333DDR vs 400DDR benchmarks

[THUGSROOK]

333DDR vs 400DDR

SYSTEM
p4 1.8a @ 2.4ghz / 133fsb / 1.65v
Asus P4S333 SiS645 chipset (bios flashed with a P4S533 bios)
Corsair XMS PC2700 cas2 2.7v - 1x 512mb

SciSoft Sandra 2002 Cas by SPD
4:5 / 333DDR - 2457 / 2422
4:6 / 400DDR - 2829 / 2787

 

[jason112]

XP1800@1734MHz, 213 FSB. 2,1V
256Mb Twinmos PC2100 2,2V
Abit KR7A-RAID

SiSoft Sandra 2002

3213/2901


Check it out at slapps.com

 

[THUGSROOK]

jason112 - awesome score!

goes to show ya what a 213fsb and a real 1:1 ratio can do for ya!
better ram would really kick up your score.
im testing cas settings right now - still havent broken 3000/xxxx yet tho.

 

[XBoxLPU]

<<

Asus P4S333 SiS645 chipset (bios flashed with a P4S533 bios)
>>



Why flashed with different bios ?

What kind of improvments does the P4S533 give ?

 

[THUGSROOK]

mostly for the 4:6 and 1:2 memory ratio options.
many reports that their systems run better with the P4S533 bios.

 

[XBoxLPU]

Dang it.. wish I had some Corsair XMS memory now

I did the flash today, and everything went fine. Well besides the OC.

I can't get any higher than 133 fsb with the memory ratio set to 4:6

Though with 4:5 I am running at 148 FSB and have about the same scores as with the P4S333 bios.

Also, Windows IS SNAPPIER, actually a little faster. I really like this P4S533 bios!

and btw, like you said in your other thread, KingMax will not boot with those memory settings.

 

[THUGSROOK]

hiya Duff,
i agree.
with the 533 bios i am able to run 4:6 upto 137fsb/411ddr max.
as weird as this may sound tho - gaming benchmarks are just as fast (or faster) running 4:5 140fsb/350ddr.

im sure ppl stuck running below 133fsb are getting better use out of 4:6 then we are.

im still waiting for a 3:4 ratio above 133fsb

 

[XBoxLPU]

Windows also seems to slower using default memory config. I am using 3,3,7, and auto. It is snappier than default for sure!


I got excited about having 4:6 ratio and 150 FSB.. that WOULD KICK! but can't seem to do it yet. Though that 3:4 ratio would also be awesome.

 

[THUGSROOK]

im not using the 4:6 ratio either.
4:5 and lower cas settings are just as good for me.

im starting to see how async memory ratios are at a disadvantage. (cas disadvantage) intel is planning on 133 and 166fsb cpus in the future. but you know us overclockers - we are already there
a 1:1 memory ratio would yield the better scores mhz/mhz, ...we need 200fsb cpus!

in the mean time - the closer the fsb/ddr ratio is the better off you are in terms of latency performance.
the wider the ratio, the worse it gets

<EDIT> BTW... sandra scores do not reflect this, gaming benchmarks clearly do tho.

 

[Athlon4all]

<< in the mean time - the closer the fsb/ddr ratio is the better off you are in terms of latency performance.
the wider the ratio, the worse it gets >>

I'm not sure where you're getting that. See, the fsb/ddr ratio is simply the BIOS' way of configuring the DDR/fsb clocks, while all that really matters is the frequency, and there are only 2 configurations where the fsb and the Memory clock are synchorous for the P4. The first is a 400MHz fsb with PC1600 DDR, the other one is a 533fsb with PC2100 DDR. See, the P4's fsb is QDR so the real frequency for the fsb is 100 (400fsb) and 133 (533fsb), and for DDR, the real frequencies are 100 (PC1600), 133 (PC2100), 166 (PC2700), and 200 (PC3200). And if anything, the latency for DDR400 is going to be worse than PC2700 because for DDR, the latency increases as frequency does. That's why in benchmarks you see minimal improvement from PC2100 to PC2700, and all why you see next to no real world improvement from PC2700 to DDR400. Just my 2c

 

[THUGSROOK]

Athlon4all - im not really sure how to explain it. but ill try.

these are purely hypothetical scores from a hypothetical benchmark, but please bare with me...

2400mhz 100fsb/3:5 166-333ddr = memory score 100, gaming score 100
2400mhz 125fsb/3:4 166-333ddr = memory score 104, gaming score 107
2400mhz 133fsb/4:5 166-333ddr = memory score 105, gaming score 109
2400mhz 166fsb/1:1 166-333ddr = memory score 111, gaming score 121

notice how the scores would go up as the memory ratio closes? its running the same 333ddr speeds in all 4 tests so why the difference?

obviously this is purely theoretical. the only way to truely prove this would be with an unlocked CPU.
but from all the testing ive done, and information from other peoples testing, im 99% sure that the above chart would be very close to correct.

when i tested 2466mhz 137fsb 4:6 411ddr setting vs. 2520mhz 140fsb 4:5 350ddr settings, the later kicked the formers butt in every test i could throw at it - except for memory tests of course.
why?

how could an extra 54 cpu mhz be so much better then an extra 61mhz ddr?
A) it doesnt really need that much memory bandwidth.
B) higher cas latency due to memory ratio makes it worthless.

i choose B

*i dont read about AMD systems too much, but from what i understand, async ratios aint so great on those systems either.

im not trying to push any theories on anyone, im basically just looking for someone else to confirm this

(Cas 2/2/2/5/1 333ddr is just as good or better for overall system performance then Cas 2.5/4/4/7/2 400ddr) or you could just say cas 2.5/4/4/7 anything, is crap

 

[WheelsCSM]

This is a pretty good review that shows how performance scales with mem bandwith. They even compare Athlons and P4s, asynch vs. synch. Someone else posted it in another thread a couple days ago, sorry I don't remember the name to give them credit.

DDR400 review

 

[THUGSROOK]

a very nice read!
pretty much confirms what ive been trying to say
the article doesnt get into tweaking the cas settings, which would most likely widen the gap even further.
(the wider the ratio - the tougher it is to tweak cas settings, if at all)

the conclusion was:
high fsb and low memory ratios (4:5 or 3:4) are best.
1:1 ratio would be optimal but we cant typically run a 166-200fsb yet.
3:5 and 4:6 are the worst.

thx for the link!

 

[MindDetergent]

I just ordered a stick of mushkin ddr400 to upgrade my crucial ddr266. I get windows registry errors or program load errors at anything higher than what I'm at now. I even upped the ram voltage (2.7) and use default ram timing while messing around with the RAM ratio. My cpu can hit 150 fsb @ 1:1 ratio while passing prime95 for several hours, yet when I start to load stuff I get those annoying errors.

I'm hoping this can fix my problem. Wish me luck? Any feedback would be appreciated.

1.6a @ 2337 (146fsb), 3:4
P4S333 (flashed with P4S533 1006c bios)
420w Channel Well psu
256 Crucial PC2100

Thanx!

Edit: I figure the ddr400 would gurantee a successful OC from ddr333. Also, hoping to use some aggressive timings on the mushkin.

 

[Duvie]

hey there is no 3:4 ratio offered even in the bios for p4s533....are you sure you don't mean 4:3???


keep us informed on how high the mushkin will oc....i would be interested, especially in the ability to use the 4:6 ratio at 138 or 140...my kingmax runs fine only under 410mhz ddr, so it limits me....4:6 @138 would be 414mhz ddr and 140 would be 420mhz ddr...

 

[MindDetergent]

oops. Just checked in my bios, its 4:5.

I just noticed I'm at default vcore at 146fsb:Q Snazzy

 

[Duvie]

that would pretty awesome...are you sure you didn't use the onboard jumper to boost volatge, cause it won't show up in the bios...you just need to take it into account it has been boosted by so much...does asus pcprobe say 1.5v???

If it is 1.5v I think that would be one of the better performers I have seen...Damn get that memory and start pushing...I know often when you hit the limit it stops fast interms of each fsb bump starts to require more and more power (vcore)...i would think you could get to 2.6ghz...

I could only get to 2.160ghz from my 1.8 at default and that is tested running prime for 12 hours...however I was able to hit 2.52ghz at 1.775v prime tested...I got an additional 360mhz after I hit my default limit....


heres to you and 166fsb...tell us all if the board switches to a 1/5pci divider at that point....

 

[MindDetergent]

I'm not using the jumper or wire trick. PcProbe reads 1.456vcore.

You're getting me hyped. I really hope this ram helps.

btw...thanx for the replies.

 

[THUGSROOK]

wow, thats one helova 1.6a ya got there! :Q