2-1-1-5.

[STEvil]

Had a thread on this a while back actually, but nobody was throwing 3.6v+ at their ram back then...

Anyways, anyone tried going past 200mhz at 2-1-1-5, or 2-2-1-5? ;)

[Orak]

I managed 2-2-1-2 with my BH5 at 200mhz @ 2.98v :) not far form it. that was with my nf2 rig.

[enzoR]

wooooaaaa talk about tight timings

[CCW]

Originally posted by Orak
I managed 2-2-1-2 with my BH5 at 200mhz @ 2.98v :) not far form it. that was with my nf2 rig.

Im not doubting you but it would be great to see a CPU-Z screenie of that :D

[jmke]

would be more interesting to see a Q3A benchmark compared to CL2 2-2-5 of that! :)

[STEvil]

I've managed 185 2-2-1-5 on my BH-6 @ 3.0v so far... havent really tried going farther since my mobo is having a hissy fit..

[Pro Kid]

From what I've heard its really not all that great to run the memory that tight because it just begins to lose stability. But with the voltages you guys are throwing at em...that changes everything. I tried to run my GeIL at 2-1-1-5 at 166 and as soon as I hit ok with the nVidia SystemUtility my pc locked.

[DOA]

At 200 FSB I get more throughput from Sandra with 2-2-2-11 than 2-2-2-6 which is as tight as the 2x 512 Corsair will run.

I think access errors come into play. The system is stable at either setting, but the slower timings are "faster".

[r4z]

how much do you earn with 2-1-2 or similar ? super pi or sandra

[Demon_Hunter]

Hey guys talking about the tRAS, it is true that put it on 6 or 11 is better than the other numbers ???
And to put timings like this you need put some more Vdimm, like 3.0Vdimm but only with the headskins it's enough to put all this voltage ???

[STEvil]

headskins?

Tras is board/bios dependant.. give them all a try and see what works best for you.

[Demon_Hunter]

sorry I don't know the name, but I'm asking about that aluminium or copper (passive dissipator) that you can put on the memory, only with that you can put live 3.0Vdimm or you need put a fan to cool the memory ??

thanks

[STEvil]

ah, ok.

They're called heatspreaders and the general opinion is they are pretty useless seeing that they use fragtape (thats what it used to be called, now its thermal tape..) and dont even touch all the memory chips.

3.0v without cooling should be fine, as long as you have a bit of normal case airflow.

3.2v+ is where its a good idea to make sure you have decent airflow around the ram.

[Demon_Hunter]

this true, these heatspreads are bad ??
but you can't trade the thermal tape ??
which one are better, the ones that come with the memory or buying new ones ???

[STEvil]

neither.

Running them bare is the best.

The tape has bad thermal transfer efficiency, and even if you removed it and attached the heatspreaders with glue they would still not touch the entire surface of each memory chip.

[Demon_Hunter]

I know that generaly the thermal tape are bad but has some one that are good, hasn't ??

even being a cooper heatspread ??

hey you can tell me a good chip for overclock that is easier and cheaper than a bh-5 ??? (with low timings)

they are very difficult to find, aren't they ???

thanks

[STEvil]

Thermal tape sucks, 'nuff said.

Copper heatspreaders cant do anything if the tape is keeping the heat from reaching them.

There are none.

[gouda96]

if you pulled the heatspreader off you have already voided the warranty...so why not just use some artic silver epoxy and attach some custom heatsinks you cut from an old slot 1 card or something like that. Same thing that you would do for video ram.

[Demon_Hunter]

good idea man, thanks

but now and about the timings ??
do you know what is a good chip for low timings (bh-5 is too hard to find and too expensive) ???

thanks

[gouda96]

bh-5 and bh-6 are probably the best. If you want to take a chance, you could try to grab some newegg refurb bh-5 or bh-6. I have had bad luck with newegg refurbs, but you could get lucky.

[Demon_Hunter]

wha do you mean with bad luck man ??? it's like I'm not used to buy stuff on the internet here in the US, I'm brazilian and I'm in the US for 6 months so I don't know to much about the sites to buy stuff here, this newegg is good ??? which more is good and I can trust ???

[STEvil]

Check the big BH-5/6 thread in the overclocking and cooling forum.

Kingston Hyper-X KHX3000 from newegg is BH-5.

[Monarch]

What would be faster 2-2-1 or 2-1-2?

[STEvil]

2-1-2 probably

[Demon_Hunter]

Originally posted by Demon_Hunter
wha do you mean with bad luck man ??? it's like I'm not used to buy stuff on the internet here in the US, I'm brazilian and I'm in the US for 6 months so I don't know to much about the sites to buy stuff here, this newegg is good ??? which more is good and I can trust ???

hey man how can I know when a memory is BH-5 or not ???

this memory have BH-5 but it get an overclock, like how much do you thing the FSB goes ??? and this memory works good and dual channel ???

thanks

[Orak]

Originally posted by Demon_Hunter
hey man how can I know when a memory is BH-5 or not ???

this memory have BH-5 but it get an overclock, like how much do you thing the FSB goes ??? and this memory works good and dual channel ???

thanks

If you have TwinMOS with winbond chips you can read it on the chips..
But if you have other moduls like Kingston etc, they change the txt on the chips so you cant read it.. But you can tell that its BH5/BH6 if the chips do 2-2-2-5 easy at low voltage. like my modul does 230 2-2-2-5 at 2.9v.. You can look at the chips to, if there is two circles on the chips its usaly BH5..

My english is baad so dont whine :)

[iddqd]

Look at my sig :D.
Yes, I do intend to run 1.5-1-1-0 at 250MHz, and over... or at least die trying. Now, all I needs is a new PSU, and I can begin volt-modding... I say about 4.2vdimm, or so should do the job.

[Mr. Tinker]

Even better. On my IS7-E with my modded bios (http://www.adamsanders.net/files/IS7A_22.BIN) you can try 2-0-0-5. 2-0-0-5 in 2005 sounds catchy.

[iddqd]

I could do 0's on the DFI as well, but That's just too low... it barely takes the timings the way it is now.

[ZMarre]

Friend of mine:

2-2-0-7 | 247MHz (http://users.pandora.be/taye/pics/super_pi_33s.JPG)

screeny SSandra (http://users.pandora.be/taye/pics/247x10%20Sandra%20MEM.JPG)

[reject]

isnt it so that is the 2 main timings are lower than 2 exra latency is introduced from the memory controller, the ram starts idling and this causes lag to the memory cotroller?
btw. that spi is awesome!

[craig588]

Friend of mine:

2-2-0-7 | 247MHz (http://users.pandora.be/taye/pics/super_pi_33s.JPG)

screeny SSandra (http://users.pandora.be/taye/pics/247x10%20Sandra%20MEM.JPG)

You have it wrong, you acctually have it at 1.5-2-2.

[Geforce4ti4200]

iddqd at 1.5-1-1-0 is the tightest timings ive seen by a long shot! that is just insane, I have never seen any ram do ras-cas1 and one of my friend said theres no ram in the world thatll do cas 1.5 at over 166fsb, I did manage cas 1.5 back on generic pc3200 at 133fsb but dont think it was 100% stable and I think it gained like 200 marks but 166fsb at 2-2-2-6 was still better anyway. can you do us a favor? let us know how much you gain in 3dmark 2001 with ras-cas1 vs. 2 because this may be real exciting too! going from ras-cas2 to ras-cas3(the only timing that matters on a64 other than command rate which I always leave at T1) costs me about 25-30fsb which is about 75MHz cpu clocks which was exactly 158 marks back on my ti4200. I never did much testing with ras-cas3 vs. 4 but from what ive seen, the hit is pretty severe, much worse than going from 2 to 3. with this pattern, id assume thered be maybe 100 marks if that going from 2 to 1 since 2 is already darn low and the ondie controller just doesnt need or cant take advantage of 1. everyone says ras-cas4 sucks yet I see lots of people run it like that, well 250+ fsb with ras-cas4 is needed just to match 200fsb with ras-cas2, sad eh?

[Preyos]

You have it wrong, you acctually have it at 1.5-2-2.


no craig he was right what he said was definitly possible. once i got it down to the 2 2 1 timings with the negitive 50 degrees celcius extreme overclocking nutstraight. U really dont know much about computers

[STEvil]

Craig is correct.

Cas latency 1.5
Ras to Cas Delay (TRCD) 2
Ras Precharge (TRP) 2
cycle time (TRAS) 0

Therefore 1.5-2-2-0


once i got it down to the 2 2 1 timings with the negitive 50 degrees celcius extreme overclocking nutstraight

what?

[mushk1n]

I thought it was physically impossible for the data to transfer in 0ns.... :/ wow that is pretty quick :P

[STEvil]

tras is only 0, which means there will be no cycle delay between clocks not that the latency of the memory is 0 (this is technically impossible pretty much).

[iddqd]

tras is only 0, which means there will be no cycle delay between clocks not that the latency of the memory is 0 (this is technically impossible pretty much).
It's possible. There is no delay at all.. think of sram.

[celemine1Gig]

It's possible. There is no delay at all.. think of sram.

Well, it's possible that's right. BUT, it would only be possible at very low temps (much below zero), when the metal and semiconductor parts of the chip would become superconductors. Then there would be no real resistance and and it could theoretically be possible.

That sounds good in theory, but reality is another question. :toast:

[iddqd]

No; when Si becomes a superconductor (and it does so at a higher temp than aluminum that you want to superconduct), it would no longer be a semiconductor; hence none of the logic gates would function properly.

On the other hand, there are types of memory where there is no delay, at room temperatures, or above. Sram = static ram = cache, and most L1 have 0 latency. You could also run 0 latency on old sdram, I beleive.

I see the error of your logic - you're thinking about the delay in terms of actual time (0ns = instantaneous). It's impossible, even if you have superconducting circuits, theoretically speaking. The electrons still can't move from point A to pointB in no time... that would be teleportation, their mass would have to be infinite. Either way, it's not. That just means that there are no CPU cycles in between the time when the memory controller asks for a read (or write) and the said read/write initiates. Yep, it happens on the SAME cycle, which takes up some amount of time - however small. On a 2Ghz chip, that time would be 1/2,000,000,000 of a second, potentially enough time...

[STEvil]

L1 and static ram still have latency...

[celemine1Gig]

Well, very interesting! You learn something new every day. :)

But, as I already said, theory and parctice are very different sometimes. :) So, theoretical there could be RAM with no latency, but as you also pointed out, it's physically impossible.

[buff]

anybody tried lower timings on VX yet?
OPP?

[STEvil]

Lowest I was able to hit on VX was 1.5-2-2-0

lowering cas, trcd or trp all resulted in no post at 200mhz 3.30v

[iddqd]

L1 and static ram still have latency...
Yes, but there's no write delay for example.

[STEvil]

There is no cycle time between writes, yes....

[iddqd]

It seems the only solution for this is for me to actually try running one of the timings at 0 on my BH-5. However, I warn you - because I'm going to need enough voltage to destroy the world :cool:

[beau_zo_brehm]

I thought that performance gains stopped when setting timings below 2 and decreased because of corrupt data and memory dumping. :confused:

[Tek]

My lowest was CL1.5 0 0 0 7 0 T1, but only at 133mhz. They will bench and memory test perfectly, but a few more mhz and they done for it. Will try at 200mhz and see how low they will go at that speed :)

[bias_hjorth]

Would you do me a favor running the above timing vs. 166 2-2-2-5 in a 3dmark01

[Tek]

Yes, ofcourse :)

I have tested CL1.5 0 0 0 7 0 T1 vs CL2 2 2 7 0 T1 at 133mhz. The thing is, for some reason DL or DH will crash the system, if I use the same settings as the CL2 one, at 166mhz.

Got the two results for now, as I think the one or two of sub timings were too tight. Will try to make a CL1.5 0 0 0 7 0 T1 133mhz vs CL2 2 2 7 0 T1 166mhz, as soon as I figure out the lowest common settings.

Hope these are alright for now. The first is with CL1.5 0 0 etc. and the second the CL2 one. ~23k vs ~22.6k, there is some difference.

OT: I get my FX-55 tomorrow(it is already at the postoffice, girlfriend ordered it without me knowing about it :D ), but haven't got a s939 mobo, yet :p:

[iddqd]

So, wait- you ran 1.5-0-0-0? 1T? Respect.

[bias_hjorth]

Thx :)

Man that ras-cas = 0 pulls out some performance. I wonder if the 133 0-x speed is faster than 2-2-2 at 200mhz.
Whats your highest mhz with rascas 1?


OT: lucky bas**** :p: :D

[Tek]

So, wait- you ran 1.5-0-0-0? 1T? Respect.

Thanks. Only at 133mhz T1, though :)

CL1.5 7 9 0 0 0 0 2 1 1 133mhz 15.6us x8 64bit disable disable x4 7ns 7.5ns 0 clks enable 133 - A64 Tweaker settings. Two 512mb sticks SC.

These Geil RAM are good. 240mhz CL2.2.2.5 T1 200mhz 15.6us with two 512mb sticks is pretty good. 245mhz only with one stick, though. They should be Geil PC3200 Ultra CL2.3.3, but they were relabeled by Kingston and sold as Kingston Ultra RAM, but kept the Geil logo on the Cu HS. Not sure what they are, they could be CH-5 or even BH-5/6(it has the two big, and one small dimples), but I don't know :)



Thx :)

Man that ras-cas = 0 pulls out some performance. I wonder if the 133 0-x speed is faster than 2-2-2 at 200mhz.
Whats your highest mhz with rascas 1?

Yeah, it does seem to make some difference. The 133mhz CL1.5 thing isn't faster than CL2.2.2.5 at 200mhz, I already checked that. I think, some other timings has to be tweaked to unleash the full potential, but I don't know much of what I am doing RAM-tweaking-wise, and I don't know if these RAM would allow for that potential to been seen. Any suggestions to what could be differently with the A64 Tweaker are welcome :)

Will check how high I can go with rascas 1, first with loose sub timings, then tight(should make a world of difference).

OT: Yeah, it so sweet :D Too bad I can't bench the beast right away :( Will post pictures at once. I hope it is loaded with leprechauns. Vapo LS will be up and running tomorrow, and there is a slight chance I can get my hands on a mobo :)

[iddqd]

Thanks. Only at 133mhz T1, though :)

CL1.5 7 9 0 0 0 0 2 1 1 133mhz 15.6us x8 64bit disable disable x4 7ns 7.5ns 0 clks enable 133 - A64 Tweaker settings. Two 512mb sticks SC.

These Geil RAM are good. 240mhz CL2.2.2.5 T1 200mhz 15.6us with two 512mb sticks is pretty good. 245mhz only with one stick, though. They should be Geil PC3200 Ultra CL2.3.3, but they were relabeled by Kingston and sold as Kingston Ultra RAM, but kept the Geil logo on the Cu HS. Not sure what they are, they could be CH-5 or even BH-5/6(it has the two big, and one small dimples), but I don't know :)




Yeah, it does seem to make some difference. The 133mhz CL1.5 thing isn't faster than CL2.2.2.5 at 200mhz, I already checked that. I think, some other timings has to be tweaked to unleash the full potential, but I don't know much of what I am doing RAM-tweaking-wise, and I don't know if these RAM would allow for that potential to been seen. Any suggestions to what could be differently with the A64 Tweaker are welcome :)

Will check how high I can go with rascas 1, first with loose sub timings, then tight(should make a world of difference).

OT: Yeah, it so sweet :D Too bad I can't bench the beast right away :( Will post pictures at once. I hope it is loaded with leprechauns. Vapo LS will be up and running tomorrow, and there is a slight chance I can get my hands on a mobo :)
Geil never made anything with Winbond in it... I've looked at the "winbond list" a lot of times, before they took it down, and never saw any Geil there... Hmm, I don't know if you bought it recently, and it's the UltraX stuff, then it's TCCD. Another clue is the dimples, Winbond used the same packaging plant as everyone else, they make two large dimples there, that's it... so it can't be Winbond. Alright, my next project is going to be "1.5-0-0-0 at 250Mhz"... I'd need to get about 6vdimm, and a phase changer on my ram.

[STEvil]

Old samsung TCB3 chips or Winbond likely.. very likely since they are just repackaged Kingston actually.

[bias_hjorth]

Geil never made anything with Winbond in it...


They surely did. Both bh5 and ch5 - Actually the pc3500 Geil Platinum was a 100% certain bh5 at a period.

[iddqd]

They surely did. Both bh5 and ch5 - Actually the pc3500 Geil Platinum was a 100% certain bh5 at a period.
Interesting. Well, too bad for Geil, their PCB is mediocre at best.

[Tek]

Interesting. Well, too bad for Geil, their PCB is mediocre at best.

My thoughts exactly. These chips with Brainpower PCB would be :banana:



Alright, my next project is going to be "1.5-0-0-0 at 250Mhz"... I'd need to get about 6vdimm, and a phase changer on my ram.
:explode:

:D

[STEvil]

Really? the old TCB3 samsung ram was the only other stuff that could do 200mhz 2-2-2 under 3.0v back when BH-6 was new.. wonder if this stuff can still do it? ;)

[celemine1Gig]

Sure, but it's quite hard to find some of the good old samsung C-TCB3 stuff.

[STEvil]

should test the 1gb sticks, see how they do.. :D

[STEvil]

lol

[STEvil]

ok, some results from tonight.. :D

XP-M 2500+ @ 11.5x227.4 2.03v (2613mhz) on Abit nF7, 2x512 CH-5 @ 3.30v..

CAS 2
TRCD-r 1
TRCD-w 2
TRP 1
TRAS 11
1-3-3-3-2-1-3 alpha timings

Sisoft unbuffered: 1701/1776
Sisoft buffered: 3494/3304
Super-Pi 1M: 40s

Old install of winxp pro and using drivers for nF3...

Thats the best I can manage, but I might be able to get a few more mhz FSB.. Setting TRCD-w to 1 causes the system to crash even with 3.75v to the mem, on my 2x256 BH-6 and 2x512 CH-5.

BH-6 can run memtest to 255mhz running 2-2-2-1-11 IIRC, i'll have to try them out again.. 10.5x252 would spit out some nice bandwidth at 2-1-2-1-11-13-15 1-3-3-3-2-1-3 :D

[funktional]

235*11,5@2,03 AN7
2x512MB Kingston Value BH6/BH5 3.3V

7
10
CAS 2
TRCD-r 1
TRCD-w 2
TRP 0
TRAS 11
0-0-3-3-3-0-3 alpha timings
CPC-off

This was gamestable. Cant give the RAM more Voltage as my 3.3V line only goes up to 3.35V ;)
Will try it on my NF7 soon.

[STEvil]

I didnt find a performance difference between 0 and 1 for TRP on my nF7.

[Rabbi_NZ]

Jus posting to get updates... Ill test my VX when I get a chance

[M.Beier]

With my A8N-SLI (last MB, sold it)...

I managed running my ballistix at 2,5-1-1-1 255 mhz.. 24x7 stable...
- The low cpu voldtage was my greatest issue... The mem ran 263 mhz - I think... At 2,5-1-1-1....

[blinky]

check this:

http://home.earthlink.net/~dzungq/images/2-1-1-11.jpg

[Sons_of_Piru]

check this:

http://home.earthlink.net/~dzungq/images/2-1-1-11.jpg

:slobber: :slobber: Which ram it is?

[blinky]

:slobber: :slobber: Which ram it is?XMS3500

its for sale if u want to buy it

[Rabbi_NZ]

XMS3500

its for sale if u want to buy it
what size module is it?
PM me if it's 256mb please ;)

[celemine1Gig]

Would you please both post some more info, 1_2_know and blinky?

That would be great! :toast:

What did it take to get from 2(2.5)-2-2-11 to 2(2.5)-1-1-11 ?

Let us know what you did to make it that far with those ultra tight timings. And how much did it help in terms of speed (synthetic and felt speed). :)

[blinky]

Would you please both post some more info, 1_2_know and blinky?

That would be great! :toast:

What did it take to get from 2(2.5)-2-2-11 to 2(2.5)-1-1-11 ?

Let us know what you did to make it that far with those ultra tight timings. And how much did it help in terms of speed (synthetic and felt speed). :)thats not mine, its a guy i knows

[celemine1Gig]

Well, then simply tell him to please post the info here. ;)

[M.Beier]

1,5-1-1-1 at 176mhz I think ;)

3,2v on ballistix PC4k

[celemine1Gig]

1,5-1-1-1 at 176mhz I think ;)

3,2v on ballistix PC4k

I thought you had them running 255MHz 2.5-1-1-11, according to your previous posting. :rolleyes:
As Cas Latency is not very important on A64 CPUs, why don't you tell us the details for 255MHz or more, instead of the 176MHz results? :toast: That would be far more interesting.
I bet that 255MHz with 2.5-1-1-11 is faster than 176Mhz with 1.5-1-1-1. :)

[tinker77]

it works fine with me.

[M.Beier]

Well...

They suck on my DFI nF4..

But A8N-SLI (wont run T1 - ASUS bug)... 277mhz 2,5-3-3-5 ;)

[craig588]

I have BH5, I have a 4V 3.3V rail, I have a clawhammer with a crappy memory controller, I have a new project: Take my ram to to 2-0-0 at the highest MHz it will take and find out if it's faster than 2-2-2 at the limit of the memory controller.

EDIT: I just woke up and did some benching 2-0-0 at 130MHz (Something that is close to the limit of the memory, it might have a few MHz more, but nothing signifigant) was a serious hit in performance compared to 252MHz at 2-2-2.

I don't think anyone needs to bother with timings any lower than 2-2-2 uaing any current memory sticks.