voltage myself.

But does it really need that much voltage to do DDR2-1000?

I would expect it to do that speed with similar voltage to DDR2-800?

If not, it's your call, but you'll find thru benching that generally, bandwidth > timings, unless the two option are really close together in speed (but different timings).
Click to expand...
so... do you have any proof?

i said real world performance. obviously 1:1 is one of the slower ratios, although back in the day it was not the slowest ratio. i am well aware that bandwidth tests show notable increases with running higher speeds vs lower timings. i posted it right after he ran everest, which i don't consider a real world performance test.

i have heard many reports that due to the core 2 architecture, lower timings (including PL) & lower speeds was better, and have never seen any non-bandwidth tests otherwise. i never bothered to test it myself, as all of my applications are gpu limited anyway so it doesn't matter.
 
I do my own testing for "proof", so can you; it's the best way to learn what improves performance.

Real world differences are pretty small.

Synthetics like Everest do tend to mirror real world results, just with more obvious results, which makes it a nice easy way to compare.

Obviously, you can run games benches to show the same thing, but it tends to be a lot more time-consuming.

Of course, that leads to the "why bother with higher RAM speeds" question then.

In response to that i'd ask why bother OCing your video card or your CPU or why the better motherboard or the faster HDD.

A lot things do indeed provide rather small real world improvements, yet we enthusiasts still do them anyway...

I have no problem with 1:1 being used, as it's easy & allows for use of cheaper RAM, or in cases where very high FSB is needed, it's sometimes the only choice.

What i have a problem with is people saying it's the best/fastest ratio, when that's simply not true; it will generally lose to all the higher ratios.

If voltages needed are similar, why would you want to run the slower configuration?
 
sardonic.n7 said:
I do my own testing for "proof", so can you; it's the best way to learn what improves performance.

Real world differences are pretty small.



Obviously, you can run games benches to show the same thing, but it tends to be a lot more time-consuming.


If voltages needed are similar, why would you want to run the slower configuration?
Click to expand...
Good points IMO. On Core 2, when I have tested 1:1 vs faster, ie 2:3 etc 2:3 is typically anywhere from 2% - 5% faster in real world testing. This was with a dual core CPU. Using the right applications, a quad core and especially a quad with a large cache can make use of good bandwidth.It shojld equal or exceed that 2% - 5%.

The testing included, video editing and rendering, audio compressions, WINRARing, etc. It was not synthetic stuff.
 
Mike Clements said:
Good points IMO. On Core 2, when I have tested 1:1 vs faster, ie 2:3 etc 2:3 is typically anywhere from 2% - 5% faster in real world testing. This was with a dual core CPU. Using the right applications, a quad core and especially a quad with a large cache can make use of good bandwidth.It shojld equal or exceed that 2% - 5%.

The testing included, video editing and rendering, audio compressions, WINRARing, etc. It was not synthetic stuff.
Click to expand...

i'd like to see tests of 1:1 with tighter timings including PL vs 2:3 with looser timings.myself
 
LOL.....I did it over 2 years ago and my copies are long gone.

Based on your signature, you have the perfect system to do the testing yourself if you'd like to find out what works best for you.
 
could you even adjust performance level 2 years ago separate from fsb strap?

the performance level used to be set by the strap, and each strap had difference ratios, infact many had the lowest PL timings in a strap that did not have a 1:1 ratio, only higher. today is a different story.
 
DoomzDayz said:
could you even adjust performance level 2 years ago separate from fsb strap?

the performance level used to be set by the strap, and each strap had difference ratios, infact many had the lowest PL timings in a strap that did not have a 1:1 ratio, only higher. today is a different story.
Click to expand...

Two years ago not likely on most boards.
On boards that automatically set tRD, you'll usually end up with a higher tRD @ 1:1, which is another reason not to run 1:1; latency is always worse as a result, even with tighter timings (due to looser tRD).

Even on newer boards though, like the P5Qs for example, you cannot run tRD as low @ 1:1 as you can at 5:6 or any other ratio really, even when using the same straps.

On my board for example, you usually have to run tRD 2 values higher @ 1:1 vs. 5:6 (333 strap) or 4:5 (266 strap), 3:4, & so on.

So by running 1:1, you are not only losing bandwidth, but you often are losing a tighter tRD as well.

Again, this can vary from board to board & will vary sometimes, but as a general rule, 1:1 being the slowest ratio is going to hold true.

There are reviews out there that have measured RAM speeds vs. timings, etc., but they tend to be lacking in details of what subtimings, etc. were used, & as different boards have different options, you really should be doing this yourself, which is what i've always done.

I don't presently have the time to try to do a comparison on my current mobo, & TBH, i have no real desire to as i already know what the results will be from past testing.

But if i find some time, i'll post the results here.
 
yeah, i was just showing how he probably didn't really have that data. but yes, i was mostly talking about tighter timings particularly tRD / PL vs higher frequencies. on my board/ram i can run tighter PL/tRD timings than higher ratios at the same voltages, and having seen some data on how important tRD timings are, thats where my justifications came in. lower timings > higher MHz
 
Everest is a good benchmarking tool for memory. Another good program is Winrar's benchmark. Run that and see which memory speed + timing gives you the higher score. I'm guessing that setting the memory to 1000mhz will beat out 835mhz under Winrar.

If the difference is negligible, run it at lower speed since it will require lot less voltage. Or, you can try to lower the voltage at 1000mhz.

Under gaming situations, 1000mhz with 5 5 5 15 timings had virtually no advantage over 800mhz with 4 4 4 12 timings(fsb:ram = 1:1) from my own testing.
 
I know, these aren't real world.

But they are quicker to run, & i was running them to show someone something, so i figured i'd post 'em here too.

P5Q Deluxe bios 1406
Q9550 @ 8.5x400
Memory subtimings are all the same other than fine delay & tREF which were on auto. Those do not affect the scores by more than a few points though.

tRD (Performance Level) was set as low as it could be for the different RAM speeds.

DDR2-800 4-4-4-12 tRD 8 (1:1 333 strap)
DDR2-960 5-5-5-15 tRD 8 (5:6 333 strap)
DDR2-1000 5-5-5-15 tRD 7 (4:5 266 strap)
DDR2-1066 5-5-5-15 tRD 7 (3:4 400 strap)

You can pretty sum it up exactly as i mentioned.
The higher the better, regardless of timings.

Read, Copy, & Latency all get better as the speeds go up.
Write is pretty much purely FSB based, so it doesn't change.


8.5x400%20DDR2-800%204-4-4-12%20trd%208.JPG


8.5x400%20DDR2-960%205-5-5-15%20trd%208.JPG


8.5x400%20DDR2-1000%205-5-5-15%20trd%207.JPG


8.5x400%20DDR2-1066%205-5-5-15%20trd%207.JPG
 
You must log in or register to reply here.
Back
Top