help FSB and RAM ratio overclocking

[∞Velocitymaster∞]

I just recently built my self a super rig
Mobo; evga 680i SLI
2 Corsair Dominator 2.2v at 1066 speed
QX6700 CPU
2 8800GTX in SLI
800watts enemrmax PSU
I want to have my cpu from 2.66 at 3.2, no more since i'm on air cool
The whole day i have been playing around with overclocking of my cpu and ram
I like to have 1:1 ratio but it doesn't happen unless with lower ram speeds
the FSB is always at 1280Mhz "settings i use is Unlink"
7:12 ram @ 1096
4:7 ram @ 1119
9:16 ram @ 1137
3:5 ram @ 1066
and finally 1:1 ram @ 638
those are the settings that my mobo will let in that range. Example 1124 or 1120 it will just be 1119 for s certain range then it will bump to 1137.
Whats the best option? I don't want my ram speed lower then 1066 or higher then 1137, I try it at 1200 but memtest shows errors after 10 min in the test. I did ran memtest at all those speeds for like an hour with no errors.
Also my latency is stock 5-5-5-15

 

[SDraconis]

I'm going to venture a guess and say that you're probably better off letting your RAM run slower and sticking with 1:1. Anyone with more experience want to chime in?

 

[WintersFury]

I've played around with memory settings with my C2D setup. And faster memory improved results with the cpu at the same speed.

I know a lot of people say to run 1:1, perhaps that was the case with older cpu's. But from what I've found the faster memory speeds give more performance.


I'd suggest testing each of the memory speeds with a few benchmarks to see what gives you the best performance and using that.

 

[BillParrish]

and finally 1:1 ram @ 638

that implies an FSB of 319 x mulit of 9 = 2.8GHz

Dude, you are UNLINKED that means you can just type in whatever memory speed you want, forget the divider stuff, not sure where that came from. Your memory clock is now completely independent (unlinked) from any relationship with the FSB for the CPU. You can just set your effective memory speed to whatever you want.

1. Enter the BIOS by restarting your computer and pressing DEL during POST.
2. Navigate to the Advanced Chipset Features„FSB & Memory Config
3. Make sure SLI-Ready Memory is set to Disable
4. Set FSB – Memory Clock Mode to Unlinked.
Now the MEM (DDR), MHz field can be changed
5. Enter your desired memory frequency value and press ENTER.
6. Press F10 to save changes and exit.

This came from Nivida and is from their overclocking guide. I dont have this board so maybe there is something simple I am missing, it dont get much easier than that

 

[BillParrish]

And faster memory improved results with the CPU at the same speed.

I QUOTE]



Cant argue with that We were talking about a bit more complicated scenario and frankly I am too brain fried to get in to it. Absolutely, you are correct in the above but we had/have an entirely different situation in mind.

HAHA on me lol I didnt even mention the 1;1 divider and synch memory operation in my reply to the OP, guess I was too stunned at how easy it is to mess with the memory on that board.

 

[∞Velocitymaster∞]

that implies an FSB of 319 x mulit of 9 = 2.8GHz

Dude, you are UNLINKED that means you can just type in whatever memory speed you want, forget the divider stuff, not sure where that came from. Your memory clock is now completely independent (unlinked) from any relationship with the FSB for the CPU. You can just set your effective memory speed to whatever you want.

1. Enter the BIOS by restarting your computer and pressing DEL during POST.
2. Navigate to the Advanced Chipset Features„FSB & Memory Config
3. Make sure SLI-Ready Memory is set to Disable
4. Set FSB – Memory Clock Mode to Unlinked.
Now the MEM (DDR), MHz field can be changed
5. Enter your desired memory frequency value and press ENTER.
6. Press F10 to save changes and exit.

This came from Nivida and is from their overclocking guide. I dont have this board so maybe there is something simple I am missing, it dont get much easier than that

Thanks that was very help full. Unlink didn't really know that there was no relationship with the FSB. My multiplier is x10 by default and haven't done nothing. FSB 319.8X10 = 3198mhz or you actaully say 3.2ghz. I have the Quad core which you can change the multiplier.
I thought SLI ready memory is better. My memory is EPP and is enable. If I disable SLI will that reduce performance in ram or relation to my SLI config?

 

[PiTHON]

the multiplier is 10, not 9, so 266mhz FSB X 10 = 2.66ghz stock, 320mhz fsb x 10 = 3.2ghz cpu speed

320mhz ram speed DDR = 640mhz, with dual channel = 1280mhz
320mhz bus QDR = 1280mhz

looks linked to me :\

OP said "I like to have 1:1 ratio but it doesn't happen unless with lower ram speeds"

well 1:1 ratio is 640mhz on your ram speed.. with dual channel is 1:1 with your bus

 

[BillParrish]

Honestly I have no idea. Hopefully an 680i expert will swing by.

I will get you the link to guide, it might be in there, I didnt read the whole thing but had found and saved it for reference several weeks ago. Take me a minute to find again on Nvidias site and i will edit this with the link and you will know as much as I do about that board, pretty slick stuff, wish I had one to play with.

be right back


www.nvidia.com/docs/CP/45121/nforce_680i_sli_overclocking.pdf

have fun.

 

[PiTHON]

the sli ready memory enables access to "Enhanced Performance Profiles", which is pretty much an extended list of the SPD settings, if the motherboard supports it and you enable it it will access this extended list of memory timings for the speeds you are running

if you are overclocking most people just set the settings themselves, also not all memory is sli ready, so maybe thats why nvidia says to disable the setting..

its completely unrelated to the sli on your video cards, i have no idea why the decided to call it sli ready memory because its fairly confusing :\

on a 680i motherboard i was playing with once, enabling the sli ready memory set the timings to tight for the clocks i was at because memtest would fail after a few minutes... disabled it and it passed.

 

[calebb]

I'm going to venture a guess and say that you're probably better off letting your RAM run slower and sticking with 1:1. Anyone with more experience want to chime in?

Here's a post I made in the memory forum earlier today - I actually meant to post it in this thread


A lot of 4-4-4-12 @ 800MHz memory will also do 5-5-5-15 @ 1066MHz.

For example, my PC8500 Ballistix:

1065MHz @ 5-5-5-15:

vs.

856MHz @ 4-4-4-12:

But real-world performance? I think that I would benefit more from the lower latency than an extra few hundred MB/sec memory bandwidth.

I haven't had time to do any real-world tests on the different settings yet...

 

[KD28]

And faster memory improved results with the cpu at the same speed.

But real-world performance? I think that I would benefit more from the lower latency than an extra few hundred MB/sec memory bandwidth.

Wondering if anyone has any more data/info on the subject of higher FSB = higher performance. I know this to be true with benchmarks that I've ran but I'm not sure about real world applications Fellow forum member ran some tests on real world applications that showed no real difference with FSB being almost 100mhz apart.

I know a lot of people say to run 1:1, perhaps that was the case with older cpu's. But from what I've found the faster memory speeds give more performance.

Yah most articles I found suggest 1:1 is more stable. I guess I also found this to be true. When I tried running 4:5 ratio I got more errors in orthos at same CPU speed and voltages then when I run them at 1:1. For example when I ran my computer at 9x340 at 4:5 (so my ram was at 850) I'd get more errors then at 1:1 (where my ram was 640).

I know that its not the memory higher speeds that caused the problems since I ran mem test on the ram at 850 (actually up to 1050) at the same voltages/timings.

 

[∞Velocitymaster∞]

the sli ready memory enables access to "Enhanced Performance Profiles", which is pretty much an extended list of the SPD settings, if the motherboard supports it and you enable it it will access this extended list of memory timings for the speeds you are running

if you are overclocking most people just set the settings themselves, also not all memory is sli ready, so maybe thats why nvidia says to disable the setting..

its completely unrelated to the sli on your video cards, i have no idea why the decided to call it sli ready memory because its fairly confusing :\

on a 680i motherboard i was playing with once, enabling the sli ready memory set the timings to tight for the clocks i was at because memtest would fail after a few minutes... disabled it and it passed.

ok cool but my memory is SLI ready http://www.newegg.com/Product/Product.aspx?Item=N82E16820145043
so you said SLI ready just has to do with 4 value timings? if thats the case I will disable it. It's rated to run 5-5-5-15 2T My mobo detects that timings. I will just go to expert so I can set or change that my self to 5-5-5-15 2T my self. I also already manually set the volts to default 2.2. There are other timings besides the cas, ras trcd further down that I don't know, guess I live it auto.

 

[WintersFury]

Wondering if anyone has any more data/info on the subject of higher FSB = higher performance. I know this to be true with benchmarks that I've ran but I'm not sure about real world applications Fellow forum member ran some tests on real world applications that showed no real difference with FSB being almost 100mhz apart.

That test was FSB speeds with Quad Core performance. If he wanted to keep the test accurate he had to keep the memory at the same speed for both tests.

As I said, give it a try and run some apps that give framerate displays. It's not a huge difference but I was able to find one. Heck run 3DMark, it's a fairly good way to see if theres any difference in changes made to the system.

 

[calebb]

That test was FSB speeds with Quad Core performance. If he wanted to keep the test accurate he had to keep the memory at the same speed for both tests.

As I said, give it a try and run some apps that give framerate displays. It's not a huge difference but I was able to find one. Heck run 3DMark, it's a fairly good way to see if theres any difference in changes made to the system.

3DMark is also a synthetic benchmark - although I agree that it is more accurate than sisoft sandra memory bandwidth scores. Framerates in a real game would be a more accurate test, IMO. (Someday I'll have time - haha)

 

[KD28]

If he wanted to keep the test accurate he had to keep the memory at the same speed for both tests.

I think the point of his tests were to show that its the CPU speed that affects performance in real world aps, not really the RAM speed/FSB, thus the RAM/FSB were the variables. He was kinda dispelling the higher FSB/RAM speed = better performance mantra that is generally accepted.

As I said, give it a try and run some apps that give framerate displays. It's not a huge difference but I was able to find one.

That was my intended point/question, sorry it wasnt more clear but IF there isnt a huge difference with higher FSB/RAM speeds AND IF there is a boost in stability by staying with a 1:1 ratio, wouldnt the OP be better served by going 10x320? Since he'd achieve his intended 3.2ghz with added stability. Another beneift could be that he may not have to raise his voltages as much with ram speeds leading to lower temps?

 

[silent-circuit]

OP: if you want 1:1 you need to set RAM to linked and synced, not unlinked... Also, I imagine with an X series CPU you'll be able to hit 3.2 just fine, but you really should test in steps, just in case -- if you haven't already. 2.8, 3.0, 3.2 for example. Just to see if it behaves in Windows, then run Orthos if it boots without issue (as it should) at 3.2Ghz for a 'real' stability test.

Now, if you really want to run your RAM at rated speed, no, you can't run linked and synced. You also have to disable all the Nvidia SLI ready crap. All of it. I know you paid for it, you want it to work, etc, but it's pointless if you're trying to push above the tiny predefined overclock (of around 10% max on the CPU) they allow with it. Disable it, disable speedstep, all that. When overclocking, at least initially, you want the board trying to do as little 'on its own' as possible, since in truth it's really just more places for things to go wrong.

Set your memory to 'expert' and find the specs for those sticks. I don't have them handy. Set them up speed and timing wise to spec, make sure they get 2.2 vmem, etc. Again, you'll have to run Unlinked if you want the RAM to run at 1066 with the CPU at 3.2Ghz.




Another option might be to figure out what multiplier would give you 3.2 ghz and a non-quad-pumped FSB speed that would give DDR2 1066 RAM speeds...

Running 1:1 you would need a 533 non-quad-pumped FSB to run the RAM at DDR2 1066 (spec). That's... high. Very high. I don't know that the board could sustain it, honestly. I run 400Mhz FSB all the time, but have never tried to go higher. You could try, but I think at that speed you'd need a lot of northbridge voltage (if you could get there at all) and probably better cooling than stock. (Someone more knowledgeable please correct me if I'm wrong)

Assuming you could somehow manage 533Mhz FSB (2132Mhz quad pumped -- again, very unlikely even your very nice board could ever manage this, so it's pretty much theoretical) a 6x CPU multiplier would get you 3.198 Ghz. To reiterate, though... probably not even something you should try.

The second half of this post is just jabbering. Pay no attention to it...

 

[∞Velocitymaster∞]

No won't try to reach over 450 or 533mhz FSB for few reasons, heat, instability just to be safe.

I'm fine now. I disable Speedstep and SLI Ready. Unlink it. I just keep my multiplier at x10 and type in 1280 value for my FSB which is 320mhz. Now I want to take my ram a bit higher then 1066 stock speeds. I will try 1124 ram speed test it for 30min so far no errors and I guess I stick with that.
Also I manually punch in the timings for my ram as suggested Corsair dominator values for this memory model.