7800 GTX 512 BIOS Mod

[HeavyH20]

A little revisit to the modded BIOS.

What does not seem to help any:
tighter timings on the memory
any type of delta
any voltage table work - needs a physical VMOD

What does help:
loosen memory from BIOS settings to Samsung spec - allows memory to exceed 1800 easily

That's about it. But, you will not see much change unless you get the core to aroun 620 or so, since the memory easily exceeds anything the core throws at it.

The rev3 BIOS is for people who want to improve scores but need to keep the core a little lower since their card does not OC well.

[Roliath]

I've been waiting for this!!
thanks!!

[Dynasty]

Basically a small change to the voltage table and a slight tightening of the memory timings.

What were default memory timings in this bios before the edit/tweaked timings.

[eva2000]

how small a change in voltage table ?

[HeavyH20]

The timings before and after. There are two tables to look at. Table 3 and 7.



As for the voltage, there is a third undefined value that is supposedly 1.5V. I have not had a chance to verify with a multimeter, but, the core clocked a little higher with it selected versus not selected. This is the same common 7800 GT BIOS mod for 1.5 V. I now get zero artifacts at 625 where I saw a little previously. I was also able to coolbits pass 620 where 605 was the limit prior.



After the hex edits, here is what NiBiToR sees:

[eva2000]

ah interesting love your work

[HeavyH20]

Well, have to try some change to pass up your x1800XT results

[VoRtAn_MaDgE]

All brands of 7800 gtx 512mb use the same settings ???
I'm assuming so...

Wich manufacter is that bios u modded HeavyH20 ???

I will save my xfx bios right now to compare

[VoRtAn_MaDgE]

HeavyH20 i have some question to you...

When you say table 3 and 7 you're refering to timmingset 3 and timmingset 7 am i correct ?

If so that's my default values of those 2 tables



Am i missing something...cuzz none of these my two tables have same values of your first table (for what i understood is one "default table values" of yours)...

Any explination would be great

[ViperJohn]

What did not work was any sort of delta. If they had built the 256 MB card with better memory, the timings would have aligned better and the delta would not have been needed. So, the 512 card appears to be very well matched as far as core speeds and memory bandwidth.

Must have been your particular card and/or cooling and/or Vcore run. Both of Tario's 512Mb cards responded well to GDC adders up to 54Mhz. What his cards didn't respond to was running Vdd higher than stock with one card taking a large memory OC drop with increased Vdd.

http://service.futuremark.com/compare?3dm05=1517679 - Single Card

Viper

[HeavyH20]

Interesting. Maybe the VCore mod has to occur for the delta to add extra performance. I did find the delta BIOS on the 256 MB card to be important. I will try some 50 deltas again and see what happens.

[ViperJohn]

Interesting. Maybe the VCore mod has to occur for the delta to add extra performance. I did find the delta BIOS on the 256 MB card to be important. I will try some 50 deltas again and see what happens.

I was able to run the GDC up to 54Mhz one on of Tario's card stone stock. Only 27Mhz on his other card stone stock.

Modded the bios images were both programmed for an HP3D ROP clock of 648Mhz with GDC's of 40Mhz and 27Mhz respectively. Tario's Card #1 is decidedly stronger than card #2. Card #1 will run all the way up to 702ROP / 742G/SE x 1008 (2016ddr) in 3DM05 slightly dirty which was good for almost 12,100 with my lowly 2900Mhz Opteron.

Every card has to be checked though as the max Raster Operations and Geometry/Shader Engines clocks a given core can run clean varies all over the map. The Delta between the ROP and G/SE clocks can vary markedly from stock to modded on the same card too.

Viper

[HeavyH20]

I noticed some weird instability with the voltage table change after 8 hours hours of solid 3Dmark05 looping. So, I am pulling that change from the rev2-580.rom until I see if it is a "good" change to keep. I also noticed some very small artifacting from the memory. I ran the cards purposely hot to determine what they are like on air to make sure the modication is safe for the stock heatsink.

NVIDIA already tightened up the memory a bit, and I tightened it a little more. Some cards may not handle much past 1750 with these timings. In light of that, I have created a BIOS that is stock 580/1750 (the BFG BIOS) with the Samsung spec timings. They are a little looser than the stock spec, so you may be able to get your memory higher if you are having trouble reaching the 1750 spec or higher. rev1-580.rom has this single modification from stock. rev2-580.rom still has the tighter memory timings if you are fortunate enough to have a card with very good memory.

[ViperJohn]

I noticed some weird instability with the voltage table change after 8 hours hours of solid 3Dmark05 looping.

Heavy were you using a fanless GPU cooling solution on that card???

Viper

[nachthymnen]

It's possible to use nvflash for flashing this bios?
The command it's:
nvflash -5 -6 biosname.rom
Tks

[ViperJohn]

It's possible to use nvflash for flashing this bios?
The command it's:
nvflash -5 -6 biosname.rom
Tks

Yes just use the lastest revision of NVflash.

Viper

[nachthymnen]

Yes just use the lastest revision of NVflash.

Viper

Tks Viper
And it's better the bios rev.1 right?
I will try tomorrow
Another info...setting 110 on PCI EXP. can give a good improvement in the results of 3dmark'05?
Sorry for my English

[HeavyH20]

Heavy were you using a fanless GPU cooling solution on that card???

Viper

Yes, I was using a couple of Innovatek water blocks. No issues with the any of the other settings. Have you seen different?

Another info...setting 110 on PCI EXP. can give a good improvement in the results of 3dmark'05?

I found that 107 on the PCI-E bus was a bit of a sweet spot. I have also had it as high as 115.

[nachthymnen]

Yes, I was using a couple of Innovatek water blocks. No issues with the any of the other settings. Have you seen different?




I found that 107 on the PCI-E bus was a bit of a sweet spot. I have also had it as high as 115.

Tks Heavy...I will try in the next days
Open windows and I hope the outside temp reach the -10°C

[ViperJohn]

Yes, I was using a couple of Innovatek water blocks. No issues with the any of the other settings. Have you seen different?

If you still had the stock Vcore mosfet heatsink and didn't fan cool it when you went to a fanless GPU cooling solution that was probably why your Vcore went unstable. You really do need to define "unstable" though the Vcore regulation is fairly loose to begin with and tends to bounce around a bit with load changes.

When the weak stock Vcore mosfer heatsink looses it's air flow the whole PS area runs hotter than hell. It is not uncommon for 7800's with fanless GPU coolers to puke Vcore PS parts over time from excess heat, expecially when volt modded. One of the first things you will see from high heat in the PS area is the already loose Vcore regulation can get fairly sloppy.

Viper

[HeavyH20]

Ah, that's what you meant. I have an 80 mm fan blowing on the VREG heatsink on both cards. That never gets over 35C or so. The whole card stays nice and cool.

As for a prior question, the default timings on set 3 are already tightened over the stock Samsung spec for the memory. So, NVIDIA has already tweaked the memory a little. That is why stock spec is 1700 versus 1800. Most, if not all, cards use timing set 3. The rev1 BIOS loosens the timings to Samsung stock spec (like in timing set 7) for those that want some extra clocking on the memory. 1900 is possible with no volt mod. The rev2 BIOS tightens things slightly, but if your memory is not up to it, the 1750 clock and the tighter timings may not like each other. Most cards should be able to take the timing bump.

[nachthymnen]

Just a question...also the 512MB have the cold bug?
Yesterday with cold room temp, the frequencies of my XFX freeze @ 501/1728(from rivatuner hardware monitoring)
There's a bios that resolve this problem?
It's a problem setting with nibitor 1.5V only when I bench?

[mike]

Is the timingset #6 for 2D?

[[XC] moddolicous]

If its any help in this thread, a lower trp and a higher tras is better than a lower tras and a higher trp (ex tras=8 & trp=2 is better than tras=6 & trp 4) and running them had the same stability (tested in SPI 1.4mod on regular ram).

[giorgioprimo]

signed