Playing with memory settings

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
With all the memory talk going on, I thought this would be a good time to throw this out there. These are just some findings, I'm not concluding anything. Thought those of you with the same setup would be interested. I wish I could've tested at higher FSB speeds, but I can't get much more than 150, so I stopped there.

For those who don't know (and I'm reading straight out of my mobo manual), here is a description of the settings. If anyone is willing to elaborate on this, please do, and do so in simple words for the simple minded like me :). Also, if anyone would like to explain how they work and how they work with each other, I'd appreciate it.

CAS Latency: controls the latency between the RAM read command and the time the data actually becomes available. After the RAM receives a read command, there is a time delay before it actually carries out the command. This clock cycle delay is the CAS Latency. Apparently, this timing setting also determines the number of clock cycles it takes to complete the first part of the burst transfer. Increasing the CAS Latency will give you a bit more overclocking room, while decreasing it will boost performance slightly. So far I've found that the performance increase you get by lowering this setting isn't much, I'd rather keep it higher and squeeze a bit more fsb to boost performance.
RAS to CAS Delay: controls the latency between the DDR active command and the read/write command.
RAS Precharge Delay: controls the idle clocks after issuing a precharge command to the DDR.
Active Precharge Delay: controls the number of DDR clocks used for DDR parameters.

System:
P4 1.6a
Asus P4B533-E
1 x 256MB Corsair XMS3200
WinXP

SiSoftware Sandra memory benches:

SciSoft Sandra - CAS . 2/2/2/6 . . . . . . 2/3/3/6 . . . . . . . 2.5/3/3/7
113fsb 1:1 226ddr = 1602 / 1602 . . . 1559 / 1561 . . . 1553 / 1555 DS6 ON
125fsb 1:1 250ddr = 1785 / 1784 . . . 1738 / 1738 . . . 1731 / 1730 DS6 ON
100fsb 3:4 266ddr = 1946 / 1945 . . . 1902 / 1903 . . . 1900 / 1900 DS6 ON
133fsb 1:1 266ddr = 1934 / 1934 . . . 1888 / 1887 . . . 1878 / 1878 DS6 ON
113fsb 3:4 300ddr = 2188 / 2189 . . . 2142 / 2143 . . . 2137 / 2136 DS6 ON
150fsb 1:1 300ddr = 2169 / 2167 . . . 2114 / 2113 . . . 2105 / 2105 DS6 ON
125fsb 3:4 333ddr = 2434 / 2434 . . . 2381 / 2380 . . . 2379 / 2380 DS6 ON
133fsb 3:4 354ddr = * FAILED * . . . . 2577 / 2578 . . . 2573 / 2573 DS6 ON
150fsb 3:4 400ddr = * FAILED * . . . . 2882 / 2882 . . . 2878 / 2877 DS6 ON

100fsb 3:4 266ddr = 1946 / 1945 . . . 1903 / 1901 . . . 1901 / 1901 DS6 OFF
125fsb 3:4 333ddr = 2433 / 2432 . . . 2380 / 2382 . . . 2379 / 2380 DS6 OFF
133fsb 1:1 266ddr = 1939 / 1941 . . . 1891 / 1890 . . . 1890 / 1889 DS6 OFF
113fsb 3:4 300ddr = 2190 / 2189 . . . 2141 / 2140 . . . 2137 / 2137 DS6 OFF
150fsb 1:1 300ddr = 2172 / 2172 . . . 2118 / 2118 . . . 2118 / 2117 DS6 OFF
125fsb 3:4 333ddr = 2433 / 2432 . . . 2380 / 2382 . . . 2379 / 2380 DS6 OFF

After I disabled all USB, LAN, and Sound:
113fsb 3:4 300ddr = 2317 / 2317 . . . 2294 / 2293 . . . 2291 / 2290 DS6 ON
125fsb 3:4 333ddr = 2564 / 2564 . . . 2538 / 2539 . . . 2536 / 2536 DS6 ON
133fsb 3:4 354ddr = * FAILED * . . . . 2736 / 2735 . . . 2733 / 2732 DS6 ON
150fsb 3:4 400ddr = * FAILED * . . . . 3050 / 3050 . . . 3048 / 3048 DS6 ON

The DS6 on/off refers to the ratio hack for the P4B533-X boards. I didn't explore all the setting combinations, I simply wanted to do some experimentation with the settings to find out what would be the best settings for me at both default speeds and overclocked.
 

SupermanCK

Platinum Member
Mar 31, 2000
2,264
0
0
ha...i have a whole table full of these @100fsb...i will post them here when i get home...
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
Originally posted by: SupermanCK
ha...i have a whole table full of these @100fsb...i will post them here when i get home...
by a "table" he means a kitchen table full of notes! - not a single piece of paper with some benchmarks :Q

;)

lol - we are a sik bunch arent we.
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
Added some more scores, boy what a difference the ratio hack made!

Thugsrook, thanks for the tip, I'll check it out during my next bunch-o-runs. Sik? Na, maybe demented....:)

SupermanCK, that would be great!

I know that these scores won't determine the best settings to use because it varies system to system, but I like looking for patterns and stuff. Plus I'm having fun...;)

Anyone want to take a crack at explaining these settings for all to understand? And the relationships?
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
Anyone want to take a crack at explaining these settings for all to understand? And the relationships?
nope ;)

1) lower is faster
2) "By SPD" in bios doesnt work correctly on most modern P4 motherboards.
3) know your own SPD specs! and be prepared to run some tests!

:D

there is also something you completely overlooked :Q
some boards have a 1T/2T option in their bios and their "speedy" specs would look more like this: CAS 2/2/2/6/1T . if you ever had a board with that "1T/2T" option you will know that going from 2T down to 1T is like going from cas 2.5/3/3/7 to cas 2/2/2/6! its an awesome improvement. (and something most ppl dont even know about - til now i guess)

;)
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
nice update!

now i can pick apart the ratio scores!
notice how @ 100 3:4 the scores are better then 133 1:1? :Q
the ratio is working very well - latency effect is not only minimal its better @ 266ddr.

i wonder how well that will scale?
say at 333ddr?
or even higher?

i think i have some tests to run....

great work Mikki! :D
 

SupermanCK

Platinum Member
Mar 31, 2000
2,264
0
0
Setup 1.6A @1614.4 MHz FSB 100.9 DDR269.1 - All readings from CPUZ
mobo: P4B266-C bio 1004 final <--- HTH :)
mem: Corsair XMS PC2700 rated cas2-3-3-6 @166FSB 333DDR
MEM timing - Sandra Mem Average of 3 Tests (int/float)
2-2-2-5 - 2030/2027 MB/s
2-2-2-6 - 2029/2027
2-2-2-7 - 2028/2027
2-2-3-5 - 2022/2022
2-2-3-6 - 2018/2017
2-2-3-7 - 2020/2019
2-3-2-5 - 2018/2014
2-3-2-6 - 2019/2017
2-3-2-7 - 2016/2014
2-3-3-5 - 2009/2010
2-3-3-6 - 2010/2007
2-3-3-7 - 2008/2004
2.5-2-2-5 - 2030/2029
2.5-2-2-6 - 2031/2029
2.5-2-2-7 - 2030/2027
2.5-2-3-5 - 2024/2023
2.5-2-3-6 - 2022/2020
2.5-2-3-7 - 2020/2018
2.5-3-2-5 - 2021/2019
2.5-3-2-6 - 2020/2015
2.5-3-2-7 - 2018/2015
2.5-3-3-5 - 2011/2010
2.5-3-3-6 - 2008/2007
2.5-3-3-7 - 2004/2002

no...i don't have any notes on my kitchen table...just food...but that was funny...:D
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
:Q

you might wanna list the motherboard on that SupermanCK - those scores hardly budge at all

:Q


nice work! :D
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
P4B533
Corsair 512mb XMS3000
Win98SE

SciSoft Sandra - CAS . 2/2/2/6 . . . . . . . 2/3/3/6 . . . . . . 2.5/3/3/7
166fsb 1:1 333ddr = 2510 / 2435 . . . 2482 / 2376 . . . 2481 / 2374
125fsb 3:4 333ddr = 2517 / 2434 . . . 2490 / 2373 . . . 2486 / 2370

133fsb 3:4 354ddr = 2709 / 2625 . . . 2685 / 2560 . . . 2683 / 2557
150fsb 3:4 400ddr = XXXX / XXXX . . . 2998 / 2858 . . . 2995 / 2854


some CAS "1T" testing in this archived thread.
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
Thugsrook, thanks for all the info!
Lower is faster, it seems to me that the two settings that affect performance most are the RAS to CAS Delay and the RAS Precharge Delay, at least for the tests I've done.
Yes, know your SPD specs! You can avoid an...eh...embarassing post.....:eek:
Didn't overlook the 1T/2T option, just didn't know about it. That's one of the reasons for this post, to learn about the settings, what they do and how they work with each other and how they affect performance and overclocking. I'm in a great situation for this post because I don't understand any of this! :) Thanks for bringing it up, it adds a lot to the post!
The ratio is working great. I'll run some more tests momentarily...


SupermanCK, great post, jeez, you hit every combo, that's awesome, I'll have to compare. Thanks!
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
Thugsrook,
Great scores, thanks for sharing those, and as always, your format is fantastic! :)
 

THUGSROOK

Elite Member
Feb 3, 2001
11,847
0
0
np Mikki :)

im very interested in this thread.
should put some memory "myths" to rest.
there is more info on that "1T" setting in that archived thread that i linked to. most i845 boards dont have that 1T setting tho. the P4B533-V board does - its called turbo. (not to be confused with p4b266 pci lock) on some other boards its called "enhanced mode" or something to that effect. usually thats the 1T setting.

wish we had control over ours :D
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
Updated some scores using 113fsb/300DDR, per Thugsrook's request. Also, I'm only doing three tests now because thru all this I haven't seen a big difference when changing CAS Latency or Active Precharge Delay settings. So i've matched Thugsrook's settings so we can compare. I may re-run my earlier tests so that everything matches.

Thugs also pointed out something strange: at 133fsb, his XMS3000 fries my XMS3200. As you can see here, my 133fsb memory scores have always been off, I would like to know why, anyone have any thoughts?

One more thing I'd like to point out: I'm using default memory voltage, and I'm adjusting my v-core to my known good settings for each FSB speed. This is just another variable that has to be taken into account.
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
Updated scores. The easier-to-read format is curtousey of Thugsrook (actually I stole it, but I'll give credit where credit's due...:) ). Am drawing some conclusions to all this, but I'm doing some research to better understand this stuff and make some statements that actually make sense. Bear with me...
;)
 

SupermanCK

Platinum Member
Mar 31, 2000
2,264
0
0
Originally posted by: Mikki
System:
P4 1.6a
Asus P4B533-E
1 x 256MB Corsair XMS3200
WinXP

SiSoftware Sandra memory benches:

SciSoft Sandra - CAS . 2/2/2/6 . . . . . . 2/3/3/6 . . . . . . . 2.5/3/3/7
100fsb 3:4 266ddr = 1946 / 1945 . . . 1902 / 1903 . . . 1900 / 1900 DS6 ON
113fsb 1:1 226ddr = 1602 / 1602 . . . 1559 / 1561 . . . 1553 / 1555 DS6 ON
125fsb 1:1 250ddr = 1785 / 1784 . . . 1738 / 1738 . . . 1731 / 1730 DS6 ON
113fsb 3:4 300ddr = 2188 / 2189 . . . 2142 / 2143 . . . 2137 / 2136 DS6 ON
125fsb 3:4 333ddr = 2434 / 2434 . . . 2381 / 2380 . . . 2379 / 2380 DS6 ON
133fsb 3:4 354ddr = * FAILED * . . . . 2577 / 2578 . . . 2573 / 2573 DS6 ON
150fsb 3:4 400ddr = * FAILED * . . . . 2882 / 2882 . . . 2878 / 2877 DS6 ON

100fsb 3:4 266ddr = 1946 / 1945 . . . 1903 / 1901 . . . 1901 / 1901 DS6 OFF
133fsb 1:1 266ddr = 1939 / 1941 . . . 1891 / 1890 . . . 1890 / 1889 DS6 OFF
113fsb 3:4 300ddr = 2190 / 2189 . . . 2141 / 2140 . . . 2137 / 2137 DS6 OFF
150fsb 1:1 300ddr = 2172 / 2172 . . . 2118 / 2118 . . . 2118 / 2117 DS6 OFF
125fsb 3:4 333ddr = 2433 / 2432 . . . 2380 / 2382 . . . 2379 / 2380 DS6 OFF

Like i mentioned in the vmod p4b533 thread...mikki your scores seem a little low to me...are you running some other programs while benching?
 

Mikki

Golden Member
Jun 13, 2002
1,488
0
0
SupermanCK, yeah, my nic, sound, and usb are all enabled. As I found out before, my usb modem is a serious hog and is probably affecting scores a bit. I'm gonna run a few tests with that stuff disabled, they should come up a bit. My purpose in this wasn't necessarily to get the greatest scores, but to show the relationship between the memory timings and fsb speeds. And it's not that I don't want the great scores, it's just that my cpu isn't one of the 'elite' overclockers, so I'm just keeping myself busy until my new one gets here. If it's a good one, then i'll get BUSY....:)

Anyway, I'll get some more test results out soon...
 

SupermanCK

Platinum Member
Mar 31, 2000
2,264
0
0
Hey...maybe i should test the exact same test except i only can do it on 2.5-3-3-7 if ddr is below 290...cause we have pretty much the same setup...at least the o/c
 

SupermanCK

Platinum Member
Mar 31, 2000
2,264
0
0
haha...sorry..never got around to it...been tried to sell some stuff over at for sale forum...
maybe i will do it just 4 ya