memory timings w/A64 article suggestions needed

[kryptobs2000]

I've heard some interest on this subject, but not much information to show for it, so I made some of my own. Any suggestions to add are welcome, hope you guys like :

Half-Life 2 Stress Test:

640x480:

DDR400:

1.5-2-2-10 - 143.29
2-2-2-10 - 143.25
2-3-3-10 - 140.94
2-4-4-10 - 134.93
2.5-2-2-10 - 140.39
2.5-3-3-10 - 136.09
2.5-4-4-10 - 132.70

DDR500:

1.5-2-2-10 - 152.63
2-2-2-10 - 152.13
2-3-3-10 - 147.22
2-4-4-10 - 144.72
2.5-2-2-10 - 147.76
2.5-3-3-10 - 145.78
2.5-4-4-10 - 139.98

1280x960:

DDR400:

1.5-2-2-10 - 137.64
2-2-2-10 - 137.07
2-3-3-10 - 135.20
2-4-4-10 - 131.48
2.5-2-2-10 - 133.85
2.5-3-3-10 - 132.41
2.5-4-4-10 - 129.83

DDR500:

1.5-2-2-10 - 142.48
2-2-2-10 - 142.97
2-3-3-10 - 140.82
2-4-4-10 - 138.09
2.5-2-2-10 - 142.03
2.5-3-3-10 - 139.09
2.5-4-4-10 - 135.67

3dmark 03:

DDR400:

1.5-2-2-10 - 12063
2-2-2-10 - 12023
2-3-3-10 - 11985
2-4-4-10 - 11979
2.5-2-2-10 - 11981
2.5-3-3-10 - 11976
2.5-4-4-10 - 11934

DDR500:

1.5-2-2-10 - 12264
2-2-2-10 - 12230
2-3-3-10 - 12197
2-4-4-10 - 12152
2.5-2-2-10 - 12222
2.5-3-3-10 - 12144
2.5-4-4-10 - 12123

3dmark 01:

DDR400:

1.5-2-2-10 20820 Lobby High - 112.9
2-2-2-10 -21118 Lobby High - 113.5
2-3-3-10 - 20602 Lobby High - 109.5
2-4-4-10 - 20019 Lobby High - 106.5
2.5-2-2-10 - 20809 Lobby High - 110.7
2.5-3-3-10 - 20357 Lobby High - 108.1
2.5-4-4-10 - 19769 Lobby High - 104.7

DDR500:

1.5-2-2-10 - 21958 Lobby High - 121.7
2-2-2-10 - 21704 Lobby High - 121.0
2-3-3-10 - 21221 Lobby High - 117.1
2-4-4-10 - 20950 Lobby High - 114.4
2.5-2-2-10 - 21702 Lobby High - 119.5
2.5-3-3-10 - 21219 Lobby High - 115.8
2.5-4-4-10 - 20750 Lobby High - 112.6

SiSoft Sandra 2005:

DDR400:

1.5-2-2-10 3116
2-2-2-10 - 3105
2-3-3-10 - 3101
2-4-4-10 - 3097
2.5-2-2-10 - 3100
2.5-3-3-10 - 3093
2.5-4-4-10 - 3088

DDR500:

1.5-2-2-10 - 3880
2-2-2-10 - 3883
2-3-3-10 - 3873
2-4-4-10 - 3863
2.5-2-2-10 - 3876
2.5-3-3-10 - 3868
2.5-4-4-10 - 3859

SuperPi 4m:

DDR400:

1.5-2-2-10 3:38
2-2-2-10 - 3:39
2-3-3-10 - 3:42
2-4-4-10 - 3:44
2.5-2-2-10 - 3:42
2.5-3-3-10 - 3:46
2.5-4-4-10 - 3:51

DDR500:

1.5-2-2-10 - 3:30
2-2-2-10 - 3:30
2-3-3-10 - 3:31
2-4-4-10 - 3:33
2.5-3-3-10 - 3:33
2.5-4-4-10 - 3:35

[jayjaya29]

SuperPi, 3dmark01, Bandwidth test in Sandra.

[kryptobs2000]

I was planning to run 3dmark 01, lobby high, superpi, sandra, HL2 Stress test @ 640x480, and 1280x960, and with the above mentioned settings, so far thats my rough idea, looking for more input, especially on the tests before I get started though.

[=[PULSAR]=]

Really looking forward to this thread krypto, should definitely made a sticky when the data is up. A good addition would be something along the lines of 2-2-2-5 @ 250mhz and its equvilant @ 2.5-3-3-10, you know what I'm trying to say because alot of people are having hard times choosing between tight timings and mhz.

[ezrael80]

i run Vx 260@2-2-2-11 (CPU 260x9) and got something like ~25500 in 3dmark2001 se now i'm runing Gskill 280@2.5-4-3-7 (CPU 280x9) and got ~23000 ... but Everest got +- the same results...

[kryptobs2000]

i run Vx 260@2-2-2-11 (CPU 260x9) and got something like ~25500 in 3dmark2001 se now i'm runing Gskill 280@2.5-4-3-7 (CPU 280x9) and got ~23000 ... but Everest got +- the same results...

I find that hard to believe. with almost 200mhz more clockspeed on the cpu, yet slacker timings, you lose 2500 points? I'm not saying ur wrong to cover, But from my experience 200mhz higher cpu clockspeed would seem to make more of a diff. than tighter timings.

But anyways, I've gotten started on all the testing. So far here is what I have:

3dmark01, 3dmark03, sandra bandwidth scores, and HL2 Stress test at 640x480 and 1280x960. And I showed the Lobby Highfps as well from 01.

I decided against doing 05 as I didn't see a significant difference to justify it and that would add about another hour of testing so I didn't think it would be worth it.

The timings being tested are:

2-2-2-10 -
2-3-3-10 -
2-4-4-10 -
2.5-2-2-10 -
2.5-3-3-10 -
2.5-4-4-10 -

I will also be doing all of the tests at both 200mhz and at 250mhz.

I kept tras the same so that the results would be more even. I will also try testing the common tras values to show 5 is not the best choice.

I might also do some tests with a cas of 1.5, and I'm considering to do a few to see which has more impact on performance, either trcd or trp. I havn't thought of how to go about doing that yet, as right now I'm trying to get the main tests knocked out, I am about halfway through with that so far. I will not do these as extensive as the rest though because this is taking alot of time as it is.

edit: If I get some tccd in the future, I will also update this with the mentioned timings with a cas of 3, and probably a higher frequency as well.

[shadow_dweller]

good job you doing over there krypto.

from my experience it's better to have higher mhz compare to tighter timing. just my opinion though. coz i ran sisoft mem test comparing 220mhz at 2-2-2-10 against 280mhz at 3-4-4-10 and the 280mhz shows alot more bandwidth compare to 220mhz although 220mhz is run at a much tighter timing.

i didn't test the 3dmark score though.

all the best in your testing

[krille]

220mhz at 2-2-2-10 against 280mhz at 3-4-4-10 and the 280mhz shows alot more bandwidth compare to 220mhz although 220mhz is run at a much tighter timing.

NO WAY!!
You got higher bandwidth with higher FSB? That was something new!
Please, don't tell me you were doing this w/ a 1:1 divider?

[Snowman89]

NO WAY!!
You got higher bandwidth with higher FSB? That was something new!
Please, don't tell me you were doing this w/ a 1:1 divider?

ust look in overclocking he ran 350fsb i think 1:1 2800mhz and got almost 8121 and with my bh-5 @ 268 1.5-2-2-4 i get 8122 :p
btw pls test the diffrent of 1.5 and cas 2

[kryptobs2000]

Ok, I've updated the original post with all my results. The cas 1.5 testing varied alot half the time it did better, sometimes no improvement and sometimes it was about as bad as 2-3-3. I found this wierd, but I think it was most likely because the tras was too high, I will also be writing an article on what I found for hwlogic, I'll post the link if you guys are interested when it's done.

So what d'you guys think, lemme hear ur thoughts. And like I said, somewhere down the road I'll be adding cas 3 and 2.5 @ higher mhz as well, so I'm still open for suggestions if you have any, I think I need a break right now from this though lol, that was at least 6 hours of benching total, I'm tired.

[reject]

another set of solid datas telling me to change my rams. thanks!

[kryptobs2000]

Article is up if you guys are interested:

http://www.hwlogic.com/modules.php?o...owcontent&id=7

[Kanavit]

nice informative postk, Kryptobs2000! there is a fine balance between high clocks and low timings.