RAM divider question

[BlueAcolyte]

Ok, I have some DDR2-800 4-4-4-12 RAM. My current E4500 is overclocked to 2.93GHz. That's 266FSB.

So I can run it at 2:3 (CPU:MEM) to get DDR2-800.
Or I can run it at 1:1 to get DDR2-533.

I would assume that I could also lower the voltage on the RAM if I downclock it to DDR2-533, since that's JEDEC specs.

Which one is better?

 

[DSF]

I'm in more or less the same boat as you. Chances are you're not getting much if any benefit from the higher frequency. You may as well run the RAM 1:1, especially if it lets you lower the voltage.

I don't know that lowering the RAM voltage has much impact on system heat though. In the end it's probably six of one, half a dozen of the other.

 

[LOUISSSSS]

yea try to lower the voltage to 1.8v and run it 1:1 @ 533mhz 5-5-5-15

 

[n7]

*sigh*

1.8V is as low as you're going anyway.

You can run at least DDR2-800 5-5-5 @ 1.8V.

2:3 with the same timings as 1:1 is always faster!*

*Assuming you're not messing w/ tRD, command rate, or other subtimings

Only possible way it wouldn't be is if you somehow were possessed to loosen tRD @ 2:3 over what it was at 1:1.
Except in reality, it works the exact opposite way.

So usually, not only do you get better bandwidth at 2:3, you have a tighter tRD, which helps to provide better performance overall as well.

Sorry if this came across as annoyed, but i'm getting so tired of hearing that 1:1 is the best ratio...someday people will learn, someday...

Simply put, if you're really worried about vdimm, set it to 1.8V DDR2-800 & run as tight timings as you can, which will likely be 5-5-5 or 5-5-4.

Or, just run it how it's supposed to be, at rated vdimm DDR2-800 4-4-4...not sure how that could possibly be a bad thing when it's running the voltage it's warrantied for.

 

[spdfreak]

I'm in the same situation, but I was wondering if there is any benefit to running 1:1 if I can use a 1T command rate (MSI P6N SLI). It will not let me run 1T in its current config of 2:3 with all other timings at default (Corsair XMS2- 5,5,5,18,22).

 

[DSF]

Originally posted by: n7
*sigh*

1.8V is as low as you're going anyway.

You can run at least DDR2-800 5-5-5 @ 1.8V.

2:3 with the same timings as 1:1 is always faster!*

*Assuming you're not messing w/ tRD, command rate, or other subtimings

Only possible way it wouldn't be is if you somehow were possessed to loosen tRD @ 2:3 over what it was at 1:1.
Except in reality, it works the exact opposite way.

So usually, not only do you get better bandwidth at 2:3, you have a tighter tRD, which helps to provide better performance overall as well.

Sorry if this came across as annoyed, but i'm getting so tired of hearing that 1:1 is the best ratio...someday people will learn, someday...

Simply put, if you're really worried about vdimm, set it to 1.8V DDR2-800 & run as tight timings as you can, which will likely be 5-5-5 or 5-5-4.

Or, just run it how it's supposed to be, at rated vdimm DDR2-800 4-4-4...not sure how that could possibly be a bad thing when it's running the voltage it's warrantied for.

I"m not saying 1:1 gives you the same bandwidth, but what difference does it honestly make in everyday use?

 

[n7]

Originally posted by: spdfreak
I'm in the same situation, but I was wondering if there is any benefit to running 1:1 if I can use a 1T command rate (MSI P6N SLI). It will not let me run 1T in its current config of 2:3 with all other timings at default (Corsair XMS2- 5,5,5,18,22).

If you can do one 1T, i'd definitely try for that, as it will indeed take extremely high speeds to beat that.
So for your case (& actually most nForce chipset situations), 1T 1:1 would be ideal, or at least 1T is ideal.

You could also try not using the sync option, which i believe you might have with your mobo, & going as high as you can in speed @ 1T.

You'll need to bench to determine if sync is better than no synced though.

Originally posted by: DSF
I"m not saying 1:1 gives you the same bandwidth, but what difference does it honestly make in everyday use?

Not much, if any

OP asked which is better though, & i guess you can interpret better in different ways.
I took it to mean better as in performance better, you took better as in easier/lower voltage.

Neither of us is wrong per say; just different goals i guess.

 

[DSF]

Originally posted by: n7

Originally posted by: DSF
I"m not saying 1:1 gives you the same bandwidth, but what difference does it honestly make in everyday use?

Not much, if any

OP asked which is better though, & i guess you can interpret better in different ways.
I took it to mean better as in performance better, you took better as in easier/lower voltage.

Neither of us is wrong per say; just different goals i guess.

Ok, just wanted to make sure I wasn't missing out on some kind of performance/spreading the wrong info. I haven't read up a whole lot on tweaking tRD.

 

[LOUISSSSS]

in conclusion... theres no observable speed increase in running any ram speed faster than 1:1. good day =]

 

[Pelu]

You should try to test it yourself... 1:1 or 2:3

 

[spdfreak]

Originally posted by: n7

Originally posted by: spdfreak
I'm in the same situation, but I was wondering if there is any benefit to running 1:1 if I can use a 1T command rate (MSI P6N SLI). It will not let me run 1T in its current config of 2:3 with all other timings at default (Corsair XMS2- 5,5,5,18,22).

If you can do one 1T, i'd definitely try for that, as it will indeed take extremely high speeds to beat that.
So for your case (& actually most nForce chipset situations), 1T 1:1 would be ideal, or at least 1T is ideal.

You could also try not using the sync option, which i believe you might have with your mobo, & going as high as you can in speed @ 1T.

You'll need to bench to determine if sync is better than no synced though.


I'll do some benchmarks and see which is better. I've read that 1T makes a huge difference.

 

[Mondoman]

Originally posted by: n7
...

2:3 with the same timings as 1:1 is always faster!*

Since the FSB is bottlenecking the bandwidth, I can't see that you'd get any bandwidth boost. As for latency, any links that you could provide to tests showing latency improvements/performance boosts with the faster memory bus would be welcome.

 

[spdfreak]

Can anyone recommend a memory bandwidth benchmark? Is Sandra still used or is there something better?

 

[spdfreak]

Using Sandra I got 5.37 and 5.39MB/s at 800 (2:3 ratio) and 2T command rate with all other timings at default. At 533 (1:1) I got 5.41 and 5.49MB/s with 1T command and all timings at default. So the 1T does have a big effect. Do you think I can tighten up the timings a bit since it is running at 533?

I think my cpu (e4500) still has some headroom so I might try running the fsb up a little and see what I get. I want to run it as fast as I can (stable) at default voltage.

 

[myocardia]

Originally posted by: spdfreak
Using Sandra I got 5.37 and 5.39MB/s at 800 (2:3 ratio) and 2T command rate with all other timings at default. At 533 (1:1) I got 5.41 and 5.49MB/s with 1T command and all timings at default. So the 1T does have a big effect. Do you think I can tighten up the timings a bit since it is running at 533?

You should be able to run 4-4-4-12 or 15 at those speeds. Then again, because of the 1T command rate, you may very well not be able to go any lower with your timings. You'll just have to find out for yourself.

 

[Yellowbeard]

Originally posted by: LOUISSSSS
in conclusion... theres no observable speed increase in running any ram speed faster than 1:1. good day =]

Actually there is if you care to take time to do the testing. It is an admittedly small difference but I have seen as much as a 5% increase in memory intensive applications running dividers vs 1:1. So, if the system is stable at both settings you certainly don't lose anything and can potentially make small gains using dividers vs 1:1.

 

[spdfreak]

Originally posted by: myocardia

Originally posted by: spdfreak
Using Sandra I got 5.37 and 5.39MB/s at 800 (2:3 ratio) and 2T command rate with all other timings at default. At 533 (1:1) I got 5.41 and 5.49MB/s with 1T command and all timings at default. So the 1T does have a big effect. Do you think I can tighten up the timings a bit since it is running at 533?

You should be able to run 4-4-4-12 or 15 at those speeds. Then again, because of the 1T command rate, you may very well not be able to go any lower with your timings. You'll just have to find out for yourself.

Yep, when I checked the SPD stuff in cpuz I found that since I had left all the timings at default, it automatically lowered them to 4,4,4,12. So if I can get the fsb up a little and keep the 1T, I should see a little more improvement. I'll play with it tonight.

 

[Yellowbeard]

Command rate is only good for synthetic benchmarks. It will certainly have less effect on real world usage than more memory speed (aka using a divider) will. Real world 1T is virtually indistinguishable from 2T.

 

[spdfreak]

So, it doesn't really matter which I use... At 533 I should have no problem playing with the fsb, so I guess I will leave it there.

 

[VulcanX]

Quick question, in my case of having the Corsair XMS2 4-4-4-12 2.1V Memory, 2x1Gb Kit, what should i do, should i possibly drop my voltage, and speed and get it at 1T seeing how it makes such a difference?
What you guys recommend i do? Bcoz i am not fully happy with that memory as it is anyway, they call it low latency but in actual fact you must run it at the speed they say otherwise it runs crud!

 

[Yellowbeard]

Originally posted by: VulcanX
Quick question, in my case of having the Corsair XMS2 4-4-4-12 2.1V Memory, 2x1Gb Kit, what should i do, should i possibly drop my voltage, and speed and get it at 1T seeing how it makes such a difference?
What you guys recommend i do? Bcoz i am not fully happy with that memory as it is anyway, they call it low latency but in actual fact you must run it at the speed they say otherwise it runs crud!

First, in your specific case you should not need to change the RAM voltage and timings to get to 1T. Command rate is a function of the chipset and a newer system should easily run 1T at the settings you have listed.

Also, 1T makes a difference in benchmarks only. There is no real world performance gain. You will lose performance by slowing your memory speed.

Third, I'm not sure what you mean about the low latency part.