Whats better: memory at 1:1 or 2:3?

[Gorgon]

Hi all. I have a question that I haven't seen answered anywhere.

I have a non-overclocked E6600 (thus with 1066 rated FSB). Right now I'm using my Corsair XMS2 DDR2 800 as DDR2 533 so that I get a 1:1 FSB relation. Should I Instead raise the voltage to 2.1 and put it working at DDR2 800 speeds and get a 2:3 ratio?

I've read in some article that only the AMDs actually get some benefit from using a higher speed memory when the ratio is not 1:1, but that on Intels its just better to keep it at 1:1. Can someone enlighten me on this matter?

Also, I suppose that I have to raise the voltage to 2.1 in Corsair XMS2 DDR2 800-C4 before enabling it to DDR2 800 speeds? Or does my ASUS P5B Delux sets the voltage accordingly if I leave the Voltage set to AUTOMATIC? (right now I have voltage set to automatic but I have set both the timmings and speed manually without using the SPD feature)

Thanks in advance for your comments.

 

[heruur]

Read this, the guys at anandtech did some benchmarks

http://www.anandtech.com/mb/showdoc.aspx?i=2813&p=4

 

[Gorgon]

So its better to let it at 2:3 (DDR2 800), I suppose?

Thank you for your reply.

 

[Gorgon]

Well, I'd say from what I've seen, running your memory in sync will yeild the best performance at it's rated speed, but running 2:3 offers far more bandwidth that would probably make a sizable difference in amd systems since they don't really have a sync as the mem controller is integrated, but for conroes, it's hard to say, but I'd go 2:3 up to as high as you can get it, but 1:1 if you get to 400mhz or so because the memory is fast enough at 800 and will perform better at 800 1:1 than 800 2:3 from what I've seen

But take this with a grain of salt as I'm not completely sure

Yes, but to run it at DDR2 800 at 1:1 that would mean OCing the CPU to get a 1600Mhz rated FSB and my FSB is 1066 since the CPU is not overclocked. So its either DDR2 533 at 1:1 or 800 at 2:3.

But thanks anyway.

 

[Gorgon]

Hehehe. I think I would be the first guy to actually give himself all that "pseudo-overclocking" trouble just to have DDR2 800 at 1:1 with a 1066 FSB CPU. :roll:

Nah, me thinks not good!

 

[3Ball]

What taco said is correct it just needs to be taken further. The syncing of the memory speed with the 1:1 ratio and the FSB speed is what is making the performance higher when you achieve 800mhz on 1:1 ratio. Intels fair well from the sync between the bus and the memory because this is where the communication between the two takes place. Unlike that of the AMD's where the lower timings helps more than speed because of the integrated memory controller, which is used to processor similar information to that of the FSB in current intel CPU's. That is why the 1:1 at rated speed is better, but even when not at 400mhz for the FSB the higher bandwidth in the Memory will help ever so slightly. As the article suggests. Look at it this way with the 2:3 ratio you get the full use of bandwidth that the 1:1 offers with your rated FSB speed, and then you get alil xtra just for kicks! Hope that helps!

Best,

3Ball

 

[Gorgon]

Actually, you wouldn't, some people lower their cpu mult down to 6 just to get to 500fsb since they can't cool their cpu at speeds far above 3ghz, so you wouldn't be the first person to mainly their fsb

Either way, with conroes memory speed isn't all that important due to the low latency, and k10 will improve on that even more with a 2nd gen memory controller

I know that, Taco, but they don't lower to 6 and put the FSB at 400 and stick with the same 2.4 Ghz just for the sake of a 1:1 ratio for DDR2 800! Thats just what I was saying. But Thanks for your help!

Also thanks to 3Ball for his post. It was usefull! So now I have my rated FSB atill at 1066 and my DDR2 at 800, with 2:3 ratio.

Thank you, guys!

 

[Gorgon]

do what suits you best, I personally am a performance guy, but you if think it's going to be that weird, it really doesn't make much of a difference to me

You mean that some people actually lower the cpu mult to 6 and raise the fsb to 400 and remain with the exact same 2.4 Ghz just to get 1:1 with DDR2 800? I though that with all the trouble involved (voltage tinkering, stability issues, etc) they would go all the way and actually raise the FSB above 400 to get a faster CPU (e.g. to 500, like Taco mentioned). If that is the case, my apologies.

You guys have been very helpfull. Thanks again.

 

[Gorgon]

Thanks for the info, Taco!

 

[flasher702]

For FSB architecture you should go for the speed that gives you the best latency while providing as much or more memory bandwidth as you have FSB bandwidth. At stock SPD settings DDR2-800 will likely yield tighter timings. (CAS 5 at 800mhz is faster than CAS4 at 533mhz for example) and give a performance bump. (CAS3 at 533 would be even faster though). Adding additional memory bandwidth that gets choked at the FSB will give some performance improvement by reducing NB memory controller latency very, very slightly but once you get enough memory bandwidth to fill the FSB it's all about latency.

Note that there are other timings also, I just mentioned one for simplicity and brevity. Also, you'll use more electricity and generate more heat with higher voltage and speed which you may or may not care about.

Better yet: bench the three scenarios similar to what I just outlined and tell *US* which is better

 

[enewmen]

1:1 personally runs better for me.
Anything higher and you need damn good RAM when overclocking.

 

[ZOldDude]

You said:

I've read in some article that only the AMDs actually get some benefit from using a higher speed memory when the ratio is not 1:1

With AMD Optrons (or AMD 64's) things work fastest at 1:1...as the memory AND cpu speed climb the latentcy drops.

Everest shows my Optron 146 and DDR1 G.Skill ram 1:1 with only a 30% OC chews up DDR2-800 ram in Read/Write/Copy/Latency tests even though it is only running at DDR1-520.

By the time I need a new system DDR3 ram will have been around for 3+ years so I will be bypassing DDR2 altogether.

 

[flasher702]

Just to let you know, there is also a performance increase running your memory in sync instead of async, and running your fsb at 1600mhz like I said is possible will take the fsb bottleneck out too anyways, so perhaps timings might make a difference, but running the memory in sync already adds its own increase

AFAIK memory dividers having performance penalties on FSB+NB-memory-controller is an historic problem. I'm not aware of any current platforms that have this problem. If you run the memory faster with as good or lower latency you will get a performance increase always. If you can find benchmark data with current chipsets that shows otherwise I would be most greatful. With a quad-clocked FSB and double-clocked RAM this also means you can get very large performance increases if you are running in single channel.

My radeon xpress 1250 chipset C2D mobo does not have "memory dividers" 8O . just "asyncronus".

 

[Gorgon]

Keep the discussion going, guys/gals. Its beying quite informative, although some contradictory info would benefit from some article/testing backing.