Ddr2 @ 1-1-1-1 by kyosen

Kyosen has been testing the Asus P5NSLI (nForce4 SLI Intel Edition rev A3). An interesting board to say the least. Can run Corsair 5400UL or 6400C3 at CL2 or even CL1.

http://222.151.145.94/c-board/file/P5NSLI_NF4-SLI.jpg

VDimm mod
http://222.151.145.94/c-board/file/P5NSLI_Vdimm-mod.jpg

http://222.151.145.94/c-board/file/P..._CL2225-2T.jpg

http://222.151.145.94/c-board/file/P..._UL-CL1111.png
http://valid.x86-secret.com/show_oc.php?id=115776

http://222.151.145.94/c-board/file/P..._C3-CL1211.png
http://valid.x86-secret.com/show_oc.php?id=115782

That´s awesome :D
Any 3dmark with that timings ?

Even considering its kyosen, I have a lot of difficulty in believing that. It seems to me like a BIOS glitch of some sort.

Don't know about 3D. I think he indicated maximum FSB is not that good on the board.

Quote:

---

Originally Posted by Gautam

Even considering its kyosen, I have a lot of difficulty in believing that. It seems to me like a BIOS glitch of some sort.

---

Yes, that could be.:) Still, it's fun to see screenshots like that. He has sent in CPU-Z validations for the 5400UL's and 6400C3's.

kyosen really pushes ddr2 to the max...insane, really insane timings :slobber:

to be honest, im pretty apprenhensive to believe its really running 1-1-1-1

大著 keysen !! my hero hehehehe

but...... what is the voltage for that?

He was using 2.0v for the 5400UL's and 2.87v for the 6400C3's.

Here is CL2. There wasn't much difference with CL3, so it may indeed be a BIOS bug.:)

http://222.151.145.94/c-board/file/P..._C3-CL2338.png

It also says Hyperthreading is enabled @ x6800.

Could you try to test 1-1-1-1 vs 3-3-3-3?

looks like CL1.5 and CL2 bug on old NF4 chip

Quote:

---

Originally Posted by sierra_bound

He was using 2.0v for the 5400UL's and 2.87v for the 6400C3's.

Here is CL2. There wasn't much difference with CL3, so it may indeed be a BIOS bug.:)

http://222.151.145.94/c-board/file/P..._C3-CL2338.png

---

I did this a few weeks ago, with my Corsair 5400UL on a Gigabyte DS1.

I could never verify that it was running ras to cas 1, cause the differences were so hard to spot.

like there was only a 5 second improvement going from ras to cas 2 to 1, with super pi 32M.

I can do this over and over with this board... and I do think it's definatly running ras to cas 2, and when switching to ras to cas 1... it is doing something, cause the overclock definatly drops fast, when trying it...

however, since intel boards are limited to 2T, I hear, that while is may be ras to cas 1... it will always act like ras to cas 2....

http://www.xtremesystems.org/forums/...d.php?t=108116

http://www.xtremesystems.org/forums/...1&d=1153599765

http://www.xtremesystems.org/forums/...1&d=1153599976

unfortunatly, there appears to be no support for it on 975X boards, so I cant even try it on this DFI infinity... not even with memset.

The P5NSLI is a fairly low-end board. Retails for around $115-120. But it does support C2D.

Anyway, maybe kyosen will stop by and post his thoughts in this thread.

wow, cool, too bad it doesnt look like any performance gain, wonder if its really setting that way

Yeah, i'd like to see some comparison benches. Still looks cool though.

Wery nice... but i was thinking that performance diference will be bigger... Anyway grats :toast:

now, that's waht i call tight timings :D
anyway congratz :toast:

I'm a little skeptical on 1-1-1-1 w/ DDR2, agreeing with what Gautam said.

I think it's worth pointing out that kyosen's intention was not to deceive. The Japanese are very methodical in their testing. He's simply reporting what he found. Take it for what you will.

Many people don't believe the results because they've been told CL2 and CL1 are not possible, and not because their actual testing proved it's not possible. I think it's interesting the BIOS would even have those options.

It's very possible this is a BIOS bug. Anyway, I'll be interested in kyosen's thoughts.

i would like to see that superpi 8m at normal 3-3-3 or even 4-4-4, so that we could see if that 1-2-1-1, that looks so far to be the fastest "is" or it isn't.

Bios may have cas2 etc but DDR2 does not support it.

It took Oskar a year to actually get performance increases from the cas1.5 setting on the expert etc...this will just be a bios option that does nothing i fear.

Kyosen has a lot of screenshots at his c-board. I didn't feel like posting all of them. Here's a comparison of 6400C3 at 3-3-3 and 1-2-1.

http://222.151.145.94/c-board/file/P..._1T_C3-CL3.png

http://222.151.145.94/c-board/file/P..._C3-CL1211.png

Quote:

---

Originally Posted by Tony

Bios may have cas2 etc but DDR2 does not support it.

---

indeed, but 3-1-1 or 3-0-0 should still theoretically be possible :toast:

thanks for sharing this sierra, i look forward to kyosen's comments on the matter :up:

interesting...

the differences on his board are much more noticable, then on my 965 board at cas 3-1-1.

does the Nvidia board offer 1T at that setting? is he running 1T?

I'm not sure if he was running 1T or 2T. The NForce Intel SLI chipset does allow 1T.

Now it could be that setting RAS# to CAS# Delay to 2 instead of 3 is mainly responsible for the faster time. That's interesting in itself because I didn't know 6400C3's could do that. He also said setting it to 1 was a no-go. But X-1-X-X was possible with the 5400UL's.

Technology sometimes makes what was once considered impossible possible. For many years, people thought the multiplier lock on Intel CPU's was unbreakable. Then along came SpeedStep and suddenly everything changed.

Interesting. I dont know if it is really 1-1-1 because that would not be supported but something is obviously making the time faster.

if the comparison shots 333 to 111 are true, I'm beginning to get far less sceptical, thats insane, ...
NOW TRULY : IS IT EVEN POSSIBLE? :stick: ih yes .... the nForce Intel is :slobber: :slobber: :slobber: :slobber: :slobber:

Tony, sure, but still, it looks cool :D

Kyosen = HW god, sierra, go away, we want Kyosen in here hehe ;)

Quote:

---

Originally Posted by sierra_bound

Technology sometimes makes what was once considered impossible possible. For many years, people thought the multiplier lock on Intel CPU's was unbreakable. Then along came SpeedStep and suddenly everything changed.

---

thats not really a fair comparison.
speedstep (EIST) is something that had to be designed and added to the chip, not something that was always there but unusable... which is why say a old northwood, can't use EIST, and still have locked multipliers.

its obvious that Intel processors are locked from a Bios level...
that became clear when the 3.4E 478 prescotts were "accidentally" unlocked to 14, on some Asus boards, if you used a bios, that came out before prescotts...
but other then this, theres never been a way to unlock all intel processors.

even pentium D's today can't be unlocked (manipluated would be a better word), cause they dont have EIST.

getting ram to cas 1 and such... thats always been a option for ram... not a easy option, but its a option, not cause something was added to the ram to make it possible, like EIST...

Quote:

---

people thought the multiplier lock on Intel CPU's was unbreakable

---

it is actually lock by mobo bios. not by intel CPU...
mobo RD can also open the intel CPU MP if RD wants to do it..... but lose job 4ever!


like x6800 is 6-20 on p5wdh but it is 6-11 on some other mobo

@Kunaak
I was speaking in general terms. Some things are definitely impossible. But every time I hear someone say something is technologically impossible, I smile. Time has a way of proving people wrong.

Quote:

---

Originally Posted by M.Beier

Kyosen = HW god, sierra, go away, we want Kyosen in here hehe ;)

---

Okay I will go away.:D But you probably would have never seen these screenshots unless I had posted them.;) I did ask kyosen to join the discussion here. It's 7AM right now in Japan. He may be asleep.

How bout 0-0-0-0 ;)

I like it, he's doing a good job exploring something many would never try.

There still is the fact that times were faster @ 1-2-1-1 which im guessing is 3-2-3-3 or am i missing somethign here..

There are no intel boards that allow for 1T?

I never knew this, and thats why im AMD fanboi, until the roe rig arrives at doorstep..

I have the Gigabyte GA-965P-DQ6 board with F4d beta bios, and even with official F3 bios, my DDR2 timing options were able to get down to 3-1-1-1-1.
But CPUZ v1.36 always report 3-3-3-9..
I really need a program from windows for screenshots etc that can be as accurate as possible with true DDR2 timings...:(
Can't believe here we are more than 2 years after DDR2 on markets and no proper program can reliable report correctly....:mad:

awesome stuff Kyosen :toast:

Nive work - what is the best CPU FSB Achieved on this board so far?

Cheers

Mav

The jump doesn't look big enough in the comparo. Looks like 3-2-3- vs 3-3-3. The 1's being 3's in reality. Just a guess. tRCD 2 is definitely very difficult on GMH and GKX, but its doable at low speeds like that.

Gautam, Im not sure, but I belive its very easy to run 2 with the 5400UL, its D9DQ (aint that the one?) - anyhow the first and well known fatty chip from the old BBmods thread, I belive my ram has that chip (havnt checked the chips, but its on the list..) and they run 3-2-2-4 at 43xmhz with just 2.4v :D - but no way of them hitting the 600mhz, tops around 530 4-3-2-4

Quote:

---

its D9DQ (aint that the one?)

---

yep....D9DQT/W are fat body D9's they are the BH-5 for DDR2.......... :)

Quote:

---

Originally Posted by Dynasty

yep....D9DQT/W are fat body D9's they are the BH-5 for DDR2.......... :)

---

Great..
Now tell me, what would you prefer, 2*1GB of D9DQT/W or 2*1GB D9GMH??

Quote:

---

Originally Posted by M.Beier

Great..
Now tell me, what would you prefer, 2*1GB of D9DQT/W or 2*1GB D9GMH??

---

I would take fat body Micron's anyday over the newer GMH/GKX stuff....:)

But, thats just my preference....

Quote:

---

Originally Posted by Dynasty

I would take fat body Micron's anyday over the newer GMH/GKX stuff....:)

But, thats just my preference....

---

Agreed.

I got a killer set of 5400UL here that does 470 4-2-2-x, but being just 2x512, gets outscored by my much cheaper D9GMH. :( Of course I'd love to get my hands on a 2 gig set of DQW or DQT, but I guess you can't always have it all. :-/

So yes, 2's are pretty easy on DQW/DQT, but GMH and GKX are completely different animals. (Considering how little I paid for mine, I suppose I shouldn't be complaining)

But..
If I can hit the freq I can, is it then pretty certain I got some of those chips on my mem, cause frankly, I must admit, I like this ram and dont wanna waste 'em, especially not since they're from USA and I live in Europe.. (I fear pulling of heatspreader will pull of chips as well, havnt done it ever myself, but seen it, and its a damn shame)

Thing is - other threads talk about BE6 or something' but I've heard that its only the old kits that has D9DQ..
But I kinda' know its not GMH or Elpida since its never been used on these kind of mem :)

Quote:

---

I got a killer set of 5400UL here that does 470 4-2-2-x,

---

what rev. are they ?........ :)

1.2, no substitute. :D

M.Beier, DO NOT pull the spreaders. You've obviously got D9DQT or D9DQW. Guard those sticks with your life. :p:

Great :D

Was a pair of 322$ sticks as far as I remember :D

Pretty neat buy, aye? :D

Quote:

---

Originally Posted by M.Beier

Great :D

Was a pair of 322$ sticks as far as I remember :D

Pretty neat buy, aye? :D

---

They are definitely keepers.......:)

If you look at the screen shot of the monitor. I have done CAS2 on C19A and XE as well.

A latency screen shot would be nice, should be very close to best A64 ODMC.

I got Corsair 8000UL's rev1.1 & 5400UL's rev1.2 - but both sets only 512mb sticks. Which ones can handle the highest bandwidth and tightest timings or are they pretty much the same...:)

I'm asking because my Gigabyte GA-965P-DQ6 mobo has option of running DDR2 @ 3-1-1-1-1. Cas 3 is as low as it can go. ;)

A memory engineer told me that CAS 1 is physically impossible. He explained it but this was a while ago

it looks nice, but seriously, 1 - 1 - 1 - 1 :)
Do you really believe this ?

Quote:

---

Originally Posted by ZX7891

A memory engineer told me that CAS 1 is physically impossible. He explained it but this was a while ago

---

Yeah, it makes you wonder. Why would mobo vendors not put it as an option, even at Cas 2 on enthusiasts boards despite there being high end DDR2 sticks on market for a while now. :)

oh sweet jebus

Quote:

---

Originally Posted by sierra_bound

http://222.151.145.94/c-board/file/P..._C3-CL1211.png

---

If it's really CAS 1, it's a slow CAS 1. (I couldn't get any 2's to complete 8M)

:eek: Only beatable by 0-0-0-0........ :p:

:toast:

I thought that nothing below cas 3 could be chosen for ddr 2 ram? I'm very skeptical about this. We need some more tests and some input from

... But who has the guts to challenge Kyosen in a tweakin' contest? - I know I dont :D

I'm more interested in the fact that the post claims he has a core 2 duo with hyper-threading enabled...

Quote:

---

Originally Posted by sierra_bound

http://222.151.145.94/c-board/file/P..._CL2225-2T.jpg

---

Good point, not sure how I missed that when I was checking out timing from that image.

Quote:

---

Originally Posted by DilTech

I'm more interested in the fact that the post claims he has a core 2 duo with hyper-threading enabled...

---

not even possible is it, i mean core 2 doesnt even have HT does it? (archetecture wise)

im scared about ht coming back

i asked around and some one told it wouldnt help with such a short pipeline in c2d because it would ad latency.

i think that i remember hearing from somewhere that ht will return eventually, but i think that in this case its just a nforce 4 bug.

I wanna see latency tests. Cool, HT... again... :D

I don't know if it has been said before, but CAS below 3 on DDR2 is technically impossible.

I can understand that you can't believe my P5NSLI results...
...at first I also couldn't believe my eyes:eek:
But these are simple facts that P5NSLI has CL1 setting and
that I can set CL1-1T with Corsair 5400UL, without any difficulties.
Several screenshots are here:
http://www.oohashi.jp/c-board/c-boar...ne;no=4798;id=

Anyway, as Gautam says, this P5NSLI is VERY SLOW even if running at CL1.
There is no reason to choose P5NSLI...except formal SLI support, I think;)

Fair enough, and nice work in trying out some very unusual parameters...but, if your eyes can believe that, can they also believe this? :p:

http://mysite.verizon.net/gautamb/memory-3101.JPG

Corsair 5400UL 1.2 at stock vdimm. :p:

CPU-Z and BIOS cannot always be trusted...

Hahaha, interesting, Gautam:)

> CPU-Z and BIOS cannot always be trusted...
--
But if BIOS setting and CPU-Z(or any software) showing are same,
and if benchmark scores vary according to BIOS setting,
I have no way to certify ｔhat they are false.

I don't know whether P5NSLI's CL1 is "real" CL1 or not,
but at least it's not same as CL3;)

haha cool! good job :)