Madshrimps Forum Madness

Madshrimps Forum Madness (http://www.madshrimps.be/vbulletin/)
-   FAQ / INFO / HOW-TO (http://www.madshrimps.be/vbulletin/f13/)
-   -   INFO: Memory Timings (http://www.madshrimps.be/vbulletin/f13/info-memory-timings-127/)

jmke 25th May 2002 17:14

INFO: Memory Timings
 
Well Im not up to much this morning so I thought Id post up some info on memory timings. It can be quite confusing for people who arent quite as into hardware as some of us with a plethora of seemingly random letters and numbers . Anyway, the key to a succesful overclock and getting the extra speed were all after is memory timings. It can make a HUGE difference in FSB and overall system performance. All the reading I took are based on sisoft sandra 2001TE as the new 2002 is a tad too random for my liking. Anyway, enough BS, onto the info.

All tests were done at 9.5 x 184mhz and 3.25vddr. Maximum timings except 2T command rate net a result of 1048/1155.

4 way interleave should always be enabled. Its very unlikely that you can make up your memory speed in FSB gain by turning this off, so Id recommend it to always be set to Enabled.

Queue depth makes very little difference to overall stability, so it should always be set to maximum or 4-Level.

Drive strength should always be set to Lo unless youre having serious stability problems. Setting it to Hi seriously compromises your overclock.

Now onto the important stuff.

Cas rating. This isnt as important as you might think for DDR systems. With Cas rating set to 2.5 you will lose approximately 15/23 from your memory score at the settings this test was run at. It will gain you approximately 6mhz FSB which is usually more than enough to gain an overall 30/30 ontop of your memory score. I recommend this to be set to 2.5 in order to gain another few mhz FSB.

TRP. Again a seemingly unimportant setting for memory performance. With TRP set to 3 instead of 2 my memory scores dropped by 6/8. It can gain you another 5mhz or so FSB giving a large boost in scores. I recommend this to be set to 3.

TRAS. This setting has no discernable effect on bandwidth. If youre running over 150mhz it should ALWAYS be set to 6T or data corruption can occur. This usually shows itself as a corrupted registry. Has no effect on FSB either. I recommend this to be set to 6T.

TRCD. This setting makes little difference to bandwidth but a massive difference to FSB. Setting to 3T loses 12/14 over 2T but gains a whopping 8mhz FSB. I recommend this to be set to 3T.

Command Rate. This setting is a hit or miss affair. Some people find they can run it at FSBs so high the air is thin, whereas others such as myself can only run it up to 173mhz. It seems as dependant on other components such as pci cards, agp cards and motherboard as it does on memory. Experiment with this setting as it gives a large performance boost. You may find that combining this with setting some of the above settings at 2 will give you the best overall bandwidth. I cannot recommend a setting for this as it is too dependant on the individuals hardware.


General stuff

Always set your ddr voltage to the maximum allowed by the board, or if Vmodded upto 3.25v for complete safety. If you dont mind a little risk upto 3.4v is safe most of the time. You will gain little extra from going over this.

Remember that it is not only your memory which limits your fsb. FSB can be limited by your cpu, board, pci cards, and very rarely agp card. Find your maximum fsb by removing all pci cards, setting all adjustable voltages to maximum and memory timings to minimum. If your cpu is unlocked then set your multiplier to around 6x to ensure you are not cpu limited. Once your maximum FSB is found then you should add your pci cards back in one at a time incase one doesnt like running at a high speed. Once this is done then change your memory timings up from minimum in this order :

Command rate first - it may work, it may not. It can cause allsorts of problems if set to 1T at too high an FSB causing an OS reinstall worst case. I always now leave this setting at 2T as most of the time I can achieve higher bandwidth by running lower timings and higher FSB.
4 Bank interleave - if this setting causes you to crash then lower your fsb until it doesnt. It is the MOST important memory timing and gives massive boosts in scores.
Cas Rating - can add a small amount to your memory scores, but the largest boost of the remaining timings. Its not worth dropping more than 2 or 3 mhz to enable this setting as thats all it takes to overturn the boost you get.
TRCD - also adds a small amount to your scores. Not worth dropping more than 2mhz in order to enable this setting.
TRP - adds an almost unnoticeable amount to your scores. Dont drop more than 1mhz to enable this setting.
TRAS - always 6T, no effect on clocking or bandwidth.
Queue Depth - always 4 level - no effect on bandwidth or clocking.


That pretty much covers everything. Theres a small table below to show you how much FSB (approximately) you will have to gain to gain more bandwidth than you lose by lowering a setting.

Cas 2.5 - gain 4mhz to increase bandwidth over setting to 2
TRCD 3 - gain 3mhz to increase bandwidth over setting to 2
TRP 3 - gain 2mhz to increase bandwidth over setting to 2
2-bank Interleave - gain 9mhz to increase bandwidth over setting to 4bank.
No Interleave - gain 11mhz to increase bandwidth over setting to 2 bank and 20mhz or more to increase bandwidth over setting to 4-bank.

After tweaking everything I managed to increase my FSB to 192mhz and gain a score of 1112/1201. A nice increase in speed overall.

Hope this helps somebody out there at least .

PiLsY.

credits go to Pilsy@icrontic

The Senile Doctor 27th October 2002 08:32

well, the thing that's cool, is that you can set everything to cas2.5, 3-6-3 and if you go up 5 or 6 mhz fsb, you regain it all?
mmmm. why did we go throug so much trubble in the old days then?

TerAngreal 27th October 2002 09:51

'caus we didn't know then ?

The Senile Doctor 27th October 2002 12:11

it's not right. there's discrepancies, sisoft mem scores suck, they don't tell you anything so how can you rely on them when tweaking your system...

jmke 27th October 2002 13:24

older versions of Sisoft were more accurate.. when they started calculating the buffered speed etc things went haywire

DUR0N 17th November 2002 13:20

when you use memtest86, the bandwidth is displayed in the upper left corner of your screen, i think that number is quite accurate. :)

jmke 18th May 2003 11:38

Intel P4 2.4 Ghz, 400Mhz FSB, 512K L2 Cache CPU
Soyo SY-P4X400 Dragon Ultra Platinum
ATi Radeon AIW Video Card
Crucial 256 MB PC2700 DDR RAM (Provided by Crucial)
Maxtor DiamondMax Plus 60GB D740X Hard Drive
Windows XP Professional Service Pack 1

http://www.techwarelabs.com/reviews/.../index_3.shtml

Observations:

*This can't be shown in the recorded results, but from my observations during testing, I noticed the general trend that as memory timings are set more and more aggressively, Sandra would reach it's steady-state score quicker and quicker (A gradual decrease from 5-7 trial runs to achieve a steady-state value to about 2-3 runs)

*The overall bandwidth increase from the slowest memory timings to the fastest memory timings (1333 --> 2603) was approximately 95% (1270 MB/sec).

Check the article for all the info!
http://www.techwarelabs.com/reviews/.../index_3.shtml

Quote:

The differences in memory bandwidth concerning CAS latency were non-existent (and it is just as likely that any recorded performance gains are attributed to random events, as performance gains were not at all consistent). There was no significant gain in memory bandwidth from adjusting CAS latencies.



All times are GMT +1. The time now is 03:08.

Powered by vBulletin® - Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO