What has Asus done?? CAS4 vs CAS3 on P5B Deluxe!

[toledo]

I've checked 3-3-3-4 vs 4-4-4-4 with same alpha timing 355x9 and guess what? 4-4-4-4 was 17s faster in super pi with 711 bios.

WTF ?? .... Seems that 4-3-3-8 is faster than 4-4-4-8 , but 3-3-3-8 is slower than any timing settings at cas 4 . As i suspected the cas 3 is triggering the odd behavior

[funkflix]

SPi32M @ 450x9=4050MHz, DDR2-900 8-4-4-4.0 --> 13:08.282

SPi32M @ 450x9=4050MHz, DDR2-900 8-3-3-3.0 --> 12:57.782

Hmm, i'm so confused with my 32M results.. At the ~ same settings as u used i always get around 14min 02.xxs, 4-4-4-8, 4-3-3-8 or even 3-3-3-8 does not really affect the times!

[bachus_anonym]

I've checked 3-3-3-4 vs 4-4-4-4 with same alpha timing 355x9 and guess what? 4-4-4-4 was 17s faster in super pi with 711 bios.

Well, as a matter of fact you're right... I've decided to burn few hours on SPi32M runs and this is what's up:

Boot @ below (DDR2-533 ratio; 1:1; timmings: 4-4-4-4.0):

355x8        400x9        401x9        416x9        470x9

15:56.625       14:09.328       14:43.328       14:12.734        12:33.625


Boot @ below (DDR2-533 ratio; 1:1; timmings: 4-3-3-3.0):

355x8        400x9        401x9        416x9        470x9

16:30.219       14:39.797       14:30.922       13:59.813        12:22.781

Now, as far as I can tell, this behaviour it NOT triggered by CAS=3.0 but TRCD=3. Until we're below 401MHz FSB (e.g that 355MHz or 400Mhz above), changing 4-3-3-3.0 to 4-3-4-3.0 results in sudden boost (I know, WTF!) which results in 32M time arrive to the level of 4-4-4-4.0. In other words, from above experiment we can see, that 4-3-3-3.0 is pointless, right?

Wrong! Look what happens when we hit 401MHz. As some of you might have read earlier, 401MHz is where many of us believe, strap change occurs or at least some sort of NB's timmings' loosening happens. Apparently, this does not occur when we're already on TRCD=3! You can see, how much loss is 4-4-4-4.0 taking by going from 400MHz to 401Mhz (whole 34s), yet 4-3-3-3.0 magically gains 9s at the same moment!

Now, as some know, we can force P5B Deluxe not to switch strap at 401Mhz, by simply booting at 400Mhz and using Clockgen to ramp up FSB. I have shown a lot of results in Tony's "More P5B secrets uncovered" thread, so I'm not going to repeat them here. That "trick" gave a heatly boost of around 30s at any given FSB above 400MHz, compared to straight boot at that FSB. Surprisngly, this does not work as soon as TRCD=3! Taking above example, if we boot at 400Mhz and force 470Mhz 4-4-4-4.0 by using Clockgen, then instead of 12:33, we're going to see 12:02 (e.g. 12:05 @ 468MHz) but same thing for 3-3-3-3.0 will NOT result in a same boost and as a matter of fact, result will be worse than above shown 12:22 (e.g. 12:32 @ 468MHz)! Keep in mind, "clockgening up" to above 470MHz (32M-stable) is not an easy task and will very likely require vmch mod (my mobo has vcore, vdimm and vmch mods done) and good NB cooling.

This shows, that above 400MHz FSB (straight boot), 32M runs at 4-3-3-3.0 are somewhat faster than 4-4-4-4.0, but that's only due to strap change apparently affect only TRCD=4 cases, not TRCD=3. But below 400Mhz FSB, 4-4-4-4.0 (or 4-3-4-3.0 or 4-4-4-3.0) wins because apparently strap doesn't change until 401MHz. I have even booted as low as 266Mhz 5-3-3-3.0 and used Clockgen to raise to 355Mhz and it was exactly same result as straight boot @ 355Mhz. Would that mean, that RAM set to TRCD=3 (e.g. 5-3-3-3.0) results in a very loose NB from the very beginning??? Why? What's the purpose of it?

I'm gonna try to add somewhat better conclusion for it later... Feel free to chime in, guys!

[alpha0ne]

Thanks bachus_anonym for your "as usual" invaluable help and conclusions

[VictorWang]

you can compare 400fsb with ram=ddr2_800
and 450fsb with ram=ddr2_900
strange to, 400fsb is faster, both pi and 3d

[metro.cl]

you can compare 400fsb with ram=ddr2_800
and 450fsb with ram=ddr2_900
strange to, 400fsb is faster, both pi and 3d

i believe that is because of the strap change.

[Vikodemous]

Thank you bachus_anonym

I was wondering was going on.

[zachy]

Yesterday i've made couple of bench:

7x450 4-4-4-5 spi16m 7m34s
7x450 4-4-4-8 spi16m 7m37s
7x450 4-4-4-8 losened alphas spi16m 7m41s
8x400 4-4-4-4 spi16m 7m25s
9x355 5:4 4-4-4-8 spi16m 7m17s

Im gonna add some more asap.

Non optymized system.