DDR3 latency

[cadaveca]

CAS refers only how long it takes for the data to get to the data pins from the ICs, when called for by the memory controller. IT doesn't really tell you much about how fast memory is going to be, just how fast it can get the first part of data. It is merely one part of the latency puzzle...

 

[Mussels]

CAS refers only how long it takes for the data to get to the data pins from the ICs, when called for by the memory controller. IT doesn't really tell you much about how fast memory is going to be, just how fast it can get the first part of data. It is merely one part of the latency puzzle...

however, it is the one that has the greatest effect on performance.

 

[cadaveca]

For reading, sure, it can have a large effect, but personally, I feel tRCD is jsut as important, as it's the time after CAS need for RAS.(CAS is columns, RAS is rows.)

tRP, the time between accessing each row, is equally important. How all three mix together really defines how fast ram is going to be, and maybe are the three most important.

 

[Hayder_Master]

Hmm... do you have Everest Ultimate? Would you mind running the benchmark tool and posting the results? 2200 is very fast but I'm curious to see what 10-10-10-28 does to it... and what voltage are you running that at? here's mine

http://i50.tinypic.com/5k3l01.jpg

Ganged mode seems to give better performance
http://i47.tinypic.com/xognqh.jpg

ok hat i just do some overclock and set the memories at the default 2200 cuz i reduce my cpu speed at 3.7 now and i will be upload the results

 

[Hayder_Master]

sorry hat im late buthere is my tests, i don't have Everest right now but this one do the job

 

[mastrdrver]

actually, i barely understand it. a guide would be good *nudge nudge*

Seconded *wink wink*

i third because im getting ready to build a new ddr3 computer i don't understand it at all im used to ddr2

All those timings that are shown for your memory in spd, xmp, bep, etc are not all the same unless different ram being compared run at the same speed. Those timings are all the number of clock cycles (in ns) needed for that particular timing to be completed. The clock cycles are dependent on the speed of the ram.

ddr3
1066: ~1.875ns
1333: ~1.500ns
1600: ~1.250ns
2000: ~1.000ns

To take something like my blue Ripjaws I had from 8-8-8-24 at 1333 and be able to get similar timings at 1600 we would do the following: (1.5/1.25) * t = nt; where t = rated timing at rated speed, nt = new timing at new speed

Given this we find that moving from 1333 to 1600 would produce timings of 9.6-9.6-9.6-28.8. Obviously since noninteger timings have not been used since ddr1 we need to round to 10-10-10-29 and thus we have come to as close a timings at 1600 as we can all the while keeping similar latency. The only thing that has been increased is bandwidth given that all other timings have also been change to reflect the increase in the base clock they all refer to. Realize too that even though tRFC on AMD is given in nanoseconds, this too is relative to the speed of the ram and 90ns at 1333 (60t) is not equal to 90ns at 1600 (72t) but a setting of 75ns would be required.

Obviously 10-10-10-29 at 1600mhz is far from preferred timings at that speed, but it does show that even with the main timings of 8-8-8-24 at 1333, 9-9-9-24 at 1600 has a lower latency though it appears otherwise (given that your sub timings follow a similar pattern). This is dealing with just memory latency though. As cadaveca pointed out earlier bandwidth that exceeds that of the L3 cache on a Phenom II is wasted as its not able to be used ever so your cpu-nb will need to be clocked higher to take advantage of this. Also too that as you increase your cpu-nb clock your latency will also improve. I think cpu-tweaker shows this as "maximum read latency" which is given in cpu-nb clock cycles. Though I'm not sure how to convert that nanoseconds.

 

[mastrdrver]

sorry hat im late buthere is my tests, i don't have Everest right now but this one do the job


http://forums.techpowerup.com/attachment.php?attachmentid=36772&stc=1&d=1278278694

If you want to see something interesting use the multithreader version of MaxxMem then compare a pass at those settings in your pic to settings of 7-7-7-21 at 1600 with the same uncore speed if possible. Then compare results.

 

[Hayder_Master]

If you want to see something interesting use the multithreader version of MaxxMem then compare a pass at those settings in your pic to settings of 7-7-7-21 at 1600 with the same uncore speed if possible. Then compare results.

i was try 7-7-7-21 at 1600 didn't work

 

[hat]

sorry hat im late buthere is my tests, i don't have Everest right now but this one do the job


http://forums.techpowerup.com/attachment.php?attachmentid=36772&stc=1&d=1278278694

I'm suprised to see ~44ns access time with 10-10-10 timings... heh

thanks for delivering

 

[Hayder_Master]

I'm suprised to see ~44ns access time with 10-10-10 timings... heh

thanks for delivering

welcome bro i will see if can i do more and i will upload it