512mb 7900GT vmod development thread

[s e t h]

here's the 7900GT 512mb versus the 256mb at exactly the same clocks. 450/330
512 above and 256 below.
scores virtually identical

512


256

[graham_h]

humeyboy can you post them please ?

[humeyboy]

Heh, I just did pencil mod to core and got 1.4v and ATI TOOL passed max core as 640mhz, problem is the 1.4v increase is in both 2D and 3D so its 24/7
Actually 1st time Ive ever heard of pencil mod for Vcore.

Yes I wll post pics but who the hell lets me upload high res pics, both are 1.7MB-3MB, rear being smaller res, as I needed more detail for reading core and ram text. (imageshack wont accept then, I had up smaller ones).

[graham_h]

so you got 1.4v when you should have got 1.3v ?
Thats was measured with a multimeter right ?

Core should go into the 670's at that voltage. Mine tops out at 645mhz on the stock 1.2v

Damn good effort.

You did the pencil mod to the spot marked in the VR zone threead ?

[humeyboy]

My card is 1.2v stock, I penciled and it read 1.8v lol, I redid and its 1.41V on a new calibrated meter

There is issue with cooler on agp cards, well Im 99% sure the backing paper on rampads is wrong and should have been removed, this causes core to be bad contact eap at side nearest memory and had to fill with extra thick artic silver 5, if paper is removed it will be less a gap (awaiting reply by e-mail) none my other NV Silencer rev3 told me to leave paper on or the 20pipe version I had before of this card, I tihnk Gainward have made mistake.

So if this is the case, I get new pads and stick less artic silver 5 needed for better core cooling. BTW the 640mhz will be higher in other apps like 3dmark05/06, cause stock I ran 600mhz core in them but only 550 in ATI TOOL, Im talking not 1 blip on screen not the ones it ignores.

Yes from VR site but that guy Sham is helping is kinds dimm, I was worried he gonna cause me to blwo up card, as he changed it a lot from post to post.

OK how and where can I up high res pics, do you actually need them ? its AGP remember.

EDIT: NVM, I seen your im/pm and sent you and other peep copy of pcis rared and viri checked, hope your ISP accepts 4.5MB emails, mine will.

[graham_h]

sure, PCB is the same you said ?

[humeyboy]

Yes look at pics of mine and one on previous page, every part exept where SLI notch is are same, all mod area chips are same, the card I have is a PCI-E card but fitted with HSI Bridge same as all 6800Ultras were AGP and later fitted with HSI Bridge for PCI-E use.

Your mail is sent, both of you that asked in im/pm.

[IMG][/IMG]




As big as site would let me put up few days ago.

[graham_h]

ok they're up (don't click if you don't want a Hi-Res image)

http://www.tigerdirect.com.au/images...0GTAGPback.jpg

http://www.tigerdirect.com.au/images...GTAGPfront.jpg

[humeyboy]

Cheers for the hosting of my Hi-Res pics

LOL, God help 56k users loading up that front pic. :P

[alpha0ne]

Looks like some better metal caps used in the AGP version (6 used), the PCI-E version only has 2 metal caps used, the rest are el cheeeepo resin

[persivore]

Can someone check the voltage at the points I've marked on the attachment please.

[s e t h]

here we go

[humeyboy]

There wont be 2.11V to ram though cause it would do rated 800/1600DDR if it had 2.0v as its BJ modules, we need know how much volts ram is really getting, then give it 2.1v , should be safe to at least 2.3v and do 900/1800DDR.

[persivore]

From those readings, it looks like the LM358 on the card is controlling a voltage at 1.24v (I don't know what it is tho). It looks like the memory is being supplied from the RT9214 and not the LM358 and the origional Vmem mod was correct . Vmem looks like its probably at 2.11v.

The 7900 series of cards doesn't seem to show much (if any) improvement from increasing the Vmem, which could be why no overclock increase was seen when increasing the memory voltage.

There is alot more to memory clocks than just the voltage. The clocks can be greatly effected by the layout of the memory traces on the card, as well as the timings. I would suspect that the memory timings have been tightened a bit, which might be why the memory doesn't seem to overclock very well.

[s e t h]

well they're hot enough to be getting 2.11 imo.
i've tried two card now and the second one has memory that does 770 if cooled actively and 760 if not cooled besides for the stock 'sinks.

i'm starting to think they've starved these cards for volts somewhere else like igpu or vddq or something
i mean there are like 10 separate points that read 2.11v on the back of the card.
that pretty much has to be the vdd imo.
anyway i'm clueless so my opinion counts about squat at this point

btw persivore in case you missed it i updated the pic above with the codes on the mosfets which are slightly different than on humeyboy's agp card.

edit: actually the memory doesn't really appear to be tighter as this 512mb card score the same in '06 when matched clock for clock with the 256mb cards.
512mb vs 256mb benchies in post#26 above

[humeyboy]

Hmm, well I can get timmings using NibiTor, this ram should hit 800/1600 with 2v anyhow never mind bench even if timmings tighter.
I doubt my agp is 2.1v now on ram, I only need exact part to measure not some chip thats putting out that volts, man you can get 12V on the card if you measure it in right spot.
They didnt put on 1.2ns for nothing, the NON+ 20pipe version I had before had cheapo Infin 1.4ns ram, it did 1475 unmodded, unmodded this aint any better so I assume its lack of volts.

I did however get increase in ram OC (aint reached limit yet stopped at1520 up from 1475) after doing Vcore mod with pencil.

[persivore]

Hmm, well I can get timmings using NibiTor, this ram should hit 800/1600 with 2v anyhow never mind bench even if timmings tighter.
I doubt my agp is 2.1v now on ram, I only need exaclt part to measure not some chip thats putting out that volts, man you can get 12V on the card if you measure it in right spot.
They didnt put on 1.2ns for nothing, the NON+ 20pipe version I had before had cheapo Infin 1.4ns ram, it did 1475 unmodded, unmodded this aint any better so I assume its lack of volts.

I did however get increase in ram OC (aint reached limit yet stopped at1520 up from 1475) after doing Vcore mod with pencil.

Timings will make a lot of difference to the memory clocks. Try clocking some TCCD on 2,2,2,x . GDDR should behave similarly.

There is also double the memory on these cards, which will should decrease overclocks. The 7900GTX's don't usually respond very well to increased voltage either (increasing voltage by 0.1v on some cards can cause artifacts at stock clocks).

The points found on the front and back of the card which read Vdd are almost definitely the correct points. They are the ouput of the voltage regulator (and associated de-coupling), and will read within 0.02v of the actual voltage getting to the memory.
Unless you are willing to de-solder a memory chip from a card to check the voltage that is being recieved at its supply pins, there is no way to varify 100% that you are reading the correct voltage. There will probably be a few de-coupling capacitors on the back of the card where the memory chips are which will read Vdd, but its not going to be 100% certain that the voltage is actually being used for the RAM's voltage supply.

[humeyboy]

HMM, I got max of 1450mhz in ATI TOOL with no aritifacts on this card and after reseating cooler I got 1475, now after 1.4v VCore mod I get 1550, I bet if Vcore was 1.5v or even 1.55v the ram would do 1600 or more

Yes I know timmings will change max OC but I meant to even set to 800/1600 never mind bench (ram should do it).

I will grab timmings and then try find out other cards timmings but not sure whats to comare too, supose I try find out if this ram has a default timming setting.

[s e t h]

comparing the timings on a 256mb card would be a good start point imo.
that ram does 900mhz in most cases with less volts even though it's rated slower.

persivore - what do you think about changing the 1.24 and/or the 1.74 upwards about 5% to start.
if iether makes a difference we'll be getting somewhere.
you got any ideas how i should go about this besides trying RT9214 again.

regarding the LM358 - are my guesses accurate in that pin2 and 6 via a VR to ground will vary output A and B respectively?

[persivore]

The specified timings for 800MHz operation are on page 56 of the datasheet.

comparing the timings on a 256mb card would be a good start point imo.
that ram does 900mhz in most cases with less volts even though it's rated slower.

persivore - what do you think about changing the 1.24 and/or the 1.74 upwards about 5% to start.
if iether makes a difference we'll be getting somewhere.
you got any ideas how i should go about this besides trying RT9214 again.

regarding the LM358 - are my guesses accurate in that pin2 and 6 via a VR to ground will vary output A and B respectively?

I've got no idea what the 1.24v voltage is used for, but it might be worthwhile increasing it.
The 1.74v voltage looks like its the PWM signal for the Vdd supply before it is fed into the LC filter to produce the final Vdd supply.

From the picture, it looks like pins 5, 6 and 7 on the LM358 are unconnected, but the connections could go under the chip to the opposite side of the card.
Can you check the voltage on the other pin on the MOSFET which reads 1.24v, and check the voltage at pins 1 and 7 on the LM358. Check the voltages with the card under a constant load (ie sat at the desktop)

[humeyboy]

NibiTor cant read timmings on this card yet.

[s e t h]

3.68 on the other mosfet leg on the one with 1.24.
i did lots of measuring of LM358. results in post#21 above
and yes there are connections under LM358.
for instance pin5 is connected to pin3 and the smd below
pin6 is connected to the closest lower smd's to the left of LM385
pin7 reads 5.5volts when system is on

[persivore]

3.68 on the other mosfet leg on the one with 1.24.
i did lots of measuring of LM358. results in post#21 above
and yes there are connections under LM358.
for instance pin5 is connected to pin3 and the smd below
pin6 is connected to the closest lower smd's to the left of LM385
pin7 reads 5.5volts when system is on

Can you check if the 1.24v voltage has anything to do with Vgpu?

[s e t h]

*whoooossssshhh* (/sound of that^ going right over my head )
scuze the dumb question but how would you suggest i go about that? :P
you want me to mess with LM358 till 1.24 moves and see if it affects vgpu?
sorry i didn't understand.

just in case with the card off the 1.24v mosfet leg reads 131ohm to ground or the vgpu read point in post#2. if i reverse the leads the reading is 214ohms to ground and 216ohm to vgpu read???

edit: once again i really appreciate your help.
i know what it feels like talking to brick walls

[persivore]

*whoooossssshhh* (/sound of that^ going right over my head )
scuze the dumb question but how would you suggest i go about that? :P
you want me to mess with LM358 till 1.24 moves and see if it affects vgpu?
sorry i didn't understand.

just in case with the card off the 1.24v mosfet leg reads 131ohm to ground or the vgpu read point in post#2. if i reverse the leads the reading is 214ohms to ground and 216ohm to vgpu read???

edit: once again i really appreciate your help.
i know what it feels like talking to brick walls

Just check if Vgpu and the 1.24v voltage read the same voltage, and check if the 1.24v voltage raises as you raise Vgpu

1.24v just looked very close to what I would expect Vgpu to be at on a stock card