Dram Frequency Ratio Question

[Zhut]

I must start off with an apology for this question. I am new to the Intel side of life and I have what is likely a simple and quick question to ask:

AMD allowed for underclocking ram and overclocking FSB. Will my E8400 be able to run at a high FSB, while setting up a divider to run the ram at a lower mhz?

For example:

Can I run a FSB of 550 on my E8400 and my ram in the low/mid 400's mhz? Or is 1:1 the lowest ratio ram can be run at? I know there is 5:6 but is there be 6:5 ?

When I was purchasing my ram in sig I was thinking that this could be done. Now I think I am wrong. I tried using search but this question may be too easy to show. Thank you in advance.

[Evilsizer]

the lowested ram ratio for intel boards will be 1:1, unless your running a NV600/700 based board. those are the only ones what would allow for ram underclocking. as well as setting ram speed independantly of the cpu fsb.

[Zhut]

Understood Evilsizer, thank you. I was wondering if I was missing an option in my bios. So overclocking with lower speed ram (ddr800) I would likely keep a 1:1 ratio or the selection 667 in the bios when attempting to reach 4.0 on my E8400? Now I am wishing I got 1066 ram knowing this, my mistake. I guess prices will come down but then again, overclocking with 8 gigs in can be unpredictable on its own I think. :bang head

I know my current ram can do 500mhz @5-5-5-12 with 4 sticks (8gigs) but so can most 1066 I would imagine. Uggg.....

[Evilsizer]

at 5-5-5-16/18 timings may get your ram to 533mhz/ddr2-1066 speed. i highly dought your going to get 533fsb on the stock multi of the cpu. yea use a 1:1 ratio and start clocking up the cpu.

[18 is # 1]

Understood Evilsizer, thank you. I was wondering if I was missing an option in my bios. So overclocking with lower speed ram (ddr800) I would likely keep a 1:1 ratio or the selection 667 in the bios when attempting to reach 4.0 on my E8400? Now I am wishing I got 1066 ram knowing this, my mistake. I guess prices will come down but then again, overclocking with 8 gigs in can be unpredictable on its own I think. :bang head

I know my current ram can do 500mhz @5-5-5-12 with 4 sticks (8gigs) but so can most 1066 I would imagine. Uggg.....

Why? Intel processors love bandwidth and perform better with more of it. Speed>Timings

Here you go...same FSB
4-4-4 @ 1:1
http://i383.photobucket.com/albums/oo280/rkduncan/cachemem4-4-4.png
5-5-5 @ 4:5
http://i383.photobucket.com/albums/oo280/rkduncan/cachemem555.png
Read Copy, and Latency all improve.
Real world improvement depends on you applications and expectations.

[Zhut]

Thank you both for this information, I will get to trying this out. So perhaps I could try for 8x500 fsb with ram running 5-5-5-x to hit 4.0 or 8.5x471 etc. Gotcha I will have to work on this, it is fun work though! I at least know to stick with 1:1 now for my current system. Thanks

[18 is # 1]

You missed it. Raising your RAM divider will raise the MHz and give better performance. Look at the above pics again. OC in this order:
1 Max Processor Speed
2 Max RAM speed (without lowering the FSB)
3 Lower timings and voltages

[Zhut]

the lowested ram ratio for intel boards will be 1:1, unless your running a NV600/700 based board. those are the only ones what would allow for ram underclocking. as well as setting ram speed independantly of the cpu fsb.

Alright, so some of the CPU-Z screenshots I see in the E0 stepping thread for temperatures using 13:11 and 11:9 FSB:DRAM ratios are using those particular boards? The flexibility sounds awesome for these especially for people like me with 4x2gigs of ram (no strong ram OC with 8 gigs it seems) yet allows for high FSB, and the cpu speed is #1 as 18 is # 1 says. I am thinking there is a negative to these type of boards that can go in either direction with ratios? Sorry if I am misunderstanding.

[Evilsizer]

zhut,
either that or they are using DDR3 based motherboards. i took a quick look and couldnt find the thread your talking about. could you link to a few of the posts your talking about from that thread?

[EarthDog]

While 18 and 1 brings up a good point, I have still yet to find anyone prove that there is a point in running your memory faster than your FSB. Nobody could prove that "x" loads faster or I can burn this ISO faster or anything. SO I say why bother. My best benches and such (sans Pi) is all on a 1:1 ratio with tightened timings.

For the record, 680i and 780i boards can run ram unlinked from the FSB. P45 chipsets cannot.

[Zhut]

zhut,
either that or they are using DDR3 based motherboards. i took a quick look and couldnt find the thread your talking about. could you link to a few of the posts your talking about from that thread?

http://www.ocforums.com/showthread.php?t=581447&page=2

It is the thread about "posting your temps E0 stepping only" page 2 the poster has 13:11 ration and I think another has a 11:9 ratio earlier in the thread. Must be one of those 680i or 780i boards. I would have gone for one if I had known maybe. Seems they offer more possible adjustments for testing? Interesting.

poster is slb785 interestingly also running with 8 gigs on his 4.5 OC. That type of board may be well suited for 8gig Overclocks.

[Evilsizer]

right
http://www.ocforums.com/showpost.php?p=5880566&postcount=56
and
http://www.ocforums.com/showpost.php?p=5896898&postcount=60
are using 780i based board for unlinking the ram. however ocing isnt the best on them compared to intel. many will tell you the only reason to get NV based board is for SLI. However i did find when i ran a 680i for a short bit, that its memory controller. Was able to offer more bandwidth then P965 or P35,P45 not out then. i no longer have the pics of the 680i benches i did.

the best thing to do is just use 1:1 and clock it up. though i would test your ram like i suggested in post 4. run a 6x multi on the cpu up the fsb or just run a higher ram ratio to find the max of the ram.

[EarthDog]

Well the memory "multipliers" on the P45 chipsets only go UP. And most of the time can be used unless you are at stock or near stock FSB otherwise the ratios boost that number WAY over useable speeds on the memory.

[18 is # 1]

While 18 and 1 brings up a good point, I have still yet to find anyone prove that there is a point in running your memory faster than your FSB. Nobody could prove that "x" loads faster or I can burn this ISO faster or anything. SO I say why bother. My best benches and such (sans Pi) is all on a 1:1 ratio with tightened timings.
Since there is no standardized method for testing ISO burn speeds or "x" loads, most of us enthusiasts use performance appraisal programs to see how our tweaks are effecting our systems. I'm sorry you don't "feel" any difference when you change RAM settings, but Everest, Sandra, and SuperPi show higher RAM speed improves performance.

[EarthDog]

Synthetically it does... there is no doubt. But I dont play synthetic apps or SPi (but I mentioned it improves that). Im not a benchmarking ***** (not a bad thing at all, just not me). All it takes is someone with the gusto to test it, thats all. If its the same size ISO it will run the same. Average 3 runs at X speed, then 3 runs at your fastest memory speeds, if it improves REAL applications, wonderful!!

So far, nobody has stepped up and I have trouble believing people on this issue as I have done it, and it garned no real world improvements for me for what I do with my PC, which in all honesty is mostly gaming.

EDIT: It certainly doesnt hurt to run things as fast as you can, but Im not certain what exactly, REAL WORLD its helping. :)

[Evilsizer]

well someone a while back in the FAH section tested different FSB speed with different ram speeds at the same cpu speed. if he was lucky it was about 1-2secs quicker with faster fsb and faster ram speeds. i didnt see it being worth it in that area, 1:1 was just as good as the rest of his runs but maybe 2 secs slower. im going to see if i can find that thread...

[EarthDog]

1-2 secs in F@H is nothing...consdiering it takes days or hours (GPU folding) to get it done.

But at least its something REAL! Prove me wrong please, I would LOVE to run my rig like that for an actual reason!!!!

EDIT: Regarding 18's last sentence... all it shows is that it runs faster in those synthetic applications. It doesnt show any real world application improvement which is all I was ever looking for in this endeavor.

[18 is # 1]

The goal of OCing is to squeeze all the performance out of our equipment that is possible. If we were average, it wouldn't matter. But we are digital wizards shredding the new frontier of electronics. We change stuff because we can!!! :attn:

[EarthDog]

LOL... why the yellow writing?

So lets drive a car at 8k RPM doing 41 MPH instead of driving the car at 3K RPM at 40MPH. Sounds logical. ;)

Seriously though, it wont hurt anything, but I dislike hearing it makes "X" faster when in reality (until proven otherwise) it only makes your synthetic apps, which directly apply to nothing, faster.

[18 is # 1]

I sermonize in LARGE YELLOW LETTERS. Seriously though, I have heard your side before, but the fact remains that all quantifiable means show improvement. One mile per hour is faster, even if it doesn't feel so.

[EarthDog]

Quantifiable in the synthetic (benching) applications you listed, thats it.

I actually was doing more research and found a quite interesting thread over at some site, Same thing there, nobody posts proof of REAL LIFE improvements. Nobody plays SuperPi, Sandra, or w/e. And you cannot take those results and extrapolate that info into REAL WORLD PERFORMANCE. It doesnt help in the games I play, that I can tell you. Its like boosting the output of your car motor and saying it will corner better now...Someone over there had a great point. Where is the bandwidth saturated? Prove that the bandwidth @ 1:1 @ x.xxGhz is saturated and needs to be higher...I like the analogy I saw, something about "just b/c the pipe is bigger doesnt mean more water will flow through it".

Give me REAL LIFE APPLICATIONS (photshop, encoding, SOMETHING!!) that show the imrovement and I will kindly spread the word. Until then, enjoy your superior Pi and Sandra scores..........but thats all I will ever give credit for (like I already did). My advice will always be to run 1:1 unless you are benching. :)

EDIT: Oh well, this is probably better left for a thread of its own. Sorry for the threadjack OP.

[18 is # 1]

Give me REAL LIFE APPLICATIONS (photshop, encoding, SOMETHING!!) that show the improvement and I will kindly spread the word.
The standard measuring devices for the computer enthusiast industry are those we've talked about. The problem with your argument is that there is no recognized quantifiable and objective way to measure those "real world" applications.
It's kind of like saying you understand water is oxygen and hydrogen but since you can't see the atoms you refuse to believe it.

[EarthDog]

Kind of...but no.. LOL

You are trying to say there are real world performance increases b/c you are seeing performance increases in synthetic benchmarking applications. I refer your analogy back to you. You are trying to make a correlation when there isnt any.

I again refer you to my other statement...Prove to me that bandwidth is saturated @ 1:1 with an overclock. Like I said, all you are doing is making that "pipe" bigger. Just b/c the pipe is bigger doesnt mean more water will flow through it. Show me the pipe is saturated and I can buy into that.

Last time though, lets start a new thread and talk about it ok? Or feel free to PM me. Thanks man! :)

EDIT: I already listed a way to test it with ISO files. Many other reviews sites test processors that way. As far as other stuff.. Measure boot times, thats an easy one... Measure the time Photoshop comes up.. thats also an easy one. Sure its not exact, but are you really going to jump for joy that Photoshop now takes 10.2 seconds to load as opposed to 10.3?

[18 is # 1]

Sure its not exact, but are you really going to jump for joy that Photoshop now takes 10.2 seconds to load as opposed to 10.3?

Do I use fingers or toes to count .2 and .3?

[EarthDog]

Finger or toe nails? LOL

You enjoy your unproven ways though and keep believing in it. Same here. That will work out best. .:bang head :santa:

[Cluster]

Regardless of how 'real-life' the application is, timings vs bandwidth comes down to one main factor. How much effort did the programmer put into optimizing his memory usage. Most programmers today dont bother with such things (except for benchmarks), for a few reasons, the gains are minimal and new hardware makes they're optimizations obsolete, optimizations for one platform hurt another, hand-coding assembler is difficult to maintain, and using advanced compiler techniques can warrant all of the above.

And to top it all off, most programs havent got a clue how to implement hardware optimizations, and they're code ends up very poor. For this reason alone, i always go for lower latency, and try to push the bandwidth as much as i can. Of course, increasing ram chip speed, increases bandwidth as well as decreasing latency, if you want to get technical :p

For the most part though, your right Caligula, the extra 5mhz isnt going to make a difference, but it does mean that my epeen is bigger than yours, and that matters! :D

Oh, and the difference between 10.3 and 10.2 can be the difference between 1st and 2nd place... so ha

[EarthDog]

LOL cluster...

I never ever denied that it affects synthetic benches. And if that is what you are going for then of course, by all means, run it that way (see prior posts).

Im simply asking for some real world results...thats all. Sandra, Pi, Everest, are not real world applications. They are benchmark programs that dont neccesarily translate to real world performance gains. If someone shows me real world performance gains, Im all about running my rig that way 24/7. Otherwise, I will run it that way when I bench those programs for superior results.

If people are measuring how fast Photshop loads (which is what the 10.2 and 10.3 was about), then jesus, thats serious bencher!!!! Who the heck cares about that was really my point! :)

[18 is # 1]

[QUOTE=Caligula420;5911084

If people are measuring how fast Photshop loads (which is what the 10.2 and 10.3 was about), then jesus, thats serious bencher!!!! Who the heck cares about that was really my point! :)[/QUOTE]

I think there are quite a few enthusiast out there that try to sqeeze that last .05 GHz from their systems. We do it because we can. Photoshop be damned! I want my hertzes!!! (sermonizing over)