• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Clockspeed vs. Timings vs. Ratios

There is a lot of debate concerning what's most important in memory, all consisting of different combinations of timings, ratios and clock speed.

Here are some scenarios, I would like to know what is the best case and if possible some reasoning behind it.


When I first purchased my PC I picked up 1066mhz Mushkin memory and to keep a 1:1 ratio with the Core 2 Quad 6600 @ 2.8ghz I under-clocked it to 620mhz rather than 1066mhz, but the benefit was the 1:1 ratio and my timings were something like 2,1,3,8. Is this really beneficial? Is it only good under specific circumstances? Is there a definitive answer to which way is the best way to run your memory, or is a different way better under different types of use?

Thanks for your participation 😀
 
it's actually not that hard to do math on RAM speeds. you can figure it out for yourself.

1 Hz = 1 cycle per second

therefore if you take memory that operates at 400Mhz (DDR2-800) and just invert it (1/400e6) then you get 2.5 nanoseconds per clock cycle. so you can literally add up all the memory timings...

for example 4-4-4-12 would be 4 cycles, 4 cycles, 4 cycles, 12 cycles, or (4+4+4+12)*2.5 = 60ns TOTAL




if you were operating at 533MHz (DDR2-1066) then you get 1.88ns per clock cycle

so with looser timings let's say 5-5-5-15, that would be (5+5+5+15)*1.88 = 56ns TOTAL.

So you see it's fairly easy to do the math... and figure out what's faster. In this case, the DDR2-1066.




IIRC the benefit to having RAM that's faster than FSB is only good for applications that do memory/memory calculations without the aid or instruction of the CPU. since these calculations are not as prevalent as CPU+memory calculations, running at 1:1 makes the most sense in general.
 
Originally posted by: wired247




IIRC the benefit to having RAM that's faster than FSB is only good for applications that do memory/memory calculations without the aid or instruction of the CPU. since these calculations are not as prevalent as CPU+memory calculations, running at 1:1 makes the most sense in general.

There's another benefit to running a higher fsb:memory ratio, atleast on an Intel platform.

It allows you to run a lower peformance level (tRD). tRD is probably the most significant timing setting you can play around with as far as gains provided. Certainly makes a larger difference than even CAS.

OP, you'd probably be better off running the memory at 800mhz (or 1066), with more relaxed memory timings but a significantly lower tRD setting.
 
I looked around quite a bit for that tRD setting, but I believe it was greyed out on my CPU-z and I could not find it in my BIOS (IIRC) does that mean it is locked for me?

 
Originally posted by: wired247
I looked around quite a bit for that tRD setting, but I believe it was greyed out on my CPU-z and I could not find it in my BIOS (IIRC) does that mean it is locked for me?

Have an option in your bios called performance level (it's the same thing as tRD)?

If not then your motherboard must be controlling it automatically. If you raise your memory ratio it should lower on its own (or lower your strap). Also lowering your CAS (tCL) can lower the performance level too.

Use memset to observe your memory/motherboard timings in windows. It shows way more than cpu-z.

Anandtech had a nice write-up about tRD in one of their motherboard reviews.

They had a simple formula to calculate valid tRD's.

[tRD - tCL/N] > [ x/N]

where x is represented by a range of FSB's:

< 266mhz, x=0
266-332mhz, x =1
333-399mhz, x =1.25
400-465mhz, x =2.00
>= 466mhz, x = 2.75

and N is your memory ratio in a fraction. 400fsb with a memory speed of 1000mhz is a 1:1.25 ratio, which in a fraction is 5/4.

 
Originally posted by: wired247
I looked around quite a bit for that tRD setting, but I believe it was greyed out on my CPU-z and I could not find it in my BIOS (IIRC) does that mean it is locked for me?

look for the following settings:
Transaction booster
DRAM Static Read Control

http://xbitlabs.com/articles/m...eluxe-overclock_4.html
check that and some of the following pages, if you see those settings in your BIOS.
(this review was about P35-based board though)
 
Originally posted by: nevbie
Originally posted by: wired247
I looked around quite a bit for that tRD setting, but I believe it was greyed out on my CPU-z and I could not find it in my BIOS (IIRC) does that mean it is locked for me?

look for the following settings:
Transaction booster
DRAM Static Read Control

http://xbitlabs.com/articles/m...eluxe-overclock_4.html
check that and some of the following pages, if you see those settings in your BIOS.
(this review was about P35-based board though)


I have those settings maxxed out. If I change them anymore, system will not post. So I assume I'm getting the most out of my ram! 🙂
 
AT review, while excellent, is kinda erroneous (to me, at least) because tRD=5 with 400FSB/1:1/CL3 was common with 975X. In AT review, tRD=5 is unattainable under 400FSB/1:1 because DDR2 cannot do CL3. That might be true for P965 or P35, but not so for 975X. (And that's why I got rid of my early P35 board because it was slower than 975X)

http://img88.imageshack.us/img.../sandra975xtrd5ta9.jpg
http://img523.imageshack.us/im...everest975xtrd5ys6.jpg

These are, of course, auto/default parameters under 266 strap.
 
Originally posted by: lopri
AT review, while excellent, is kinda erroneous (to me, at least) because tRD=5 with 400FSB/1:1/CL3 was common with 975X. In AT review, tRD=5 is unattainable under 400FSB/1:1 because DDR2 cannot do CL3. That might be true for P965 or P35, but not so for 975X. (And that's why I got rid of my early P35 board because it was slower than 975X)

http://img88.imageshack.us/img.../sandra975xtrd5ta9.jpg
http://img523.imageshack.us/im...everest975xtrd5ys6.jpg

These are, of course, auto/default parameters under 266 strap.

I'm thinking that maybe with P35 boards, if your tRD setting doesn't follow the equation anandtech gave, they won't post period even if that tRD would be stable. Maybe the reason the boards aren't posting isnt because they're unstable with whatever tRD, just a self imposed tRD restriction set by Intel.

On my DFI P35 board, I can adjust the tRD's phases individually. At 450mhz fsb 5:6 memory ratio I cannot post at a tRD of 6 (tRD 6 doesn't pass the equation anyways). However I can post with a tRD of 7 and my phase adjustments all set to -1, meaning I'm running at a tRD of 6 just set another way (thus bypassing the restriction).
 
Back
Top