Sign in with
Sign up | Sign in
Your question

Which ram:fsb ratio will see best performance

Last response: in Memory
Share

What will win out? (timings are optimistic estimates, may not be achieved)

Total: 22 votes (9 blank votes)

  • 667 MHz @ 3-3-3-10
  • 16 %
  • 667 MHz @ 4-4-4-12
  • 0 %
  • 667 MHz @ 5-5-5-15
  • 0 %
  • 800 MHz @ 3-3-3-10 (probably won't hit these timings)
  • 0 %
  • 800 MHz @ 4-4-4-12
  • 16 %
  • 800 MHz @ 5-5-5-15
  • 16 %
  • 1000 MHz @ 4-4-4-12
  • 47 %
  • 1000 MHz @ 5-5-5-15
  • 8 %
March 12, 2008 8:44:58 PM

I'm sitting here with 8GB of OCZ Platinum 1000 MHz RAM, and am going to be putting it into my rig tomorrow. Because I haven't been able to find any definitive answers, I'm going to bench memory performance at a variety of ratios to find the one that works the best. My system normally is heavily overclocked but for the sake of stability I'll be bringing its clock speed down to match the system listed below:

Mobo: Asus P5KC
CPU: Q6600 @ 3.0 GHz (333x9)
RAM: 8GB (4x2GB) OCZ Platinum PC2-8000, rated at 5-5-5-15
OS: Windows Vista Ultimate 64-bit

I'll be posting results as I get them, but I'm just looking for some inital input.

What do you think will win out?
Any benching software you'd recommend? (freeware preferably)


Note: I'm not going to exceed the 2.2v maximum allowed by OCZ's lifetime warranty.
March 12, 2008 9:57:07 PM

I'm not an overclocking expert, but so far I've seen only extremely minor performance increases when pumping memory frequency or timings.
I'd go with a 1:2 ratio (333/667) and use the best timings you can get.

And as for benching: superpi or Cinebench
(Cinebench is a raytracer and uses mostly cpu and internal memory except for the little opengl test)

regards,
Grunchlk
March 12, 2008 10:05:22 PM

667 is a 1:1 ratio because it's DDR (double data rate) memory, ie its clock speed is 333 (667/2)
Related resources
March 13, 2008 6:01:46 AM

Would be nice would you share the result :)  Can't wait to see :) 

My best bet is the winner would be 1:1 ratio +1 :)  we'll see :D 
March 13, 2008 1:01:43 PM

I'm tempted to not run the 800 MHz tests out of laziness but I guess people running 800 MHz ram might be interested in seeing how it stacks up against 667... if 1000 MHz is marginally better than 667 I'll overclock the ram to 1066 to get results from the next bracket up, more than likely won't be able to pull anything better than 5-5-5-15 out of them though
March 13, 2008 8:31:05 PM

Update: I got home too late to run any serious tests, I'm just after installing the 8gb with no problems so tomorrow I'll be hitting benches
March 13, 2008 8:38:02 PM

If you can get all 8 to be stable when OC that will be interesting. Test out 2 then 4 to see if 4 cause crashes.
March 13, 2008 8:48:26 PM

This is odd... when I push the RAM up to 1000 MHz my system refuses to start regardless of what I try... stock CPU settings, more power to the RAM and Northbridge, really loose timings (6-6-6-18)... maybe my mobo just doesn't like 1000 mhz ¬.¬

Looks like I won't be able to test 1000 MHz, at which point it's safe to say that when it comes to 800 mhz vs 667 mhz @ 333 MHz FSB.... 667 MHz is the winner, as has been proven many times around the internet. I just wanted to check if making the leap from 800 to 1000 would tip the odds :( 
March 13, 2008 9:00:40 PM

Set it to a 1:1 and get the timeings tight as they will run.
When useing 4 sticks the voltage regulator for that feed is strained..so good luck.
March 13, 2008 9:05:14 PM

Trying that and loose timings @ 1:1 but whenever I set them manually it just doesn't seem to want to post. When I set it to auto it runs fine, but at 800 MHz
March 13, 2008 9:21:08 PM

Update: got it runing at 667 MHz (ie 1:1) and at 4-4-4-12. SuperPi to 1m places: 18 seconds.

Going to try to tighten the timings to 3-4-4-10

You guys are right about it being tough to push 4 sticks... not easy
March 13, 2008 9:25:12 PM

Final word: Ok I think it's time to call it a day, haven't been able to tighten the timings beyond 4-4-4-12 at a 1:1 ratio with 4 sticks, nor will it post at 1000 MHz with 4. Wasn't expecting this big of a disaster but whatever.

I set my CPU back to 3.6 GHz pretty happily, and have the ram running at 800 MHz, 5-5-5-15. Sorry I couldn't give you guys the results that I had promised, I underestimated the strain 4x2gb sticks would put on my system :( 

March 13, 2008 9:38:29 PM

ZOldDude said:
Set it to a 1:1 and get the timeings tight as they will run.
When useing 4 sticks the voltage regulator for that feed is strained..so good luck.


Agreed, although some think that higher ram to cpu ration helps it will make little if any difference. Although if your timings are lower then youll get more performance increase

As said though 4 sticks will be hard to get stable when changing off stock timings
March 13, 2008 9:51:05 PM

PLZ HELP...where is a good site where i can find out what you guys are acutally talking about...thanx for help....
March 13, 2008 9:53:39 PM

try the memory FAQ at the top of the memory secion
March 14, 2008 12:43:28 AM

@thebluedaw g http://www.directron.com/fsbguide.html (it's covering all the terms and things)

@kraynor are you able to compare 667 vs 800. As it seems that the performance difference is really tiny if any :)  My personal believe is that [1:1]+1 ratio is slightly better. IN this case of FSB 1333, RAM at 800 is a slightly better than 667. I didn't do a proper test I must admit, at least I can not see the difference in my personal experience. But I read somewhere on the net that it's like that. Darn, I miss the urls.
March 14, 2008 1:16:15 AM

Unfortunatly I can't run tests efficiently because for whatever reason my mobo refuses to post pretty often when tweaking timings and whatnot. It took close to an hour just to find it was stable at 667MHz 4-4-4-12... and a further 45 mins getting it stable at 800MHz 5-5-5-15
March 14, 2008 1:44:26 AM

thebluedawg said:
PLZ HELP...where is a good site where i can find out what you guys are acutally talking about...thanx for help....



Haha, you are a straight shooter son, I like that. [:mousemonkey:3]


Kraynor said:
Unfortunatly I can't run tests efficiently because for whatever reason my mobo refuses to post pretty often when tweaking timings and whatnot. It took close to an hour just to find it was stable at 667MHz 4-4-4-12... and a further 45 mins getting it stable at 800MHz 5-5-5-15


Are you trying to OC with 4 sticks initially? It might work best to use 2 to get the settings honed in and then throw in the other two to see if the system takes them. You might be hitting some settings that 2 modules can handle but overlooking them if you have 4 in.
March 14, 2008 2:05:03 AM

If I were you I would use the 1:1 ratio. (Q6600 overclocked to 1333 system bus.)
You said you were stable at 667MHz 4-4-4-12. Since you are using all four DIMM sockets I would make sure the command rate is 2T.
March 14, 2008 2:13:44 AM

just an aside... the MHz you listed are the effective rate. the actual speed is half of that. thats why 1333 front side bus (actually 333 X 4) is 1:1 with DDR2 667 spec. I realize that you probably knew that, but many here who are just reading don't.
March 14, 2008 2:14:15 AM

Yea... but my Q6600 normally runs at 3.6 GHz, I brought it down for stability. I have it back at 3.6 now with 800 5-5-5-15
March 14, 2008 2:17:53 AM

Kraynor said:
Yea... but my Q6600 normally runs at 3.6 GHz, I brought it down for stability. I have it back at 3.6 now with 800 5-5-5-15


Just to clarify... 9 (mulitplier of the q6600) X 400 system bus (or 1600 FSB) at 1:1 with the ram (400 MHz or DDR2 800 spec). Lots of lurkers who get confused.
March 14, 2008 2:24:12 AM

so are you going to get around to your battery of testing? I have done a little of my own and I find the best results for the apps I use is to set the ram as fast as it will go (stable, of course).
March 14, 2008 2:28:22 AM

It takes too long to get the 8gb to a stable combination of clock speed and timings. As stated above, I also can't seem to get it to run at the ram's rated 1000 MHz, as such I'd only be able to test at 800 MHz and 667 MHz, which has already been done (at Tom's as far as I remember, might have been Andandtech) and proven that 1:1 is marginally faster (about 1-2%) at the same timings, faster still at tighter timings attainable by underclocking the ram
March 14, 2008 3:05:00 AM

Yo hairy :)  I like your added info :) 

I think the 1:1 is faster is still arguable. That's why I really want to see the result. I read on some other threads and articles and it's said vice versa.

Personally I agree with hairy: "set the ram as fast as it will go".

I'm running E6750 2.66GHz FSB1333 (333MHz actual) with RAM set at DDR2-800 (400MHz actual) and 5:6 ratio (DRAM:BUS) and I can tell you that it's 100% stable and have a good benchmark (with sisoft). I must be honest that I'm not comparing to 667 myself.

Short explanation on my system:
Proc E6750 2.66GHz stock speed w/ FSB 1333
RAM Corsair PC2-6400 (DDR2-800)
FSB 1333 means that i has actual bus speed = 1333/4 = 333MHz
The RAM is PC2-6400 = DDR2-{6400/8} = DDR2-800, it has actual speed of 800/2 = 400MHz
In order to achieve the same speed with the processor bus speed, I set ratio to 5:6.
The number 5:6 are Bus : DRAM ratio = 333:400 = 5:6. If you calculate 333 = 400x5/6. (use your calculator, it won't give you exact same number, but close to it).
I'm using processor multiplier of 8 to achieve proc stock speed at 2.66 (=333x8=2.66GHz).

Gee, guess I have to do it myself. Let's see in the coming weekend if I have time (prolly no since my wife is going out of town for two / three weeks starting next Monday, so this weekend will be her time, meh :)  ) If I do I will come back here with the result.

Personally I don't really care since the difference won't be much, and not crucial for me since I'm not doing any memory intensive applications such as video/image editing and currently satisfied with my systems. But I'm really curious with this :) 
March 17, 2008 12:16:47 PM

:bounce: 
I just did a test of my own. I do CAD work and some of my routines take a long time (very long routines to clean up the drawing for model work). First I put my ram at 2:3 for a DDR2 800 spec. I ran the timings at 5-5-5-15. The test file took 870 seconds to run the routines on. (14 min 30 sec).

Then I re-booted my computer and I put my ram at 1:1 with my Q6600; My routine on the test file took 895 seconds (14 min 55 sec) I put my timings at 4-4-4-12 for the 1:1. Which is a DDR2 533 spec.

So if we take 870 X 1.03 (3%) it equals ~ 896. So basically running the ram faster helped this routine by ~ 3%. I still think that files load more crisply with the faster timing too. That would be tougher to gage though.
:bounce: 
March 17, 2008 1:35:45 PM

hairycat101 said:
:bounce: 
I just did a test of my own. I do CAD work and some of my routines take a long time (very long routines to clean up the drawing for model work). First I put my ram at 2:3 for a DDR2 800 spec. I ran the timings at 5-5-5-15. The test file took 870 seconds to run the routines on. (14 min 30 sec).

Then I re-booted my computer and I put my ram at 1:1 with my Q6600; My routine on the test file took 895 seconds (14 min 55 sec) I put my timings at 4-4-4-12 for the 1:1. Which is a DDR2 533 spec.

So if we take 870 X 1.03 (3%) it equals ~ 896. So basically running the ram faster helped this routine by ~ 3%. I still think that files load more crisply with the faster timing too. That would be tougher to gage though.
:bounce: 



Interesting...

On a somewhat different note I disabled my paging file on XP this weekend after timing startups. I found that when entering the password at the login screen until everything in Windows started up I did not see a single second increase in loading time. Was an interesting results to say the least. Overall startup is only 2 seconds quicker. I figured it'd shave 4 or 5 off.
March 17, 2008 1:42:47 PM

keeping my ram at DDR2 800 spec, I pushed the Q6600 to 2,700 (9 X 300) and found that the test file completed the routine in just over 12 min. I lost the exact time but it was barely over 12 min. I'm currious now what the difference would be if I adjusted the ram speed with the faster system bus. I doubt I'll have the time today though. ~ 12 min per test plus the startup and not being able to use the computer in the meantime = sucky experience. LOL
!