*RAM* (FSB vs Timings) in P4C800 w/ 3.2 P4

I'm building a new rig :rocker: ...

I already have a 3.2 P4, Asus P4c800 -E Delux that I will be running with water cooling and Swifttech MCW5000-PT. I am completely torn as what type of memory will give me the best performance. I could buy 3200/3500 memory and get some good timings or go with 3700/4000/4200 and get very high fsb. Am I right in saying to an extent having better timings is better than a huge fsb?

I am considering either (Kingston, corsair, or ocz) and would like to get a few opinions as to what ram (brand and model) would give me the best performance with my setup.

A frined just told me that Intel isn't about timings

I would go with corsair twinx, because you can easely run 27X fsb on 1:1 with that proc with only 2.8V en standard timinjgs !

So I would be better off if I went with some some of the higher fsb ram than with the quicker timmings then eh? Should I go with some corsair 4000 or ocz 3700 gold?

:x

ocz needs alot of ram to reach what corsair does on 2.8v

i would go for the pc4k corsair

Definately get some OCZ PC3700 GOLD....or wait a bit and OCZ Should have PC4200 right around the corner...doot doot

OCZ 3700 GOLD is performing the best right now but it will put you out in the pocketbook. TwinMOS/Winbond BH-5 3200 is also worth a consideration and it is cheap.

I have mine doing 480Mhz + with 2-5-2-2 timings and 2.8v :thumbsup:

Quote:

---

Originally posted by Bleam!
A frined just told me that Intel isn't about timings

---

It isn't?

Guess I should tell this to my Intel too :D
http://www.solidhardware.com/macci/timings.gif

Quote:

---

Originally posted by macci
It isn't?

Guess I should tell this to my Intel too :D
http://www.solidhardware.com/macci/timings.gif

---

I think what was meant by that is that you get more performance out of a higher FSB vs tighter timing

Say your ram will do 225 mhz at 2-2-2 or 250 at 3-4-4 250 is faster, just like your post , 300 at 2.5-4-4 is faster than 285 at 2-2-2

Usually, FSB wins out over timings.

How did you figure out that 300 2.5-4-4 is faster than 285 2-2-2?

Well... your SSandra scores, and Pifast are anyhow... Also , not every platform or system will get psyco overclocked scores like you ;)

In an average/modest overclock the difference is more noticeable... I ran a few tests back when I first got my IC7, but have since loaded 2 raptors in RAID, and I lost the data :(

But... suffice to say, (depending on the ram) if lowering the latency can get you 15+ more mhz, its going to be faster, if it can only get you 5-10 or so... its not.

canterwood and P4 are bandwidth beats... give em high fsb speeds thats what theyre built for.

Sandra unbuffered (real) is not better and 25.98 vs. 26.00...
within margin of error. Point to be taken...timings are everything.
So much for that "I'm gonna git me that new ddr540 that runs cas3,4,4,8"!!!!!!!!!!
Use the divider...run tightest timings possible.
The correct caveat is "Intel isn't about the divider"
C

Here is a link where a dood (our EVA2000) tested 233 at 2-3-3-7 vs 250 at 3-4-4-7 (the 1st and 3rd lines)
250 at 3-4-4-7 beats 233 at 2-3-3-7 in all tests.

http://www.fileshosts.com/pentium4/D...d/summary.html

Granted, since Intel CPU's are locked its getting more CPU at a higher FSB, but, that IS the way the Intel platform is working, and in the end more FSB+more CPU, is faster.

the whole thread is here... http://www.xtremesystems.org/forums/...415#post185415

Once I get some OCZ PC 4200, (like the day its available:D) I will see how fast I can get at 1:1 ratio with 2-2-2-6 vs 3-4-4-8 and post all the results... I think maybe 5:4 and 3:2 ram ratio's may yeild different results... Or perhaps different mobo's/bioses yield different results as well. :confused:

General rule of thumb: #5
If you cant beat the mem speed buy 30-35mhz with relaxed timings.
Then you should go with the tighter settings.
I thought everybody new this?:slobber:

Hasn't anyone ever noticed this before? Run a Sandra mem unbuf benchie at like 225fsb, cas2, 5,2,2 then 230, then 235...the numbers go up nice and neatly. Now go up to 240, but relax the timings...guess what? Lower Sandra. You've gotta' get yer fsb to like 245 before it matches 235 and tight...
This high mhz RAM is a SCAM. For example, the only processor where this ram MIGHT be useful is 3.2C and certainly the upcoming 3.4ghz at 17 multi and even then...maybe not. But for ALL you 2.4-3.0 guys out there...run high vdimm, pC3500/3700 like BH-5 at cas2, 2,2,5
C

Trdd has a higher effect on memory bandwidth than cas, so there are some trade offs that can be made....

Quote:

---

Originally posted by SpicyHuevos
General rule of thumb: #5
If you cant beat the mem speed buy 30-35mhz with relaxed timings.
Then you should go with the tighter settings.
I thought everybody new this?:slobber:

---

True... I'd say its more like 15-20 mhz, but still true. :toast:

Quote:

---

Originally posted by SpicyHuevos
General rule of thumb: #5
If you cant beat the mem speed buy 30-35mhz with relaxed timings.
Then you should go with the tighter settings.
I thought everybody new this?:slobber:

---

Not quite a general rule. With an 875P/865PE, you can beat well over 100MHz of memory speed with tight timings over relaxed timings.

Quote:

---

Originally posted by Zroc
Not quite a general rule. With an 875P/865PE, you can beat well over 100MHz of memory speed with tight timings over relaxed timings.

---

ok what am i missing here?


asus p4c800 mem @ 235 cas 2-5-2-2
asus p4c800 mem@270 cas 3-4-4-7
The 270 anhilates the 230 by huge margins.
same mem used for both settings
pcmark mem bench for
235 11932
270 12449

Quote:

---

Originally posted by SpicyHuevos
ok what am i missing here?


asus p4c800 mem @ 235 cas 2-5-2-2
asus p4c800 mem@270 cas 3-4-4-7
The 270 anhilates the 230 by huge margins.
same mem used for both settings
pcmark mem bench for
235 11932
270 12449

---

Alot of those points are from your video card. When your running your card at 235 compared to when your running it with cpu/fsb way higher at 270 the Video points will add up.

OPP

Quote:

---

Originally posted by SpicyHuevos
ok what am i missing here?


asus p4c800 mem @ 235 cas 2-5-2-2
asus p4c800 mem@270 cas 3-4-4-7
The 270 anhilates the 230 by huge margins.
same mem used for both settings
pcmark mem bench for
235 11932
270 12449

---

I'll show you what I mean...
Is that P4C800 mem@270 3-4-4-7 you mentioned at 1:1?

Quote:

---

ok what am i missing here?

---

The CPU speed (and L2 cache speed) also affects the score. And at 270 your CPU/L2 is quite a bit faster than at 235.

No doubt, alot of the improved scores are due to CPU/Cache/FSB, not memory, but with this chipset, and these multi locked CPU's that DOES play a part in improved performance, and you cant make a perfectly equal test. faster=faster, so if your ram will crap out a a certain FSB with tight timings you MAY well be able to get a better score if you loosen them and raise the FSB, each system is different, and you should test both ways to see where your own sweet spot is.

I just ran a quicky test at 220 and 230 bus to see the results... The specs are as posted in my sig, I changed only the FSB and set the ram ratio to 1:1 for test purposes

P4 2.4C@ 12x220 (2640mhz) 2-2-2-5 - 1:1
Sisoft mem - 5264/5298
3dm2k1 - 18373
pcmark2k1 mem - 9800

P4 2.4C@ 12x230 (2760mhz) 2.5-3-3-7 - 1:1
Sisoft mem - 5454/5455
3dm2k1 - 18667
pcmark2k1 mem - 10072

Only 10mhz faster for the bus speed(with higher latencies), and its quite an improvement in overall speed.

macci and opp you silly bastads
im talking 1:1 vs 5:4
280fsb 235mhz ram
270fsb 270 ram
pcmark scores at 272 1:1 vs 272 5:4
cas3-4-4-8-turbo
vs cas2-5-2-2-turbo

5:4
11189

1:1
12559

Quote:

---

Originally posted by SpicyHuevos
macci and opp you silly bastads
im talking 1:1 vs 5:4
280fsb 235mhz ram
270fsb 270 ram

---

Well there you go then, 1:1 RULES!! :D
Can't wait to try some of the new Kingston 4000, Monday:)


OPP

Quote:

---

Originally posted by SpicyHuevos
macci and opp you silly bastads
im talking 1:1 vs 5:4
280fsb 235mhz ram
270fsb 270 ram
pcmark scores at 272 1:1 vs 272 5:4
cas3-4-4-8-turbo
vs cas2-5-2-2-turbo

5:4
11189

1:1
12559

---

Here's what I'm getting at...

http://www.monsterboxpc.com/roundup2/latency.gif

I've always dug comache and QIII 1.17 Fastest, because they're still to this day really good indicators of proc and memory changes. And I always mix in 3DMark2K1 default, cuz it's so recognizable and does actually change a good deal with CPU and memory (where 3DMark2K3 is like 90% video card).

Anyway, the jist is, the hit from 2/2/2/6 to 2.5/3/3/7 is about 3% (in real benchmarks), and then the hit from 2.5/3/3/7 is about another 3%.

Now, 3% isn't a big deal in real-world usage...anything under 5% is generally imperceptible in usage. But it is a big deal in benching and overclocking.

For example, back in the roundup I did in May, I found the hit going from 1:1 to 5:4 to be 2%, and the hit from 1:1 to 3:2 to be about 5%, all other settings being equal.
Using that as a general guide, this would tell you that one could overcome those ratio hits simply with better timings.
Well, let's see:

200.5 FSB/200.5 Mem (DDR401) 2.5/7/3/3
Sandra- 4681/4683
QIII- 363.3
Comanche- 49.35
3DMark- 16031

200.5 FSB/160.4 Mem (DDR320.8) 2/6/2/2
Sandra- 4392/4398
QIII- 363.9
Comanche- 49.43
3DMark- 16123

Works out pretty well...not exact (the settings aren't the same...timings are different), but 5:4 2/2/2/6 does beat 1:1 at 2.5/3/3/7 in real benchmarks. Note that it doesn't in Sandra...but I really don't 'play' Sandra ;) I really don't like PCMark memory bench, because it's not a reliable memory bench (varies too much, and is too dependant on FSB). Besides, it's you can't 'play' that either ;)
Point here is...going after 1:1 DDR500 at bad timings is practically futile...you can beat it with better timings at 5:4.

Now, notice that was only 2/2/2/6 against 2.5/3/3/7...it will be much worse against 3/4/4/8 (my Kingston actually won't run at 3/4/4/8 in 1:1). I'd be willing to bet that 2/6/2/2 at 3:2 would equal or beat 3/4/4/8.

In a nutshell, the recent trend of trying to hit really high 1:1 scores (250+), but with bad timings, is a bad bet. 1:1 DDR500 at 3/4/4/8 would get beaten soundly by 5:4 DDR400 at 2/2/2/6. There's a really good chance that even 3:2 DDR333 2/2/2/6 would beat 1:1 DDR500 3/4/4/8 at 1:1.

Come to think of it...all this new PC4000 3/4/4/8 memory is pretty pointless. You could literally beat it with old PC2700 memory capable of 2/2/2/6. Hmm....

Good stuff Zorc:)

I'll do some testing with some new 4000 tomorrow.
The hope is that it will run a more aggressive timings then speced when useing higher voltage.

OPP

OPP my man,

Is that Kingston available for the masses already? I need to order some of the good ram....

Quote:

---

Can't wait to try some of the new Kingston 4000, Monday

---

I talked to Kingston on Thursday, because the 4 modules of 4000 I ordered that their site said was in stock wasn't. Anyway, I asked the tech if the 4000 was any different than the 3500, and their response was NO. It is the same modules, just tested with relaxed timings. So needless to say, I just ordered 2 more 3500 to go with my existing 2.

Quote:

---

Originally posted by OPPAINTER
Good stuff Zorc:)

I'll do some testing with some new 4000 tomorrow.
The hope is that it will run a more aggressive timings then speced when useing higher voltage.

OPP

---

Thanks ;)
Yeah, I hope more voltage will score you some better timings...they're needed to even justify 1:1. For reference, in my testing, RAS Precharge and RAS-to-CAS seem to have the biggest impact. Either of those above 3 seems really, really bad.

Quote:

---

Originally posted by QuadDamage
OPP my man,

Is that Kingston available for the masses already? I need to order some of the good ram....

---

I don't think so, I heard somthing about haveing 128 sticks ready:D Thats not real massive.

OPP

LOL, yeah i thought so:)

Thanks.

ZROC:
YOU can only compensate so far with timings my man.
Im talking you could maybe compensate for 25 mhz with the best over the worst timings.
ANything passsed 25mhz its just all gravy for the 1:1 setting.
Like i said when we see more people being able to acomplish this it will be accepted as the new standard for benching ang gaming on a p4 rig. The 5:4 setting will be used buy the people running microsoft office everyday:)

P.s
you dont need pc4000 to do 270+ almost any pc3500-3700 will do it.
The best memory in the world is no good without finding the sweet spot with timings.

oK,
But where is this a good thing for the "average Overclocker" with a P4 2.4C, 2.6C, 2.8C or even 3.0C?????
THis fancy shmancy PC4xxx out there won't help.
Example:
Prommy
2.8C
PC3700 OCZ Gold
Nice setup...Jonny Overclocker can post at 293fsb and bench at 291fsb. His OCZ pC3700 Gold will not post at 291fsb 1:1 (ddr582)
so Jonny runs 5:4 at 233mHz. DDR 466, how nice :) his OCZ runs it at a slick cas2.5, 7,3,3. So with his 1:1 "wunder ram" he can only run 1:1 at 264fsb...nice but <3700mhz
Running 5:4 he's at 4074mHz.
How odd...Jonny's neighbor Tommy Tweaker has the same rig with Kingston HyperX PC3500. He voltmodded his P4C800 and runs 3.1v through his sticks. He runs 291fsb at 5:4 for a nice round ddr466 at cas2, 5,2,2.
Now...Tommy's bangin' Jonny's GF cuz' Jonny's benchies and 3dmk SUCK.
Moral of story...PC4XXX ram is only worth something with 3.0C that top out <270 and 3.2C's with 16X multi. Upcoming 3.4C too of course.
Silly rabbit...this stick's for kids.
C

Quote:

---

Originally posted by SpicyHuevos
ZROC:
YOU can only compensate so far with timings my man.
Im talking you could maybe compensate for 25 mhz with the best over the worst timings.
ANything passsed 25mhz its just all gravy for the 1:1 setting.
Like i said when we see more people being able to acomplish this it will be accepted as the new standard for benching ang gaming on a p4 rig. The 5:4 setting will be used buy the people running microsoft office everyday:)

P.s
you dont need pc4000 to do 270+ almost any pc3500-3700 will do it.
The best memory in the world is no good without finding the sweet spot with timings.

---

I disagree completely. The 5:4 hit has always been tiny...around 2%. The current P4 platform has a glut of memory bandwidth. The relaxed timings hit is simply greater. Look at my numbers again...anybody in here should be able to produce the same results.
Again, here:

200.5 FSB/200.5 Mem (DDR401) 2.5/7/3/3
Sandra- 4681/4683
QIII- 363.3
Comanche- 49.35
3DMark- 16031

200.5 FSB/160.4 Mem (DDR320.8) 2/6/2/2
Sandra- 4392/4398
QIII- 363.9
Comanche- 49.43
3DMark- 16123

Again, that's just 2/2/2/6 against 2.5/3/3/7...you can already tell how much worse 3/4/4/8 will be.

Don't take my word for it, Spicy...try it yourself.

-Run you machine up as high as it will possibly go at 1:1 with the loosest timings (3/4/4/8, like you said). Run 3DMark 2K1. If you have them, run Quake III, UT2K3, Jedi Knight, Comanche...any game with a benchmark. If you don't want to, just run 3DMark2K1.

-Now, do nothing but flip it to 5:4, at the best timings you can (which you said is capable of 2/2/2/5). Run 3DMark 2K1 (and any other game benchmarks you want).

Just based on my results (and this has been repeated on many boards), your 5:4 2/2/2/5 should destroy your 1:1 3/4/4/8.

Looking forward to your results!

So whats the deal?

I have a 3.2 p4 and need to buy some new memory. Will I get more performance from some 4000 at relaxed timings or 35-3700 at tighter timings. It sounds like the 35-3700 tight timings are good for the less than 3 ghz chips while the 4000 is better for 3 ghz and above.

Is this correct?

zroc:
I have tried and tried again just like the title to the post with a 3:2 i cant beat 1:1 with my 5:4 with any combo of of mem settings:)
all my scores go down period.
be it sandra,3dmark,pifast <-suffers the most
Besides maybe Macci ,and opp there are not to many people that will be running 270+1:1 anytime soon .
and thats assuming they get a 3.2 processor.

Just like I said in my previous post, the 4000 is the EXACT same modules as the 3500. Why pay the extra $$$$$?

NeYoung,
Considering you will be using watercooling, I think it'll work well.
But you'll probably top out at about 250ish if you're lucky. More likely 245ish. Which brings up another problem...if your max OC is like 250mhz fsb, some PC35/3700 will run that 1:1 too, but at better timings. To hit 255+ on the 3.2 takes phase change.
If I was you I'd probably buy some HyperX3500 or some GEIL platinum pC3700 or something with BH-5 chips (XMS3500C2)
If your rig exceeds the capability of your ram, ebay it and buy the OCZ.
C

DJ,

I'm not saying that this 4xxx stuff is what i'm waiting for, just wanna see what it can do (timings/voltage) For example i had some of this Twinmos pc3700 ram doin 253fsb 1:1 on IS7 at 2.7v, but will not work at all on P4C800, will not do more than 212fsb on MSI, or IC7 even at 3.3v.

Quote:

---

Originally posted by QuadDamage
DJ,

I'm not saying that this 4xxx stuff is what i'm waiting for, just wanna see what it can do (timings/voltage) For example i had some of this Twinmos pc3700 ram doin 253fsb 1:1 on IS7 at 2.7v, but will not work at all on P4C800, will not do more than 212fsb on MSI, or IC7 even at 3.3v.

---

I here ya man i got the same issues:)
some works great on asus and some on abit
hell my pc3700 wont even boot period in my abit.
But works spectacular in asus.
I havnt tried pc4000 yet either.
supposed someone is sending me some but we will see;)

I'm still willing to bet, when OPP, and Macci and the rest of the OC champs get their hands on the latest PC4000, and more seriosly, OCZ PC4200, they will be setting new speed records, even with the lesser latencies involved.

yeah, but did you notice that none of these :spectacular" pc3700/4000 stuff can do cas2? Most of these sticks are cas2.5/3/3/3/7 spec'ed which means i would still choose my XMS3500 over that ocz stuff.

Twinmos pc3700 has probs with tras at 2 (if i remember correctly) and if set to 2/2/2/5 won't even boot at stock fsb on some mobos. xms/hyper x are still good stuff considering there both amd/intel compatible.

but DJ man, your 271fsb 1:1 is pretty damn impressive, just wondering what's the voltage you are pumpin thru it, 3.6v maybe?:D

Quote:

---

Originally posted by QuadDamage
yeah, but did you notice that none of these :spectacular" pc3700/4000 stuff can do cas2? Most of these sticks are cas2.5/3/3/3/7 spec'ed which means i would still choose my XMS3500 over that ocz stuff.

---

Yes, I did notice that, and I would gladly trade my XMS3500 for any of it, I would even pay extra for it. :toast:

At some point, (depending on your setup) the extra FSB+CPU makes up for the little perfomance that is lost on better latencies.

Here are my earlier test results...
P4 2.4C@ 12x220 (2640mhz) 2-2-2-5 - 1:1
Sisoft mem - 5264/5298
3dm2k1 - 18373
pcmark2k1 mem - 9800

P4 2.4C@ 12x230 (2760mhz) 2.5-3-3-7 - 1:1
Sisoft mem - 5454/5455
3dm2k1 - 18667
pcmark2k1 mem - 10072

230 at 2.5-3-3-7 beats 220 at 2-2-2-5 (for me) Now, the OCZ pc4200 stuff is rated to run at 2.5-4-4-7 at 270mhz bus... This will allow me to run 1:1 at 270(or higher) :D

isn't that ocz ram cas3 rated? anyway, i don't think it will do 270fsb 1:1 given less that 3.0V. Also tras at 4 is slow as hell and i don't think 270/tras 4 will be faster than 250/tras 2, so IMO it's not worth it. but then it's your money and it's your choice.

Its rated at 2.5-4-4-7, but OCZ has been really generous with their ratings lately.

True 2-2-2 at 250 may be faster... but what Ram is gauranteed to do that ? All of the pc4000 I have seen is either 2.5-4-4-7, or 3-4-4-7

Here are the ddr500 sticks I've heard of:
GEIL PC3700 Platinum (my own, ddr500 cas2 ,3,3,6)
Kingston HyperX PC3500, look at MrZee and others.
tWINMOS bh-5
Adata
possibly XMS3500C2, my XMS3200C2 BH-6 even does ddr480, cas2,3,3,6!
Just slip the voltage up a bit. I run 3.10v
C

Quote:

---

but DJ man, your 271fsb 1:1 is pretty damn impressive, just wondering what's the voltage you are pumpin thru it, 3.6v maybe?

---

Corsair pc3700 @ 2.85 :)
Still have not even done mem vmod to it:D
Thank you i can run it at 280 not for long though:(

280 at 2.85? looks like i'll be loyal to Corsair for another six months at least:) wondering what it can do at 3.0V+ :)

I tried 3.0 and 3.2 on my modded asus and it didnt do a damn thing. These sticks i have just dont respond to voltage like winbond sticks.

Quote:

---

Originally posted by retrospooty

Here are my earlier test results...
P4 2.4C@ 12x220 (2640mhz) 2-2-2-5 - 1:1
Sisoft mem - 5264/5298
3dm2k1 - 18373
pcmark2k1 mem - 9800

P4 2.4C@ 12x230 (2760mhz) 2.5-3-3-7 - 1:1
Sisoft mem - 5454/5455
3dm2k1 - 18667
pcmark2k1 mem - 10072

230 at 2.5-3-3-7 beats 220 at 2-2-2-5 (for me) Now, the OCZ pc4200 stuff is rated to run at 2.5-4-4-7 at 270mhz bus... This will allow me to run 1:1 at 270(or higher) :D

---

Yes, it should be, as it's 10 higher FSB and 120MHz more clockspeed, but that's not what I'm talking about.

Take that 12x230 2.5-3-3-7, drop it to 5:4, and put the timings back to 2-2-2-5. Your 3DMark2K1 will be higher (as will Quake III, Comanche, UT2K3, Jedi Knight...almost any application). Your Sandra's and PCMark mem may be lower, but who cares if everything 'real' is faster?

Here's a review out today on the Corsair TwinX-3700 stuff:
http://www.3dxtreme.org/Corsair_xms3700_twinx_p1.shtml

It is a very good read, because he uses an unlocked processor to test it with, keeping the CPU speed out of it.
He tests at 1:1 3GHz/200 2-2-2-5 against 3GHz/233 3-4-4-8 and 3GHz/250 3-4-4-8, and gets the same conclusions I do. Better mem scores, but all the real apps just take a dump.

Quote:

---

Originally posted by Zroc
Yes, it should be, as it's 10 higher FSB and 120MHz more clockspeed, but that's not what I'm talking about.

Take that 12x230 2.5-3-3-7, drop it to 5:4, and put the timings back to 2-2-2-5. Your 3DMark2K1 will be higher (as will Quake III, Comanche, UT2K3, Jedi Knight...almost any application). Your Sandra's and PCMark mem may be lower, but who cares if everything 'real' is faster?

Here's a review out today on the Corsair TwinX-3700 stuff:
http://www.3dxtreme.org/Corsair_xms3700_twinx_p1.shtml

It is a very good read, because he uses an unlocked processor to test it with, keeping the CPU speed out of it.
He tests at 1:1 3GHz/200 2-2-2-5 against 3GHz/233 3-4-4-8 and 3GHz/250 3-4-4-8, and gets the same conclusions I do. Better mem scores, but all the real apps just take a dump.

---

I see what your saying, actually , right now, and for the past week or so I have been running at 275 5:4 2-2-2-6, so I am set. One of these days when I get some time I'll play around with it some more. :toast:

I have used this board, ASUS P4C800 for months with a P4 3.0 800Mhz using Mushkin 1 Gb Dual Pack PC3200 Level 2 RAM. Combined with the Prometeia I’m able to run the CPU stable at 4032 and have reached speeds above 4200.

Mushkin's Level 2 memory jumped in price recently due to the lack of Winbond chips. Mushkin continues to use Winbond until they can find something suitable to replace it.

Only a few resellers are carrying inventory as the best has also become rare.

Quote:

---

Originally posted by SpicyHuevos
I tried 3.0 and 3.2 on my modded asus and it didnt do a damn thing. These sticks i have just dont respond to voltage like winbond sticks.

---

Just curious, what voltage would I need to feed a pair of HyperX 3500 so I can run it at 2-2-2-5 at 301fsb and 5:4 divider?

Pretty easily...maybe even 2.85v, for sure...stick it with 3.1v
C

I'm stuck in no mans land at present with my P4 3.0. It will go up to 3750 in winter which is 250FSB, and at present it goes up to 244FSB.My memory is Corsair PC3500 which will do 2-5-2-2 all day at 220Mhz but not much over 225 with any more relaxed settings, weird.

Of course I am running at 5:4 at 244 at present. If I turn it down and run 1:1 at 220 then I get 100+ more 3dmark points than at 5:4 at 220 so I want to go 1:1, however ....

When running 220 at relaxed timings of 2.5-7-3-3 I drop 200 points, so I have to make sure I get 1:1 and 2-5-2-2 and 250Mhz memory ( ie PC4000 at 2-5-2-2 ).

Of course no body does that yet, instead you get PC3200 memory 2-5-2-2 in effect. I'm tired of the memory people claiming bigger and bigger numbers for memory when they are not providing anything better than we have had for the last few months, because the chip speeds are still 5ns or 4.3ns.

I want some 4ns chips that run 250 not out of spec at 2-5-2-2 and shall sit tight until someone provides it.

Even reading this page it seems to be a lottery whether it will work well or not.

Regards

Andy

ZROC:
look at corsairs results:)
http://www.dvhardware.net/article.php?sid=1643

Quote:

---

Originally posted by SpicyHuevos
ZROC:
look at corsairs results:)
http://www.dvhardware.net/article.php?sid=1643

---

And?
None of that has any relevance to what I'm talking about...it is simply an overclocking chart...in each case, they're simply raising the FSB and clockspeed to go with it. Of course it will be faster.

None of this changes the fact that 250FSB 5:4 DDR400 2/2/2/6 will beat 250 FSB 1:1 DDR500 3/4/4/8 in 3DMark2K1, Comanche, Quake III, Jedi Knight II, UT2K3...basically any gaming benchmark or real application.

It all stems from the simple fact that the hit from 1:1 to 5:4 has always been around 2%, and the hit 2/2/2/6 to 3/4/4/8 is greater than 2%...it's more like 6%.

sorry zroc.
After many times trying to understand and after multiple experiments with settings . Im failing to see how less bandwidth = better performance .
Maybe its just me.
But i will take on any contest you want at 250 1:1 vs 200 5:4
Cause every test i do indicates the 1:1 is bettter.
on my system.
Thats both being at equal 250 fsb

I think his point , in testing only the memory, is that testing equal CPU and FSB by lowering the ratio to 5:4 and improving the timings to 2-2-2 will yield better results in most games.

However, in most systems (depending on your CPU's multiplier and top speed) you can get more performance out of the overall system (not just the memory) by raising the CPU speed and FSB.

Im my test above 230 at 2.5-3-3-7 beats 220 at 2-2-2-5, but its due mostly to the faster CPU and FSB, not ram. While I do agree with that, if my CPU couldnt do better than that I would stick with 230 2.5-3-3-7, even though its not the best memory performance, its the best overall performance.

Quote:

---

Originally posted by SpicyHuevos
sorry zroc.
After many times trying to understand and after multiple experiments with settings . Im failing to see how less bandwidth = better performance .
Maybe its just me.
But i will take on any contest you want at 250 1:1 vs 200 5:4
Cause every test i do indicates the 1:1 is bettter.
on my system.
Thats both being at equal 250 fsb

---

Less bandwidth doesn't = better performance. Less bandwidth with better timings CAN = better performance. Significantly.
Sure...here:

200.5 FSB/200.5 Mem (DDR401) 2.5/7/3/3
Sandra- 4681/4683
QIII- 363.3
Comanche- 49.35
3DMark- 16031

200.5 FSB/160.4 Mem (DDR320.8) 2/6/2/2
Sandra- 4392/4398
QIII- 363.9
Comanche- 49.43
3DMark- 16123


Anybody in here can repeat those results.
At the same clockspeed and FSB, 5:4 2/2/2/6 will match or beat 1:1 2.5/3/3/7, and slaughter 1:1 3/4/4/8. Not in Sandra or mem benches, naturally, but in real apps.

You use a stopwatch for real apps?
ALl i do with this rig is play games and bench man its the only apps i got;)
I have a normal computer or 2 for everyday use!

Quote:

---

Originally posted by SpicyHuevos
You use a stopwatch for real apps?
ALl i do with this rig is play games and bench man its the only apps i got;)
I have a normal computer or 2 for everyday use!

---

DJ, got some questions for you with the IC7.
http://www.xtremesystems.org/forums/...112#post188112

OPP

okay I just linked here from the [H] and thought I'd add my 2 cents.

In another thread there, someone asked also if they should buy pc3500 or pc4000.

I responded saying the pc4000 is a waste of money, and my pc3500 rocks with 2/2/2/5 timings at 272 fsb at 5:4 with turbo mode enabled on the IS7.

Then someone jumped in and told me they had a 275 fsb 1:1 overclock and that of course it was better... He didn't mention ram timings but I will assume they were not real tight.

so I asked him to lower his fsb to 272 and do some benchmarks.

Amazingly, our Sandra scores were nearly identical, both just over 5.9 gb/sec of bandwidth, yet my cpu score was about 100 MIPS higher than his.

He is using an IC7-G so there are some variables, but my point is that using the pc3500 at tighter timings and 5:4 can pretty much equal the benchmark results of the pc4000 running 1:1 at lesser timings, given equal cpu speed.

I have no real app benchmarks available at the moment, but I would bet money that my system would rock his to death considering I outscore him in 3dmark 2k1 by like 2000 points, using a 9500pro over his ti4200, and tighter ram timings at the same cpu speed.

seems to me, save the money on the ram which gains you nothing, and spend it on a video card, which generally means EVERYTHING when gaming.

I can testify to what Zroc explained in this thread, he was right.
now I run at 280 5:4 2-2-2-5 , it Does beat 280 at 1:1 3-4-4-8.

BTW thanks Zroc. :toast:

first of all, if you guys would run 2:2:2:7 instead of 2:2:2:5 which potentially kills the burst right in the middle of transfers, the differences between the low latency and the high latency settings would be quite a bit bigger.

Second, with the P4 on a "dual channel platform", the processor bus interface is the biggest bottleneck. That is, if you are overclocking, it actually pays to stay at a memory frequency that the memory can handle at acceptable latencies and running in asynchronous mode.

FYI, here are some interesting scores that need to be beat:
http://www.lostcircuits.com/memory/ddr400/10.shtml