Nv 7800 GS AGP Volt Mods

[liquidcooled]

I did the vgpu mod on my bfg from vr-zone to no gain.
used a 100k VR to ground, and over current mod.
should i try a 250k VR instead?

There seems to be some confustion as to where to attach the VR. I have the BFG 7800 GS AGP @500/800!!! Mem speed good but neeeeeed moooore V_CORE!!!

please help.

thank you very much

[liquidcooled]

Anybody have a GPU volt mod that is safe to use. Vr Zones mod dont work on my BFG 7800 AGP 256mb.

I want to try it but peeps here said it was unstable.

Help?

[PAIIITET]

Vr Zones mod dont work on 7800GS [core G71]

Use metod

http://www.xtremesystems.org/forums/...d.php?t=118340

[liquidcooled]

my core is G70, i thought....will it still work?

[jimmor]

my core is G70, i thought....will it still work?

Read the linked thread for peoples experiences doing both G70 and G71 versions of this mod !

http://forums.guru3d.com/showthread.php?t=200420

[Maviryk]

VR-zones volt mod are the same as the ones here. When Shamino posted them initially he had the wrong volt points! O.o He corrected them when I pointed it out.

I'm guessing my old card had the earlier G70 cores. I found that the volt mods did no help OC at all, but just added to instability, with respect to the vcore, but the vdimm, I'm not sure if they hurted or helped the ram clocks. But compared to everyone else, they were unusually high(the ram clocks with the volt mod). Vgpu clocks depend on the quality of the core on the card and the luck of your draw.

As for the newere 7800gs cads with g71 cores I cannot say because I've since upgraded to PCI-E and a 1900xt.

I do not believe cooling is an issue with the 7800gs, within reason. Certain cards have a cold bug if you attempt to cool the card below 38C, throttling kicks in.

[OCPerformance]

I have my own mod

[Nuevototem]

Hi to all,

Here you can see my watercooled nVidia geForce 7800gs overclocked.



More info at my signature.



CYA!

[OCPerformance]

nice i've get some innovatek watercooler

I hope, that I can attack 800 Mhz @ core

[jimmor]

There appears to be a lot of people out there, myself included, with G70 based 7800gs cards which start to suffer from "black screen lockups"; more usually subsequent to any overclocking after voltmodding the core volts regulator?

Following this up, and after much testing, tweaking, and a bit of reverse engineering, have now identified the "black screen lockups" appear to be related to certain incorrect pcb circuit tracks ---> therefore design/manufacturing fault?

One example is that an important voltage ref line of the 6568 chip appears to be connected to a "current/droop sensing" output of the regulated supply, rather than the real output?

There may also be some less than optimum components fitted. However, no real way to determine this without access to nvidia original circuit design?

So the affected cards essentially have a noisier and less stable core regulator. Fortunately, for the manufacturers, the "black screen lockups" problem usually only happens when overall regulator loading exceeds a specific level. On my particular G70 card, it periodically locked up when any combination of core volts and core/memory speeds achieved 3dmark05 of 7500+, or 3dmark06 of 3900+.

Also found that artifact severity and lockups were greately influenced by how mobo AGP communications bus was set. Using ATI tool for overclocking, found that setting AGP above 66Hz and 1.5v dramatically reduced the volts/overclock level at which artifacts and/or lockups would start.

And although using NGO_8456 driver "appeared" to fix the problem for some, any influence on my card was only because that particular driver was limiting overall performance and thus reducing the loading on the core regulator supply. With my card, the NGO_8456 typically caused a drop of around 150 points in both 3dmark05 and 06 scores!

Also, raising the value of the overcurrent OCSET resistor can help in a small way. However it also is not a cure, since completely removing the resistor does not stop card lockups.

The real "cure" has been determined to be the cutting and rerouting of certain pcb tracks. Unfortunately, because of the pcb's complex multilayered design, that particular option is not currently considered the way to go. However, all is not lost, since simply connecting an easy to fit wire link will more or less achieve all that is necessary for stability (see red wire link at attachment below). Although by doing this, and not cutting tracks, it has to be accepted that the built-in "droop" and "current sense" functions will not operate correctly ---> however considered a small price to pay for achieving stable core volts and better overclocks.

With my G70 card modded as above, there are no more lockups. Currently at 540/1600, typically achieving 7650+ 3dmark05 scores; and 4050+ 3dmark06 scores. And an excellent knock-on effect of the more stable core volts supply has been an ability to more easily achieve 1600+ memory overclocks. None of which was previously possible!

And I expect to achieve more. However, because of still using card's original heatsink, must be carefull and thoroughly test so as not to overtemp anything !

[homoculus]

Jimmor, what's the theory behind that wire trick? How did you determine that it would help with stability? What was your previous maximum stable overclock? Sorry about all the questions, but I've hit the wall with mine. I do however have an aircooling solution for you (all) which may help you achieve more:

Remove all stock cooling from the board together with the alloy PCB reinforcing strip. Clean thoroughly with Isopropanol.

For the HSI chip you need a Zalman ZM-VHS1 6600 / 6600GT AGP graphics card heatsink and the adjustable arms from a Zalman ZM-NB47J to fit the wider mounting holes on the 7800GS, use AS5.

Fit Thermalright V1 Ultra with AS5 as per instructions, it fits perfectly.

I used Zalman RAM sinks because you get four low profile versions for the topside of the board where I had issues with fouling my DIMM slots, I kept the included Thermalright RAM sinks for fitting to all my MOSFETS and voltage regulators (they have the perfect footprint).

Finally I fitted an additional 80mm fan to the V1 Ultra and ran it on 5 volts for peace and quiet.

However, because the PCB is carrying so much weight you need to attach a zip tie to the unsupported corner and attach it to the chassis somewhere otherwise the card bows dangerously.

The result is that the cold bug has now reared it's ugly head, but full load temperatures are down to 55c

This has helped me hit 500 on the core where previously I could only get 473. I know my board is still suffering badly from insufficient voltage but I have a modded 7800GT (G70) bios to try out with a 20 Delta and some other tweaks first.

[jimmor]

Jimmor, what's the theory behind that wire trick? How did you determine that it would help with stability? What was your previous maximum stable overclock? Sorry about all the questions, but I've hit the wall with mine. I do however have an aircooling solution for you (all) which may help you achieve more:

Remove all stock cooling from the board together with the alloy PCB reinforcing strip. Clean thoroughly with Isopropanol.

For the HSI chip you need a Zalman ZM-VHS1 6600 / 6600GT AGP graphics card heatsink and the adjustable arms from a Zalman ZM-NB47J to fit the wider mounting holes on the 7800GS, use AS5.

Fit Thermalright V1 Ultra with AS5 as per instructions, it fits perfectly.

I used Zalman RAM sinks because you get four low profile versions for the topside of the board where I had issues with fouling my DIMM slots, I kept the included Thermalright RAM sinks for fitting to all my MOSFETS and voltage regulators (they have the perfect footprint).

Finally I fitted an additional 80mm fan to the V1 Ultra and ran it on 5 volts for peace and quiet.

However, because the PCB is carrying so much weight you need to attach a zip tie to the unsupported corner and attach it to the chassis somewhere otherwise the card bows dangerously.

The result is that the cold bug has now reared it's ugly head, but full load temperatures are down to 55c

This has helped me hit 500 on the core where previously I could only get 473. I know my board is still suffering badly from insufficient voltage but I have a modded 7800GT (G70) bios to try out with a 20 Delta and some other tweaks first.

It's a long story !

The need to fit the red marked wire link was because my xfx7800gsee suffered from "black screen" lockups when voltmodded and overclocked. The problem appeared to be core loading related, since it happened much more regularly during the higher overclocks achieved when core volts was modded above 1.4v.

Through much testing, determined that "black screen" lockups were directly related to a combination of Driver, Core_volts and Overclocks levels. Or to put it another way, almost guaranteed to happen if achieving over 7500 in 3dmark05. This however is for my particular setup, so would expect other computer setups to have different lockup threshhold !

From reading around, found "black screen" lockups are quite common to 7800gs usage. In some cases the problem appeared to be cured by using NGO drivers. There are even posts at NGO forum proclaimng NGO_8598 as the overclocked 7800gs's saviour?

However, all I ever identified about NGO drivers was that their tweaked configuration was reducing overall performance and thus reducing core loading. So for me, the NGO driver was only allowing more stable operation, or higher overclocking (570+ levels), because with those drivers it was impossible to breach the apparent "black screen" threshhold of 7500 with 3dmark05?

Finally tracked my particular "black screen" problem down to "faulty pcb manufacture" involving the core volts regulator circuitry. Because of what appears to be misconnected reference/feedback pcb connections to the ISL6568, core volts is much noisier, ie, jumping around more than it should. For normal card use, this shouldn't really be a problem. However, when volt modded and overclocked, and thus forcing GPU loading to breach a specific and critical level, instability and black screen lockups is what happens.

The "red wire link" basically changes how the ref/feedback lines relate to the ISL6568 in a way that makes the core volts output more stable under heavy loading. This however was just an acceptable compromise solution, since a proper fix would have involved cutting and rerouting of pcb tracks.

I no longer get black screen lockup problems, and can also easily achieve 7750+ scores in 3dmark05; depending only on set CPU/GPU volt/overclock levels. So just as it should be, my overclock limits are now found by testing against artifacts. Unfortunately however I am now at a point where the true potential of my current 7800gs setup is being somewhat limited by my P4 CPU!

Mind you, having overcome my stability/lockup issues, achieving stable 570+ with a G70 based 7800gs typically requires 1.6+ core volts levels. And then of course controlling temps becomes the next important issue? So for now, whilst still using original heatsinking, I am currently satisfied to limit my overclocking to around 550.

Unfortunately the 7800 series chip was never a good design. It's promoted improved performance capabilities can therefore be simply put down to it having more pipes than the previous generation. However since the standard 7800gs is supplied with pipes fixed at 16x1,6vp, it could be argued that it is nothing more than an slightly higher clocked 6800gt/u. Ridiculous when you consider the very high for what you get cost of a 7800gs. So not really a good upgrade choice for anyone already owning a 6800gt/u. Unless of course you are lucky enough to get a G71 (7900) based 7800gs, since it can apparently be overclocked to very extreme levels; 700 being typical.

On the other hand; ATI are currently producing very good price/performance AGP cards. So maybe now (or christmas) is a good time for people with older AGP setups to consider putting their hard earned cash into manufacturer's that supply better value?




EDIT,

To remove "cold bug" problem, mod your bios with nibitor v3.1 ---> tick "disable temp monitoring" at the "Temperatures" tab !

Also, you are likely to achieve higher overclocks with delta's greater than 30. Although maybe more important to using the "delta" facility, a 7800gs chip doesn't generally like the ROP/Shader domain speeds to be higher than 513MHz.

[homoculus]

Jimmor, I too have the XFX 7800GS EE, and it's junk, but I blew nearly £200 on it and the missus won't countenance an upgrade until next year at the earliest.

I can get the core over 500, but the memory has to be slowed or vice versa. I believe the 91.47 Nvidia drivers are superior to the 93.71 and don't rate the NGO's at all. Because the core clock operates in multiples of 27 MHz I have settled on 486 core, but I believe the ROP and shader domains operate with half steps, so a Delta of 14 might be better. All the evidence I've read suggests that G70 dies tie the core clocks quite closely to the ROP/Shader clocks whereas G71 can use Deltas of 40 or more. I think Vdimm is my biggest problem, can you suggest a working mod for this? I have the same samsung chips as the 7800GTX but I simply cannot get them above 1400 even for benching.

Mind you AM3 = 103,368 and 3dm06 = 4071

Thank you for the cold-bug tip, I assumed that this option was only related to overheating, there must be a bug in the firmware.

[jimmor]

Jimmor, I too have the XFX 7800GS EE, and it's junk, but I blew nearly £200 on it and the missus won't countenance an upgrade until next year at the earliest.

I can get the core over 500, but the memory has to be slowed or vice versa. I believe the 91.47 Nvidia drivers are superior to the 93.71 and don't rate the NGO's at all. Because the core clock operates in multiples of 27 MHz I have settled on 486 core, but I believe the ROP and shader domains operate with half steps, so a Delta of 14 might be better. All the evidence I've read suggests that G70 dies tie the core clocks quite closely to the ROP/Shader clocks whereas G71 can use Deltas of 40 or more. I think Vdimm is my biggest problem, can you suggest a working mod for this? I have the same samsung chips as the 7800GTX but I simply cannot get them above 1400 even for benching.

Mind you AM3 = 103,368 and 3dm06 = 4071

Thank you for the cold-bug tip, I assumed that this option was only related to overheating, there must be a bug in the firmware.

The G70 doesn't really care what delta is used. Have tried everywhere from 10 to 100 with no problems. Other than of course an expected hit on performance once the delta is set too high. The performance "sweet spot" for any card is just a compromise between "max 3D speed" and "delta", so only way to determine anybody's best performance settings is by a bit of trial and error with those two variables?

But maybe more important is that ROP/Shader speeds don't appear to like being higher than 520-530? Mind you, for my situation this could be a temp issue. And since still using stock heatsinking, I tend to keep ROP/Shader at no higher than the 27MHz jump of 513MHz.

However, before "red wire link" mod, card wouldn't run artifact free with ROP/Shaders above 500. Which, using any delta value, meant ROP/Shaders had to be fixed no higher than 486MHz jump point. Mind you, even with ROP/Shaders at 486, and "black screen lockups" permitting, had no problems running with delta of 60ish and thus core geometric domain speeds of 560-570.

And just in case you are interested, I normally run with 3D core volts at 1.55v. Which means my bios's 3D volts is set at 1.4v, plus VID1 and VID2 on the pcb are grounded!

To increase memory volts and thus overclocked speed, the VR-Zone VDD pencil mod (http://sg.vr-zone.com/index.php?i=3317)works as good as any. Have read posts of modding to around 2.3+ volts levels. However, suggest you keep in mind that someone has to be the first to kill their card by overvolting memory too far---> so don't let it be you?

Also, soldering a 3.9k resistor over the one indicated for VR-Zone VDD pencil mod, should set memory at 2.05v. Which for my card allows overclocking to 1700+.

And yes, the "cold bug" issue has to be a hardware/firmware problem. but also obviously something that neither Nvidia, nor their partners, are interested in fixing ?

[jimmor]

With current Core 2 E6600 setup, my modded XFX7800gsee, running 1.55v and 545/1600 typically achieves the following:-

3dMark05 = 8520
3dMark06 = 4644
AQuamark3 = 119050

And at least to me, the "red wire link" mod has proven itself because regardless of which driver is used or max overclocks achieved, "black screen lockups" have become nothing more than a bad experience of the past!