:thumbsup:
Printable View
:thumbsup:
I can confirm the new method works :)
I took out the old Vdd=Vbt mod and put in the newer mod .
It works just as well , but it appears my board just won't budge over 240fsb no matter what .
Cheers
Im just concerend the 4th way once again pits 2 MOSFETs against one another.
We need to trace the Vtt signal, does it go to the northbridge? If so how big is the etch? If they used too small an amount of copper, then at high FSB/heavy loads the terminations resistors at the NB end of the transmission line might not get enough current, and the voltage at one end would DIP and lead to errors when you really push it.
I think ABIT needs a design review by a serious engineer. You'd be suprised what differences a little more etch here, and extra cap there can make.
Though at production rates of 1.4 MILLION boards a MONTH every penny does add up fast :D
Sounds like you are just the guy to help them see the light!! :hehe:Quote:
Originally posted by uwackme
Im just concerend the 4th way once again pits 2 MOSFETs against one another.
We need to trace the Vtt signal, does it go to the northbridge? If so how big is the etch? If they used too small an amount of copper, then at high FSB/heavy loads the terminations resistors at the NB end of the transmission line might not get enough current, and the voltage at one end would DIP and lead to errors when you really push it.
I think ABIT needs a design review by a serious engineer. You'd be suprised what differences a little more etch here, and extra cap there can make.
Though at production rates of 1.4 MILLION boards a MONTH every penny does add up fast :D
I have a concern where we are adding the wire between the Vref and Vtt pin as in this pic:
http://downloads.modlabs.net/bios/pic/4th-Method.JPG
I mean, the Vtt is already getting voltage supply from a point, and now we are introducing another source via Vref. What effects does this have and its not like the stock Vtt signal is just going bye bye here just because the Vref is introduced.
Dont get me wrong, its a brilliant idea when compared to having to cut a trace...but I wonder what long time use will bring. I can see that you have updated the site where you list people who have ahd success with this....has anyone had any serious issues with this mod?
Also, with the Vdd Booster mod, which is basically shoving the 5v rail into that cap, what are the benefits of this? Is it stopping spikes in the Vdd supply?
Thanks.
What is the voltage of the cap before the booster? If its within .1v of the 5v rail already then its probably not going to hurt it at all.
I`ve messured the Voltage, its not the cap its the induktive, and 5.08V on the ATX was also on this point.
But it could be that i`ve messured wrong pin, but i dont think so.
emmm... u still wana cut the trace ha?
ok give me high reso pic on this spot
i will show whicj trace to cut after the vref-vtt mod
cheers :toast:
if still want to cut the trace
and for sure this move will void motherboard warranty
cut em up..... :stick:
I take it the Vdd boost mod bypasses the ferrite bead filtering the 5V supply to the Vdd regulator circuit. The bead is to keep noise out, but if the value is too high, at HIGH current demands....like mod'd Vdd setups... it wont allow enough current into the circuit fast enough to meet demand, leading to a voltage DIP at the 5V input of the Vdd regulator circuit.
What etch are we cutting now TicTac? That little 3 prone gizmo is a transistor...a discrete mosfet.... was that supplying Vref on that board? I'll have to pull the back up out again and re-evaluate the Vtt Vref connection.
it is pin 1 of winbond ic
the vbt strenght set by winbond ic
so the vbt mosfet wont have their own volt source anymore
there will be only1 volt source it is Vref no more voltage from vbt mosfet itself... :(
it looks strange to me, but who cares....... :rolleyes:
tictac, you dont recomend new user to try your 4th method.
why? i wont do the old mods, only the new. it looks easy and "safe"
coz i have kill 3 board on the 3rd method... :(
i dont want to kill more board... :(
use it at your own risk if u use it... :(
I checked both my boards, the Vtt and Vref are equal, and track when I turn the pot. Weird. Unless Im not even LOOKING at Vtt at all, only Vref....but then, this IS what the pot/etchcut are diddling.
Excellent Mod Gentleman! :)
btw, i dont understold it well... was the 4th method (vtt=vref) tested?
EMPShock
I'v test it (the 4th method), and it's work fine, it's the 3rd day from the first time i mod Vref=Vtt, it's fine... but the 3th method VTT=VDD, I'v test it too and is fail, and it's make your VTT mofset on fire... heheheh (I mean Fry...)
Just got 2 NF7's. :) Finally had the chance to try it myself.:D
Ok, tell us more about those babys :cool:Quote:
Originally posted by Malves
Just got 2 NF7's. :) Finally had the chance to try it myself.:D
I can see that u have grounded the 3 leg of VR, why?
What the heck are there 3 connections to the pot for?
Malves pre mod, can you confirm that Vtt and Vref are tied together or not on the boards? And do either track then, after the Vdimm mod?
Just did Vdimm and Vtt=Vref , looking good so far. Gonne try and meassure now. That little thingie got very hot, i've put a bga sink on it.
My hands are to shaky, i just messured3.3 rail and the pin next to it schould be vtt or vbt right? It gives me 3.66 for 3.3 and 1.7 for the other. So i gues their tracking :banana:
edit: what did i just said :rolleyes: Vdim is 3.4 and 3.3 rail says 3.66
At least it tracks the Vdimm correct
Have you ever seen VR with 3 connections:confused:Quote:
Originally posted by uwackme
What the heck are there 3 connections to the pot for?
my vr got 3 leg :stick:
All of my millions pots have 3 legs:D
I have done now to my new NF7: lapped NB, golden Zalman on SB, high sinks for mosfets/IR, VDD/"normal" Vdimm mod and cut-trace VBT mod, sinked every *ucking mosfet on the board:p:...and now, were can I found Vcore mod for this board:toast:
I hope this board can do some SuperPi 32s.:stick:
this board bios sux a big time....
ROMSIP mod is not enough... :(
if i can locate the memory timing table we may have bigger chance to hit higher fsb and higher memory memory bandwith :(
Nah, great BIOS, great mobo but Vcore fluctuate like mad:mad:, we need some Vcore mod and maybe Vcore sense mod:confused:
With REAL 2.3V-2.4Vcore we can reach the sky:toast:
sorry i must admit that.... that is fact... the bios sux a big time.... :(
the bios is still limiting the fsb performance :(
i manage to locate 1 more table... but i cannot assure it is memory timing table :)
but i check it from D10 to D23
here is the result:
D10
D11
D12
D14beta
D16
D17=D18
D19=D20
D21=D22
D23
So there is 9 difference table.... should i mod this? and the question is how do i mod this? the hex code killing my brain... :(
Yes thats the truth, some more FSB we can get some nice scores even when our CPU MHz is'nt so BIG:)Quote:
Originally posted by tictac
if i can locate the memory timing table we may have bigger chance to hit higher fsb and higher memory memory bandwith :(
THat's lot :slobber: , but if you found that little secret that hold FSB down I come down there and we take some beers:toast: :DQuote:
Originally posted by tictac
sorry i must admit that.... that is fact... the bios sux a big time.... :(
the bios is still limiting the fsb performance :(
i manage to locate 1 more table... but i cannot assure it is memory timing table :)
but i check it from D10 to D23
here is the result:
D10
D11
D12
D14beta
D16
D17=D18
D19=D20
D21=D22
D23
So there is 9 difference table.... should i mod this? and the question is how do i mod this? the hex code killing my brain... :(
Jeah, I know, it needs time to search and guys to test:(
keep back on topic:
Good Luck Malves ...... :thumbsup:
p/s: Dani... ya it takes time to solve the hex code... :(
:toast:, how far can I go with Vcore? 2.4V?Quote:
Originally posted by Malves
Vcore for the NF7 is the usual HIP6301. Pin 7 to pin 9 (GND) via 50k trimmer.:)
If 2.4V so in reality that will be someting around 2.28-2.3V under pressure?:cool:
NOOOOOO, I got a BIG sink on that chip:eek:
very big heatsink dani.... hehehe
Nice pic, I remember those days, some day more CO2:D
But I cant get it off and I cant solder under that sink:mad:
Hmm, however, I will twist it off..
Can someone please explain to me why the 3.3 rail leg gives more than i messeure on the 3.3 atx connector. Vdimm is between 3.34 and 3.37 it seems . Vbt/vtt is 1.7 so thats fine with me. On the atx connection 3.3 gives 3.34 like it should equal to Vdim. But where comes the extra volts from ? The rail leg gives 3.66 volts . Should i be worried about this or is it normal?
Oh and my lvl2 ram that primes happy on another nf7-s 238 5-2-2-2 with "old school Vmem mod" . Now it gives me nothing but errors in memtest86. KHX3500 c2 runs wel on it and not on the old one , go figure :D
And please forgive my bad englisch as well :)
It's not normal cos thats impossible, you cant have more Vdimm that PSU's +3.3V can give out...
The pin on the bottom right side gives my dmm 3.66 volts. If i had company at this time i would take a pic of its display. So its not oke then. Most have done something wrong :( . Gonne undo it for now. Will try the Vtt=Vref later on my other board.
Malves... on the 3 pin thing... DONT. What you do by grounding the 3rd pin is connect a resistor from the Vdimm mod point, to gnd....that resistor being an R= full value of your POT. While also having a sliding scale from the 2nd pin. It sets up a voltage divider ON TOP OF a voltage divider. You really dont wanna be doing that, trust me. Just disconnect the 3rd-pin to ground lines you put on, the rest will work just fine....you will prolly have to re-adjust your pots. Best to set them BACK to start point, remove ground lead, and re-adjust your V's.
Tic'y!!! Dont despair old buddy, you'll get to the bottom of it.. you're bigger than any bios. The D10 and D23 tables are the only ones that matter.....see I just saved you tons of work :D
******** SADDENZ :rolleyes:
Damn, I looked under the heatspredder... my 2x 256 are CH5 too, just damn nice ones. I NEED 2x 512M BH5..... any ideas.
********
PC... you measure Vdimm on the LEFT leg of the Vdimm mosfet...the BIG one near the Vdimm mod resistor. The RIGHT leg is part of the flyback/regulator circuit and isnt really relevent, it will be some voltage higher than Vdimm or 3.3Vrail. Only yhr 3.3Vrail at ATX connector, and the Vdimm leg (and Vdd pins of DIMM connectors...pin7,15,etc) matter. 3.3Vrail needs to be HIGHER than the Vdimm you want to set, or the regulator wont operate correctly. Vref is measured on pin 1 of DIMM socket.
Usually this involves a POT inside the PSU, or a mod to the BROWN 3.3V sense line on the ATX connector. Read about 3.3Vrail mod's in other threads here.
just dropping in so i get regular updates.... keep up the great work!
________
VN1500G
*****WELCOME TO XTREME RABBI_NZ*****
good ot have you here, welcome to the addiction!!
uwacme, believe me, 2x512 MB BH-5 ( i have 1GB kit of PC3500 LvL II lying around waiting for Socket939 :D ) sucks big time as far as OCing goes... i just pulled 'em out again to test it two days ago and I'm still dissapointed. here's a cocnclsion about 11-2-2 cas2 tests:Quote:
Originally posted by uwackme
Damn, I looked under the heatspredder... my 2x 256 are CH5 too, just damn nice ones. I NEED 2x 512M BH5..... any ideas.
1. before any mods - it hit the wall at at exactly 217FSB (2x512 Dual) at 2.9v... that's primestable for 8hrs (i stopped ir after)
2. after vdimm and vdd mods - guess what? same thing... even 3.4 volts can't make it to work at anything over 220FSB (i'm still talking prime95 stability)
3. after vtt mod - still dissapointed...
but... 1 stick 512 passes 3dmark01 with flying colors at 242FSB...
so one double-bank BH-5 is still sort of OK, but adding second one brings whole setup down down down...
2x256 + 1x512 maxes out at 232FSB...
so 2x512 BH-5 at 11-2-2 cas2 on NF7-S is no OCer's dream...
my pc3500 level2 did 2-2-2-11 229FSB @ 2.9v MemTest stable... didnt try Prime tho..
________
TS90
The truth is nForce2 mobos dont like double sided memorys, one side memory --> :toast:
With VBT/VTT mod Prime95 stable @ 240.5*10.5 ~2h cas2-2-2-5 Dual 1:1 and Prime95 is still running:cool:
:p: sorry, but memtest means nothing, even new one.... i get no errors in memtest at 251FSB ATM 2x256 11-2-2 cas2, but it's nowhere near stable. goldmemory, that's what stresses memory.Quote:
Originally posted by Rabbi_NZ
my pc3500 level2 did 2-2-2-11 229FSB @ 2.9v MemTest stable... didnt try Prime tho..
My KHX3000*2 did 247FSB cas2-2-2-5 Dual 1:1 for 3h MemTest 1.11 Test #5 but 248FSB could not pass 1 minute:sick:
Yes, I agree, its hard to get it there, but I'd be happy with 250Mhz x10 at 11,3,2,2 from BH5 that's still almost 10% faster than my 230 now. Man that last 10% is a beeatch...but isnt it always ;-) But I figure if my setup is getting this much from CH5 then BH5 can only do better....one would hope? Effectively I have 2x doublesided at the 1Gig setting now.
Other factors come into play, BA you may just have a northbridge that needs more hand holding....lapping, cooling, more/less Vdd. But I know what you mean about memtest....245@ 11,2,2,2 with my 2x 256 but they crap out the second I go to do something real beyond SiS memory bench or CPUZ, browsing, etc. Any 3D and crashola. Where as 245@11,3,3,2.5 is dead bang solid. But yes, the high FSB leaves less and less time for data transfers per access, so it may very well be the northbridge OR CPU that are holding me back, regardless of what memory I ever use. Going to 3,3,2.5 leaves waitstates and more "time" per access.
Also, this harkens back to the point about Vtt at the northbridge end of the memory bus, if we dont have a balanced transmission line we may be having our performance ceiling artificially limited by incorrect voltage. But a 250Mhz FSB is a 25% overclock of the northbridge's bus, it is alot to be asking.
I dont think cooling has any impact on my setup, the northbridge and CPU are well under control.....swifty barely gets warm, and the NB is lapped at 1000 grit and AS3'd. CPU stays in the 40's when being thrashed.
But people have done it....though I wonder sometimes how "stable" they are when they say it. To me stable is running all tests without error....memtest/winmemtest/prime/3D2K+03 and gaming for hours on end.
In the meantime, if you decide you dont like your 2x 512M BH5... give me a call :D
honestly, i've been playing with mu current setup for about two weeks, after vtt mod... and i'm about to retire on this hunt for 250 FSB 11-2-2 cas2 Dual STABLE :D
vdd 1.6-2.2v, vdimm 3.0v-3.5v, vtt 1.50-1.75v... being there and done all that... even 80mm tornado blowing on SB or NB, all Integrated with SB devices OFF...
at this point i think it's my mobile (for which i feel 244 is hell of a stretch already!) not catching up with FSB request or sth :D
but i see folks on my native ( Polish ) forum having 250+ FSB with 2x512MB of some Hynix43 stix at lousy 11-4-4 cas2.5. but still, that's over 250FSB on NF2...
or like u say - there's something else that needs to be modded on this board to take BH-5 / CH-5 higher at tight timmings...
i wish we knew DFI's secret to high FSB :brick:
oh, my 2x512 feels pretty comfy here and doesn't like to be separated from it's owner and 2x256 buddy... i need those for upcoming upgrade - Socket939 + FX-53 + R420 :banana:
Hi there,
since Vtt isn´t the secret, what should be the secret?
In my opinion the secret is in chipset... imo there are many strange things with this chipset,
-Vdd don´t have any influence on NB stability :confused: anyone can explain why, or any1 have any theoric idea to explain it?
-Many ppl have their NF7 stable at certain FSB @ agressive timmings (me for example, 243MHz @ 11-2-2-2 (7 hours memtest #5 loop) and 1 MHz more, give error on memtest after 2 minutes of test, even slowering down timmings (11-3-3-2,5) and increasing Vdimm won´t help, with this I conclude that RAM ain´t the problem and "messing around" on Vtt/Vref/Vdimm... won´t help, specially for ppl with BH-5.
-Could it be the BIOS? Really don´t know, but I start to believe it could be, due to the fact that DFI nF2 have options for other timmings besides Tras, RAS, RAStoCAS and CL and maybe this timmings are the secret.
-Maybe, like uwackme as been saying, Vtt at the northbridge end of the memory bus need to be upped aswell.
I just know that I can´t run stable more than 243 MHz and my BH-5 handle more than 243...
Just a side note, I was looking at the A7N8X board and found that it sets things up....Vref drives Vtt and Vdimm circuits, so they all track. The Vdimm mod for that board actually mod's the Vref which cascades to the Vtt and Vdimm.
someone should lap their nb down to the core and direct cool the core.
my friends can run
252MHz
2 x 256MB single side Hynix D43 Rams
2.5-4-4-11
Prime 95 ......3 hours stable...
no vdimm mod or vtt mod
seem it is something related to the bios limitation
i think the bios is not build for 2-2-2- timing as DFI Infinity bios
;)
That´s a great boost, tictac, is it a new mobo? :)
What Vdd and BIOS have u used?
And what kind of memory bandwidth do u get with those timmings?
Cheers
to be honest with u tictac, this is not really unusual...Quote:
Originally posted by tictac
my friends can run
252MHz
2 x 256MB single side Hynix D43 Rams
2.5-4-4-11
Prime 95 ......3 hours stable...
no vdimm mod or vtt mod
seem it is something related to the bios limitation
i think the bios is not build for 2-2-2- timing as DFI Infinity bios
;)
and i'm also thinking that NF7-S not hitting over 250FSB 11-2-2 cas2 is due to BIOS not being tweaked for BH-5...
on my native (Polish) forum there's actually many people that are withing 240-250FSB range stable in Dual. there's even a guy that has 255FSB 11-4-4 cas2.5 2x512MB!!! on NF7-S with Hynix D43 sticks... most people there don't have that much money to get BH-5 so they are "stuck" with those...
maybe there's something more u can do as far as BIOS tweaking?
P.S. I wish i knew about that "mobo action" for u... i would definetely participate in it as i really, really appreciate your work, especially multis fix...
Yes, I think we are missing alot of bios tweaks that the DFI bios has in it. Oskar over there is doing his best to find the right combo of settings.....but he works for DFI, so I doubt they'll let him tell the world. Abit is putting in ZERO effort in this, the design is done and the product is "dead" to them, as they see the world moving to A64.
Too bad, AMD really fk'd up the memory controller design.... the act of building it into the CPU was utter stupidity. They messed up the whole AGP/PCI lock issue, severely complicating goosing the A64's. Maybe 939 will address these things, as they curve-the-learn.
Im sure there is "gold" left in the NF2 northbridge, but we need to find it. Ill email NVidia and try to figure out how to get the technical reference manual. How can it do them HARM be letting that information out? The device is a synchronous state-machine, it's all a matter of programming it to take the correct steps on an access cycle vs the ram you are using. 2x512M just needs a pause in the right place probably, and DFI has stuck the pause in to make 2x 512 happy at higher FSB. We'll find it somehow.
In the meantime, Bestbuy Sunday has a Kingston 512M valueram $54 (after rebate) that suposedly has Hynix xxxBT-D43 chips on it. Thats good for 250Mhz 11,4,4,2.5. Not BH5, but high FSB for real cheap.
BUFF.... that is my friends board not mine
Bachus_Anonym... no problem man... when i get my board... bios modding will get some serious modding job... I have xtra bios chip here... may test 5-10 bios per day.... :thumbsup:
My board still on shipment... hope will arrive soon
UWackMe... ya i think i will just buy those cheapo hynix-d43 rams when i get my board....
:D
250MHz 2.5-4-4-11.... bandwith not good enough... but hey its 250...:D
bachus,
sorry bout not letting you know we were getting a board for tictac... we sent out a lot of PMs and lost track of who had been informed and who hadn't.
tictac,
those new Memory Timings (?) tables you found... how big are they?
And is it possible to open then in a text file?
It shouldn't be difficult to write a program in Delphi/C++ or whatever else that compares the files and locates any similaries and differences between them.... might help decypher something :D
Also, do the DFI boards have a similar table... or is their BIOS structure completely different?
________
buy silversurfer vaporizer
Out of curiosity, why is building the memory controller onto the CPU bad?
Another possibility - Can the infinity bios be flashed to the nf7? 8rda+ and a7n8x's can be used for reflashing DFI's, so maby just flashing an "abit" to a "dfi" could be a simpler route?
Makes me wonder if flashing my NF7 bios with an Infinity would open those other ram timing options?Quote:
Originally posted by STEvil
Out of curiosity, why is building the memory controller onto the CPU bad?
Another possibility - Can the infinity bios be flashed to the nf7? 8rda+ and a7n8x's can be used for reflashing DFI's, so maby just flashing an "abit" to a "dfi" could be a simpler route?
I have 2 other NF7s here that I am modding for people. I may as well try it to see what happens. Can always do a hotflash if needed.
What ya think STE...?
its a small table compare to romsipQuote:
Originally posted by Rabbi_NZ
bachus,
sorry bout not letting you know we were getting a board for tictac... we sent out a lot of PMs and lost track of who had been informed and who hadn't.
tictac,
those new Memory Timings (?) tables you found... how big are they?
And is it possible to open then in a text file?
It shouldn't be difficult to write a program in Delphi/C++ or whatever else that compares the files and locates any similaries and differences between them.... might help decypher something :D
Also, do the DFI boards have a similar table... or is their BIOS structure completely different?
the d10 value = 10-11
the d23 value = 11-12
the dfi infinity = 17-18 /18-19
I'm no expert bout BIOS modding but I think flashing with an entirely different BIOS (from another board) is not a long term solution to anything....
But... if tictac or someone else could locate the NB timing tables in as many BIOS's as possible (E.g. NF7, Infinity,LanParty, etc) then we may be able to experiment with different timing tables....
There must be a portion of the BIOS code that is dedicated to the nForce 400 Ultra.... the more we learn about that the more we can narrow down where the timing options are...
But flashing an entirely different BIOS cannot be good... the boards have different integrated peripherals that are driven by the BIOS settings.... in my opinion that's asking for trouble
Can't wait till you get your board tictac... hopefully we'll see some more breakthroughs when you have your very own baby to experiment on! :D
________
volcano vaporizer classic
are those values addresses or is that ALL there is to it?Quote:
Originally posted by tictac
its a small table compare to romsip
the d10 value = 10-11
the d23 value = 11-12
the dfi infinity = 17-18 /18-19
seems a little simple????:confused:
________
buy iolite vaporizer
I agree totally with your reasoning, and wouldnt have offered to do it if I did not have a way to fix the issue if something happened.Quote:
Originally posted by Rabbi_NZ
I'm no expert bout BIOS modding but I think flashing with an entirely different BIOS is not a long term solution to anything....
But... if tictac or someone else could locate the NB timing tables in as many BIOS's as possible (E.g. NF7, Infinity,LanParty, etc) then we may be able to experiment with different timing tables....
There must be a portion of the BIOS code that is dedicated to the nForce 400 Ultra.... the more we learn about that the more we can narrow down where the timing options are...
But flashing an entirely different BIOS cannot be good... the boards have different integrated peripherals that are driven by the BIOS settings.... in my opinion that's asking for trouble
Can't wait till you get your board tictac... hopefully we'll see some more breakthroughs when you have your very own baby to experiment on! :D
Sometimes people have to be guinea pigs and test out ideas to see what happens.
I was thinking, as was STE I am betting, that if the bios flash indeed worked, then we atleast know that the tables can be brought over from one bios and put into another. Or however it is that you bios modders do your magic. I dont know if its as simple as grabbing something from one bios and adding it to a new one. My thinking was that if a bios flash opened up the other ram timings, then maybe that knowledge would help tictac and the others out.
I was just willing to be the guinea pig if it helped someone have a breakthrough.
Experimentation is the root of all achievement. Sometimes a bold move breaks a plane to a higher ground, rather than siutting and studying for hours on end.
Not that both methods arent good, on the contrary, doing it hte proper way is always best, but sometimes you just gotta suck it up and give different ideas a try.
here you goQuote:
Originally posted by Rabbi_NZ
I'm no expert bout BIOS modding but I think flashing with an entirely different BIOS (from another board) is not a long term solution to anything....
But... if tictac or someone else could locate the NB timing tables in as many BIOS's as possible (E.g. NF7, Infinity,LanParty, etc) then we may be able to experiment with different timing tables....
There must be a portion of the BIOS code that is dedicated to the nForce 400 Ultra.... the more we learn about that the more we can narrow down where the timing options are...
But flashing an entirely different BIOS cannot be good... the boards have different integrated peripherals that are driven by the BIOS settings.... in my opinion that's asking for trouble
Can't wait till you get your board tictac... hopefully we'll see some more breakthroughs when you have your very own baby to experiment on! :D
You got skills in the bios dept there tictac.
I have no idea what I am looking at there.
Are those ^^ timings tables?
________
Harley-Davidson FLSTC
DFI Infinity 12-18 bios mem table compare with 2-17 beta 2
:toast:
well cant really confirm it is memory tableQuote:
Originally posted by Rabbi_NZ
Are those ^^ timings tables?
until we experiment it... :toast:
ops sorry.... the d23 on the left & the d10 on the right
typo error....
hope u can decrypt it well...
lol, that's quite a challenge! :)Quote:
Originally posted by tictac
ops sorry.... the d23 on the left & the d10 on the right
typo error....
hope u can decrypt it well...
a lot of numbers are in the same place for Abit and DFI tables tho... that's got to be a good thing... I'm just wondering whether or not these are timings (I HOPE THEY ARE :D)
If they are not timings tho... the numbers that differ between different tables may have adverse effects on other BIOS functions... for example if they are related to register variables used throughout the BIOS code it may stir up some sheeet later on :cussing:
i guess the only way to learn is to experiment and take one for the team now and then like Hell-Fire said.
tictac,
do you want an apprentise? lol :p:
once i get my pc back together i wouldn't mind trying to learn a few things from the masters!
________
ZG1200A
Hopefully a hotflash is all thats needed ;)
I think an 8rda+ would make a good test subject for the DFI flash seeing as the bios chips are for sure compatable.
Well, its a bad idea.... memory controller in the CPU.... because it limits all further design to what can be done with the memory controller AMD supplied. Any changes in memory, and innovative design ideas of AMD's customers....well, fk'em.
It was dumb, but I see why they did it.... so few good engineering outfits left on planet earth, the only way to assure a working controller for thier product, was to build it in :slobber:
In the meantime, the correct method would be to make the A64 a 64bit differential databus....a 64bit instead of 8bit "hyperlink transport". Then the memory controller would do what I used for a Macintosh accelerator over a decade ago.... 4 DIMMs in a 256bit wide memory array. The CPU will ALWAYS access memory as a (8) 64bit word BURST. Knowing this, the memory controller would access 4 DIMMs in parallel....DDR, so 2 accesses per DIMM... and pull in all (8) 64bit items....4 in parallel, twice in a row. In the meantime, feeding the data to the CPU at....800-1000Mhz FULL BLAST over the 64bit differential bus. Right now we see the Athalon's limitations in the 250-300Mhz range MAX. While as a hyperlink-transport bus, a REAL 800Mhz bus is possible....er NOT Intel marketing BS 800Mhz, but REAL 800Mhz.
This isnt rocket science, this is what ATI and NVidia are doing right NOW on thier video boards.
The principle bottleneck is the memory bus, here we are struggling to get our 500Mhz achieved... when AMD can blow this all out of the watter anytime they feel motivated. Problem is, they aren't motivated. And only AMD/Intel can do this, 3rd parties dont control the CPU design.
Just my opinion.
BTW: my Mac accelerator... had a memory bus just shy of 8X the motherboards built in memory controller. 32bits wide, and alot of waitstates in a very conservative setup dont compete well with 128bits wide and low waitstates.
PS: with a 256bit memory controller....4 sticks of PC133 Sdram could beat the NF2 with BH5 at max.
whats to say the memory controller doesnt have some of those capabilities built in already?
Ummmm... cause it doesn't. At least it does use a 128bit wide data bus, and does sort of "dual channel".... unlike the NF2 which doesnt really make the most of the concept at all. NF2 really uses the 2 banks to try and hide memory accesses by subsystems between CPU accesses...audio/lan/ide/pci, etc. Which is cool, but doesnt pull off dual channel as well as the hype might lead you to believe.
Excercise for student.... using SiS memory test, turn off buffering, and SSE/SSE2, etc, settings so we are down to pure raw MBytes/second transfer rate. Then compare a P4 on Canterwood, or A64, with NF7-S. You'll find they both outpace the NF2 board because both implement "dual channel" much closer to FOR REAL. if NF2 was more truly dual-channel it would be a much closer speed comparison.... 250Mhz 128bit bus will run close to the same MB/s regardless of AMD/Intel, unless not implemented well.
Inside the A64 it is a 64bit path from memory controller to CPU guts... per AMD block diagrams. P4/Canterwood same diff, all 3 are very similar in arrangement, but diverge in how well they make use of the layout.
It looks like Canterwood applied to Athalon's would truly KICK ASS, combining an excellent memory controller with the Athalon's more sensible core design.....nothing like branching in Word to upset a 25stage pipeline. It is the internal design of the Athalon that lets it keep pace with P4's in everyday uses... business/gaming. BTW, there is no reason one could NOT slap an Athalon/Canterwood board together... from an engineering design, but of course Intel prohibits this in their contracts. And also prohibits P4's from being slapped onto NF2's.... there is no technical reason prohibiting any of it, the bus interface of either chip is nearly identical.
When comparing apples-to-apples you need to stick to this sort of test across the platforms. You'll see a real difference and it's all pure MB/s data movement.
So hat's off to AMD for that, but they can do SO MUCH BETTER and blow Intel completely out of the water.... so for the life of me, I dont undersand holding back deliberately. Weird.
BTW, with the memory controller on-chip, they dont need to use a differential databus. But to talk outside the chip more than 250-300Mhz it will be required.... hence the 8bit diff bus between northbridge and southbridge in NF2, SIS648, Canterwood, etc running a good 800-1000Mhz.
The engineering world has gotten a bit FK'd up. Years ago Motorola and Intel would be falling over one another to get you to use thier product.... now you need to be granted permission to even THINK about using one of thier products... welcome to the world run by semi-monopoly's.
Hi to all,
I've done the 2nd method of the Vtt mod from Tictac's webpage and it worked for me as well! Thanks Tictac.
Now i can bench into windows, but the stability issue is still on for me... don't know why tho, i'll have a look at those BIOS's and some extra cooling on the Vtt chip, the rest is watercooled :D (except mosfets).
Btw, how i can give some more volts for the mems?
Saw on Tictac webpage the Vdd booster mod ?!?!, by adding only one cable, and by using another cable you connect the 3.3v rail directly to the Vdimm's (is that safe m8???)
Cheers,
Byron
My Mobo won't go prime98 at 240 anymore.... :eek: Why...?
I don't know,... is it because the 4th mod of VDD,... ? I don't know yet,...
But I see the performance is drop time by time,... drop again and again,... Maybe U're right Dani, the best way to mod VBT/VTT is the "cut trace" method... :o
Hmmm I'don't know is it because VTT mod, VDIMM mod or VDD Boster mod,... but every time I use this 3 in one mod my Board Goes down time by time,... the 1st day work so fine, it's can prime at 243, but I've got a big problem on stability at 4th day
Whua..... my RMA rate just increasing time by time,... it's about 8 Dead Board here... :( in about 19 day....:rolleyes:
Is there anyone got the same problem with me ?
Please ... I need some advice for it (PM_me maybe....):slobber:
Can the people that did the cut-trace method tell me if their Dual N mosfet is getting hot?
Has there been any board deaths because of the cut-trace method yet?
________
Kawasaki VN1500G
oh yeah, it is by far the hottest single component on my mobo...(excluding vcore mosfets). measured with temp probe under load at 3.4vdimm and 1.7vtt i get 45C...Quote:
Originally posted by Rabbi_NZ
Can the people that did the cut-trace method tell me if their Dual N mosfet is getting hot?
Has there been any board deaths because of the cut-trace method yet?
still live and kicking, though... i'll be getting dremel kit soon to cut another hole in my side panel so i could mount 120mm fan blowing over mem 'n this mosfet area :D
bachus, have you had any trouble at all with the current flowing thru that Vtt VR?
I know that this is an issue on the IC7 line, so just checking.
Just finished doing the trace cut method, and very happy with the results. But want to make certain that the current isnt going to overwhelm the poor little VR and cause a melt-down.
I tried the "Vref = Vtt" ... a no go for me.
not 100% sure what u mean...
but i was wondering if out of the blue the mosfet got this hot then maybe it has some impact on that pot too... maybe i will measure the temp on it too ;)
i'm not at home right now but i will do that later...
ATM, i'm struggling to get my setup prime stable at 244FSB... i mean it is prime stable but for some reason 6th hour finishes with BSOD... i think it's time to revise my 2.1vdd :D or my mobile can be the one to blame too...
glad u done it too.... i will retest it tommorow but i 'm 95% sure that vtt mod didn't help me much... i will confirm that or not later...
Well, to put it more clear:
On the IC7, to get 1.6v or more on the Vbt, the VR is practically at 0 ohms...it is around 1/2 ohm, which puts the current at a whooping 3.2Amps...oucherz. Those little VRs dont stand a chance against that amount of current.
But, MadCat put together a fix for that...so all is well.
I am not sure what the resistance is on the VR when Vbt is at 1.6v or so..can you check that easily and let me know.
no problem :) i will check that for u tommorow when i get home...
I thought with the cut-trace method once you set the Vtt VR to approx 200ohm it will track a VDimm mod of approx 3.2v max as a 1:2 ratio..?
If I do this with mine (eg, limit vdimm to 3.2v) my VR will never get to 0ohms will it?
So i won't have to worry about the overcurrent issue as much...?
________
Suzuki Katana
Actually, you set the VR to zero ohms and increase the resistance to increase the Vbt voltage.Quote:
Originally posted by Rabbi_NZ
I thought with the cut-trace method once you set the Vtt VR to approx 200ohm it will track a VDimm mod of approx 3.2v max as a 1:2 ratio..?
If I do this with mine (eg, limit vdimm to 3.2v) my VR will never get to 0ohms will it?
So i won't have to worry about the overcurrent issue as much...?
It does not auto-track as far as I am aware..you have to adjust hte VR to the correct resistance to achieve the needed voltage.
As far as i've observed, the Vtt increases the same when you increase the Vdimm automatically.
For example, i did the Vtt with the cut trace method (and 1K resistor), set up the Vdimm to the lowest (2.6v bios) set up my Vdimm resistance to give me 2.90-91v's and the Vtt to 1.46v's.
When i raise the Vdimm to 2.8 (from bios) which is approx. 3.11v, the Vtt is changed automatically without doing anything to 1.56v (half the Vdimm).
Byron
Once done, and set, the Vtt will track Vdimm... but the whole "window" of setting moved from stock bios range...2.6V-2.9V to a new range... 3.0V-3.3V (as set in my case) because of the Vdimm mod.
Within this new range, they track correctly. if I alter the Vdimm pot... I must then also alter Vtt pot, but once set, they track within the new range of bios adjustment.
The circuit the pot is in is a low current feedback loop, no risk of burning the pot up. Set for 3.0-3.3 range, my pot value was 90ohms.
It seems an incovenience, but once you dial in the new range, you wont have to bother touching the pots again....so it isnt really.
Did you do the Cut-Trace method or Vref=Vtt?Quote:
Originally posted by bachus_anonym
oh yeah, it is by far the hottest single component on my mobo...(excluding vcore mosfets). measured with temp probe under load at 3.4vdimm and 1.7vtt i get 45C...
still live and kicking, though... i'll be getting dremel kit soon to cut another hole in my side panel so i could mount 120mm fan blowing over mem 'n this mosfet area :D
Does the MOSFET get hotter with some Vbt Mod methods than with others?
________
vaporizer
i have set up 2.9v in BIOS to be 3.4v in reality. the vtt is set to 1.71v and the resistance is 259 ohm.
lowering vdimm in BIOS or using nf7svcore in steps to 2.6v results in vtt tracking vdimm down in 1/2 correct ratio...
so everything works just fine.
of course it doesn't track down when u change vdimm using pot rather than BIOS or nf7svcore at the same. e.g. 2.9v preset in BIOS.
Yes once you do the cut-trace, the mosfet gets warmer than it had been, but that's only natural, 1.3V to 1.6V is a 20+% boost in the power dissipation. I slapped microcool heatsinks on both mine, no probs since, warm not HOT. Make sure you have air flow over the rams.... my LX-6A19/SLK3700/Superlanboy/D8000 case design has a fraction of intake air moving over the rams, the way the HDisk cage is laid out.
true... that's why i'm ordering dremel kit and 120mm fan to make little modification to my side panel.. yeah, 2x120mm fans that will be ;)
Here's my setup...Quote:
Originally posted by uwackme
Yes once you do the cut-trace, the mosfet gets warmer than it had been, but that's only natural, 1.3V to 1.6V is a 20+% boost in the power dissipation. I slapped microcool heatsinks on both mine, no probs since, warm not HOT. Make sure you have air flow over the rams.... my LX-6A19/SLK3700/Superlanboy/D8000 case design has a fraction of intake air moving over the rams, the way the HDisk cage is laid out.
Antec SuperLanboy - modified to have a 120mm Blowhole exhaust aswell.
Bottom front (120mm) blows in, top back (120mm) blows in through radiator, top blowhole (120mm) is exhaust, plus PSU is exhaust.
Making up a bracket for my Smartfan2 (80mm) so it blows directly on the ram and mosfet.
Got a bunch of microcool sinks to put on all the mosfets and voltage regs.
I know nothing is "safe" or "gauranteed" with oc'ing... but from experience, would you say my setup should be adequate for 3.2vdimm / 1.6vbt?
________
cheap vaporizers
This is very useful information.Quote:
Originally posted by uwackme
Once done, and set, the Vtt will track Vdimm... but the whole "window" of setting moved from stock bios range...2.6V-2.9V to a new range... 3.0V-3.3V (as set in my case) because of the Vdimm mod.
Within this new range, they track correctly. if I alter the Vdimm pot... I must then also alter Vtt pot, but once set, they track within the new range of bios adjustment.
The circuit the pot is in is a low current feedback loop, no risk of burning the pot up. Set for 3.0-3.3 range, my pot value was 90ohms.
It seems an incovenience, but once you dial in the new range, you wont have to bother touching the pots again....so it isnt really.
Thanks.
The automatic vtt mod didnt seem to work for me...
I have used vdimm mod where I removed resistor and connected the closest to ram solder point to the farthest leg on the mosfet though VR.
I've set it up to that the bios to 2.9 and checked with voltmeter...it showed 3.31v on vdimm and then checked vtt voltage ont he double mosfet, 2 left pins showed it as 3.41 but 2 right ones showed them as 1.45
Is there anything i could've done wrong that would cause it?
BTW, about 2 mins later i went to check voltages again and shorted 2nd and 3rd leg of the mosfet... and pooof... gonna RMA the board tomorrow :\ so now i have some time to try to figure out what went wrong...
Thanks
Better hope your ram is still ok. Sounds like you messed up and had 3.XV on the Vtt/Vref line ..... badness. Vtt/Vref not to exceed 1/2 Vdimm period.
Key to no shorts...SHARPEN your multimeter probes...literally, use a file. Also, scotch tape, or coat in silicon glue all but the tippity tip. I use Huntron Tracker probes...cost me $15 at the time, been with me for 18 yrs now....needle sharp, insulated except the very tip.
Also, I have a Weller EC2001 soldering station spare I'll let go for $100 to someone in the USA.... costs $300ish new.
Friend just tested my ram for me, its fine.. phew...
The board is also already on the way to ABIT...
Next time ill be ALOT more careful :)
Guys, one question, why my Vdimm mosfet is DAMN cold????
I'm feeding them 3.2v's.... and its on room temp! :)
Tought it was the fan which i've blowing air on the Dimm's but even if i stop it, it still doesn't get even warm.
The only thing which gets warm is the Vtt tiny chip, and the thing is that i can't fit properly a small heatsink i've got with thermal tape...seems i've to add some artic epoxy or something to be perfectly in contact.
Cheers,
Byron
I thing this mod didn't help me much for higher FSB, I need VDD more than 1.85V to run perfect prime at 240 Mhz....
Please can you show me the effect of this VTT/VBT mod ?
I have fail at Vbt=Vdd mod (burn your Vtt chip) and VBT=Vref mod (make Your board goes down time by time), but maybe anyone here got a great thing to share about the "cut-trace" method...
or can you find some other way for it ?
More indepth testing..... and weird things appear.
On D10, with 2x 256M CH5 HyperX3500, multi 10x or 10.5x.
The rams pass memtests and ALL...ALL 2D testing at 11,2,2,cas2 @ 3.3Vdimm 1.65Vtt 1.85Vdd 1.875Vcore (2.520 Ghz)
BUT... heres the weird part. 3D testing fails. Isolated finally to the ram timing setting... xx, x, 3, x running top-to-bottom as listed in D10 bios. At 11,2,3,2.0 everything is fine, but 11,2,2,2.0 3D stuff crashes.
So the xx,x,3,x setting is touching on a northbridge/AGP issue. if I use a PCI 3D card, 11,2,2,2 runs fine.
No setting related to the AGP bus made any difference in making xx,x,3,x work at xx,x,2,x. I tried every combo of Vagp, 4x/8x, fastwrite/not sideband/not, AGP frequency, AGP latency that I could think of.
So the traditional 11,(3),x,x need of CH5 was gone at high Vdimm but what I hadn't bothered to notice previously was the AGP related need for 11,x,(3),x setting for happyness.
From the way I see it, both NVidia and the OEMs have done a piss poor job of configuring the memory controller.... and more so, letting skilled endusers get access to all the settings we need to make the most of the product.
Still no BH5 for me to test, I'll be interested in seeing if thier is any difference with everything else the same, vs CH5 in this matter.
So far, longterm, the cut-trace/pot version of the Vtt mod is fine, the mosfet isnt burning out, the Vdd-boost mod is fine, as are the motherboardFAQ's version of Vdimm and Vdd. 5 NF7-S's done up, all reporting no problems.
I did the 3 in 1 mod
So far no problems. I made the vdimm = v i/o detachable.
The best I could do with these mods is 242Mhz FSB.
Screenies:
http://kaine.studentenweb.org/pifast52.45.jpg
http://kaine.studentenweb.org/nf7-1.jpg
http://kaine.studentenweb.org/nf7-2.jpg
More info:
So I completed testing at 240Mhz x10.5. I turned all the goodies back on, 80Mhz agp (40Mhz PCI), 8x, fastwrites, agp apeture at 256M, etc. Ram settings at 11,2,3,2.0 cpuint enb. 9800pro at 450/410.
Totally 3D stable at this point, works like a charm, IE: fastwrites on/off, AGP freq 66-80Mhz, AGP apeture 64-256M all have NO bearing on stability. So I've hit my highest so far's.....
3Dmark2K1... 20796
3D03.... 6810
SiSoft mem: 3702/3463 cpu: 9631/3981
My hypothesis is the following.... the CPU to DDR interface in the northbridge is perfectly happy with 11,2,2,2 but the AGP to DDR interface needs 11,2,3,2 to be happy. The bios settings dont allow seperate entries for the two access modes. So we have to settle for the lowest common "happy" point.
So 3D instability can easily be NEITHER the CPU nor the RAMs fault, even though you might be led to "think" it was related. Something to keep in mind in all our future OC'ing efforts.
Proof positive will come from testing with a PCI based ATI9200 card, so NO agp access occurs at all.... I dont think there is a NVidia video PCI card that does all the DX9 tests.... GF2's dont.
Hence why you can have MEMTEST give perfectly stable results, no errors, etc, and yet you are NOT 3D stable..... now this makes sense. I dont know what about RAS-precharge causes this, but it's a datapoint for us.
EDIT: hmmm, spoke too soon? Dunno, went and tested FARCRY and it crashes at 11,2,3,2 but runs fine at 11,3,3,2.5.
Now at 250Mhz x10, 11,2,3,2 and everything passes for benchmarks, Ill have to set D3D back to no AA/AF and bench 3D again, but 3Dmarks pass with AA/AF on high ... 4x/8x... without crashing.
Hal, on the vdd booster that you did you soldered to a different point than what is on tictacs site, I guess it doesn't matter which point it is soldered to?
Ehm ... really? That's odd, everything is working though :)Quote:
Originally posted by rc7
Hal, on the vdd booster that you did you soldered to a different point than what is on tictacs site, I guess it doesn't matter which point it is soldered to?
Maybe I found out a new mod? :)