Did you say you had 1.74v somewhere ?, cause Im told there is rumour Gainward crippled the ram with 1.7v instead of default 2v.
Did you say you had 1.74v somewhere ?, cause Im told there is rumour Gainward crippled the ram with 1.7v instead of default 2v.
my vgpu is at 1.37 at the moment while the mosfet leg in question reads 1.24.
i twisted vgpu up to 1.41 with no change to the 1.24.
it seems to me the legend cards have stock volts at 1.3 while the gainwards are around 1.2v.
Yes exactly 1.2v stock on 2D and 3D
where did you here that bud?Originally Posted by humeyboy
and persivore if you got any ideas how i should start attacking the 1.74v reading please let me know. i'll feed it 5% more and look for any change.
One of guys on Guru3d where we got a few threads going about these cards although no mods, again it was hearsay, I wont say its fact, I kinda knew as soon as I lifted cooler and say the 1.2ns Samsung it would be crippled.
Last edited by humeyboy; 05-11-2006 at 04:44 AM.
ok well my friend with a pencil-mod on his card sais his resistor (marked below) is at 1.4kohms with a 1.5vgpu result.
default on our legend cards are 2.3kohm on the marked resistor with 1.3vgpu default
any similar results from gainward/pencil-mod users?
he also reported that at 1.5vgpu the memory speed had improved +- 10mhz to 391mhz (upfrom +-380mhz with 1.4vgpu)
lastly he has some real ghetto cooling which i'm trying to get hold of pics of
Last edited by s e t h; 05-11-2006 at 01:19 AM.
Not sure, the 1.4V pencil mod got me more than 10mhz on ram more like 75mhz+.
Some guy on VR is claiming you may get more not using pencil (I cant post cause account was lost in downtime as it was only few days old and and I'm still waitin on email reg info to actually let me post), well you can tell him I used a mechanical pencil and it was 1.8v lol, I had to remove and do with normal wooden with graphite HB or 2B, idle temp was 59C, I turned PC of immed.
Last edited by humeyboy; 05-11-2006 at 07:11 AM.
i tested so far up to 1.45vgpu with no extra ram speed. i tried with a VR not a pencil though.
in other news i have again tried pin#6 of RT9214 to ground. this time i pencil-modded it to give me 2.25vdd to the memory voltage read points.
no change again even though i'm actively cooling memory 'sinks.
while going through the RT9214 datasheet i noticed what appeared to be OCP.
i just though that since i have tested 7.5% extra vdd without the memory failing earlier the OCP might therefor not be the problem
i could of course be wrong again.
http://www.datasheetarchive.com/AdFr...rticle=2877975RT9214 datasheet - OPS (OCSET, POR and Shut-Down) (Pin 7)
This pin provides multi-function of the over-current setting,
UGATE turn-on POR sensing, and shut-down features.
Connecting a resistor (ROCSET) between OPS and
PHASE pins sets the over-current trip point.
Pulling the pin to ground resets the device and all external
MOSFETs are turned off allowing the output voltage power
rails to float.
This pin is also used to detect VIN in power on stage and
PHASE (Pin 8)
Connect this pin to the source of the upper MOSFET and
the drain of the lower MOSFET."
^ thats the datasheet if anyone would like to check and comment on the OCP maybe being the problem.
also tested pin#6 of LM358 with a 50k VR to ground.
managed to push output B (pin#7) to 5.9v up from 5.5v without any success
(from this datasheet: http://www.national.com/ds/LM/LM158.pdf)
so that leaves like only 5 possibilities for potentially fixing the ram issue
1. OCP on RT9214 is the issue
2. output A on LM358 is the issue (pin#1 controlled by pin#2)
3. there's another output on NCP5424G (vgpu regulator) that is the issue (can't find datasheet to look for another output)
4. the regulator on the other side of the core is the issue (Fairchild PG1AB - 74ACT08 on these gainwards)
5. or lastly - something i havn't thought of is still the issue.
anyone who has some thoughts or can track down some of the missing datasheetes please do so.
safest next step for me now will be to test output A on LM358 and wait for comments on OCP issue/non-issue and/or datasheets on NCP5424G and Fairchild PG1AB)
Last edited by s e t h; 05-11-2006 at 11:16 AM.
some news from another friend.
in the picture in post#56 above if the resistor is pencilled to 2.2kohm on the legend cards the output is just under 1.4vgpu.
no gainward results yet unfortunately.
i give up
going back to 256mb cards. at least they have one easy voltmod and the memeory actually works
so far these 512mb cards are completely useless for benching.
maybe some games will enjoy the extra ram and for grandmothers running stock they're faster than the 256mb cards.
but for ppl like us so far they're a waste of time.
i'll keep an eye on them and see how they work out still.
anyway hopefully the memory voltage regulator on the 256mb cards will be strengthened in some of the newer pcb designs.
let me know please if someone figures out wtf is wrong with these cards.
maybe i'll find a nice golden one in this pile i have hear. all i want is frikken 800mhz ram.
is that so much to ask
Dont't give up, just keep going...
well i don't know what else to try right now.
i varied every output voltage available in all the datasheets i can find.
i probably come close to killing a card like 5 times in the last 2 days.
i can't take it anymore
these things are too stubborn.
in a few weeks when there are thousands around someone will hopefully find the issue.
lol
there is only 1500 of my agp made, thats a issue with mods
For the convenience of all, I post the easy-to-do vmod for the GPU that works 100% (from vr-zone):
Where to pencil (no conductive pen, just a plain graphite pencil):
default resistance - 1,66kOhm - gives 1,2V GPU
slightly pencilled - 1,5kOhm - gives 1,3V
heavily pencilled - 1,4kOhm - gives 1,4V
More I didn't try and probably won't, since this is one good card and I don't want to mess it up.
on the legend cards the default resistance on that resistor is 2.3kohm with 1.3vgpu
pencilled to 2.2 gives 1.4vgpu
any lower should be tested carefully.
if i get any more results i'll post them.
Last edited by s e t h; 05-11-2006 at 10:08 PM.
It very easy takes little graphite, I ended up 1.8v on 1st attempt, use either on of the 2 resistors above it to measure on side nearest big long black chip.
Im 1.41v got 650core and ram has improved due to this at 1550
Did Pencil MOD on Legend card got following:
resistor 2.19kohm = vgpu 1.39vgpu
resistor 2.11kohm = vgpu 1.42vgpu
3D Mark 2006 succesful @ 700mhz core... 69degC - stock coller
No clock on the ram yet... If I have a stable clock @ 700mhz (damn hot though!) on the core & start to push the ram, will it cause the core to fail @ lower than 700mhz?
Last edited by TraX; 05-12-2006 at 01:04 AM.
hardware flux
New build of NibiTor just out that can read timmings for our memory now
ok then somebody please post 256mb card ram timings and humeyboy please post your gainward 512mb timings.
i'm becoming convinced that if voltage starvation is not the problem then we will find the solution in memory timings.
and damn i just sold all 5 of my cards so i'll kick myself if the solution is found to be a simple bios flash.
edit: nevermind. here are the timings from the legend 512mg 7900gt. anyone with the 7900gt 256mb version of any brand please get NiBiTor 2.9 from here > and post your timings please: http://www.mvktech.net/index.php?opt...filecatid=1658
Last edited by s e t h; 05-12-2006 at 01:25 PM.
[IMG][/IMG]
Last edited by humeyboy; 05-12-2006 at 01:45 PM.
Your card is basically Gainward USA, I think they pulled out of there but thats possibly why using that brand.
I think you guys are making a mistake...you should first hit "Autoselect timingset" and then "Detailed timings". Here is my reading (timingset 7):
I thought that auto selected a new set of timmings, hmm
According to this one has to choose the timingset the card actually uses (I guess all potential timingsets are written in the BIOS, but only one of them is used).
If so, it would be sufficient that everyone just named the set used by his card (e.g. "7" in my case).
If OCP was the problem, I would expect the voltage to just drop down to 0v instead of stopping the memory overclocking.so that leaves like only 5 possibilities for potentially fixing the ram issue
1. OCP on RT9214 is the issue
2. output A on LM358 is the issue (pin#1 controlled by pin#2)
3. there's another output on NCP5424G (vgpu regulator) that is the issue (can't find datasheet to look for another output)
4. the regulator on the other side of the core is the issue (Fairchild PG1AB - 74ACT08 on these gainwards)
5. or lastly - something i havn't thought of is still the issue.
anyone who has some thoughts or can track down some of the missing datasheetes please do so.
safest next step for me now will be to test output A on LM358 and wait for comments on OCP issue/non-issue and/or datasheets on NCP5424G and Fairchild PG1AB)
There is a datasheet for the NCP5424 here. It looks like it only contains one voltage regulator on the chip.
The 74ACT08 is just an AND gate, and probably won't have anything to do with voltage regulation on the card
Can someone check the resistance between the 2 points I've marked in the attachment please.
Connected!
Posting Permissions
Reply With Quote
Bookmarks