Tutorial: How to softmod to ANY VGPU you want without hardmods! (1.3~1.6v or more)

Printable View

Show 100 post(s) from this thread on one page

01-31-2009, 08:28 PM
STEvil

Quote:

Originally Posted by Falkentyne

I kept editing my post several times through reboots and testing (different computer).

Yes the monitoring worked fine, as long as I didnt run AMD GPU tool. as soon I ran it, (without even changing settings) the sensors 'vanished' or something or if already active, froze. Using ATT to overclock was fine.

The current went up with the vid (at idle) and apparently alot more at load.

anyhow, same result..can't go any higher in firefly forest, regardless of the vcore/current. My max is 830 mhz regardless of voltage. (835 mhz works for about 3-4 runs then lockup, etc).

AMD GPU Tool does weird things.. it shuts down CCC and breaks fan speed control on 4870X2 cards..
02-01-2009, 03:00 AM
celemine1Gig

Quote:

Originally Posted by fornowagain

Any idea what going on here then?

I know for sure is the vcore measured from the back of the card never exceeds 1.35v volts no matter what value I use. This is a trace from 'in game' loads compared to resting loads at different voltages. It doesn't pull any more current in 3d 1.52v than 3d 1.35v. And were it moves to 1.52v in resting mode the current is well under even the 1.26v stock load current. So no evidence OCP. And OVP should always be above the set VID.

http://img218.imageshack.us/img218/1585/55424494zm5.jpg

1.06v (2d register) to 1.55v gives ~1.35v. Upto 1.35v its very close to the DMM values in 0.0125v steps. But something is limiting it, voltage increase is linear till it hits the wall.

http://img292.imageshack.us/img292/533/87865169yv3.jpg

OK, now, besides the fact that I don't really understand why you would really want to run your card at 1.55V, here's what I'm thinking.

You are simply hitting the power circuits physical limits. ;) If the monitored data is correct, what you see is that the current actually decreases while the voltage is set to 1.55V, in comparison to 1.35V. And at the time the voltage falls back to 1.35V the current increases again.
It would seem as if the circuit just can't deliever any more current than what is available at 1.35V. That's why the current decreases when the voltage rises. You hit a point where the power output is limited to a certain amount. It's like Intel's VDroop function. Once the current draw gets too high, you either have to drop the voltage to deliever the current, or drop the current to get the voltage. Both won't happen, as it's a physical limit. ;)
Perhaps this could be modified in some way, but I doubt that it would be very healthy for the card.

Oh and BTW, what is the actual DMM reading of the voltage, when VID is set to 1.55V? I didn't fully understand that just from reading your post. It's a bit confusing. But maybe that's just because English is not my native language. In one passage it reads like it was 1.35V and then you talk about 1.52V. What is the correct value? :shrug:
02-01-2009, 03:59 AM
benko

i have 4870 1gb giagbyte, tried this on cmd but nothing, everthing is not recognized, maybe this is because 64bit vista sp1??

i run cmd as adminitrator.
02-01-2009, 05:43 AM
benko

got it on 4870!!!!

yepp i did it(i hope so) in riva tuner i got new voltages but havent tested yet.

here is pic, how can i fix this that gpu dont change voltage every second(it went from 2d voltages to 3d voltages every second) i have chanhged both 2d and 3d voltage card has also before doing this chnged this voltges every secon(before was 1.2-1.26v)

http://img518.imageshack.us/img518/9...odedgpulo5.jpg

how can i get it in 2d that it is always 1.1v??
02-01-2009, 05:50 AM
fornowagain

Quote:

Originally Posted by celemine1Gig

OK, now, besides the fact that I don't really understand why you would really want to run your card at 1.55V, here's what I'm thinking.

I don't. I want to know why 1.35v is the maximum. A hardmod would take voltages higher, this method should as well.

Quote:

Originally Posted by celemine1Gig

You are simply hitting the power circuits physical limits. ;) If the monitored data is correct, what you see is that the current actually decreases while the voltage is set to 1.55V, in comparison to 1.35V. And at the time the voltage falls back to 1.35V the current increases again.
It would seem as if the circuit just can't deliever any more current than what is available at 1.35V. That's why the current decreases when the voltage rises. You hit a point where the power output is limited to a certain amount. It's like Intel's VDroop function. Once the current draw gets too high, you either have to drop the voltage to deliever the current, or drop the current to get the voltage. Both won't happen, as it's a physical limit. ;)
Perhaps this could be modified in some way, but I doubt that it would be very healthy for the card.

Its not hitting any limits, OCP causes instant shutdowns. There are two load states. One is as the game is running pulling the most load and the other as it idles at the desktop. If there was any current limit actuating at the higher VID-1.52v, whilst at idle there is still more current headroom. When switched to heavy 'in game' loads at the higher VID-1.52v then it could reach higher currents and crash, it doesn't. In fact the heavy load current is the same as 1.35v. Indication that 1.35v is the maximum voltage.

Quote:

Originally Posted by celemine1Gig

Oh and BTW, what is the actual DMM reading of the voltage, when VID is set to 1.55V? I didn't fully understand that just from reading your post. It's a bit confusing. But maybe that's just because English is not my native language. In one passage it reads like it was 1.35V and then you talk about 1.52V. What is the correct value? :shrug:

Its in the picture as well, the actual voltage never exceeds 1.35v regardless of the VID set.

Another, you see a little variance in load depending on what the game is doing. But loads currents 1.55v=1.35v. Finally default load current @1.26v is a lot lower.

http://img129.imageshack.us/img129/921/43730291lt6.jpg
02-01-2009, 06:25 AM
kemo

Gigabyte HD4850 here , no Voltage reading nothing at all :D:
02-01-2009, 06:45 AM
Hardass

Running GTX 260 I can change voltage using this, but for some reason in RT Monitor my voltage regulator output never changes even when going to full load. Current increases but not output.???
02-01-2009, 07:20 AM
celemine1Gig

Quote:

Originally Posted by fornowagain

I don't. I want to know why 1.35v is the maximum. A hardmod would take voltages higher, this method should as well.

Its not hitting any limits, OCP causes instant shutdowns. There are two load states. One is as the game is running pulling the most load and the other as it idles at the desktop. If there was any current limit actuating at the higher VID-1.52v, whilst at idle there is still more current headroom. When switched to heavy 'in game' loads at the higher VID-1.52v then it could reach higher currents and crash, it doesn't. In fact the heavy load current is the same as 1.35v. Indication that 1.35v is the maximum voltage.
Its in the picture as well, the actual voltage never exceeds 1.35v regardless of the VID set.

Another, you see a little variance in load depending on what the game is doing. But loads currents 1.55v=1.35v. Finally default load current @1.26v is a lot lower.

http://img129.imageshack.us/img129/921/43730291lt6.jpg

I think you didn't quite understand what I meant to say.

There are circuits that just keep on supplying higher currents to drive higher voltages, until they reach the OCP and/or OVP trip point and shut down to prevent damage.
And there are circuits that have another way to deal with this problem.
They will only supply a certain amount of current. To get a higher voltage you will need the right amount of current to drive the voltage (higher voltage means higher current). In your case it seems that at about 1.35V you have reached the circuits' current limit. Now you can set any higher voltage via VID. There'll never be enough current to drive the higher voltage, i.e. it'll always remain at ~1.35V. Of course the current draw at 1.26V is lower (see above).
BUT, I'm not quite sure if this is the case with your card, because normally in idle state the higher voltage should show. It would just drop down in load state. Maybe it's because the card switches to a lower VID in idle anyway, or maybe on your card there is a different implementation that doesn't even allow VIDs higher than 1.35V. I have no datasheets here so I can't tell you for sure. You need to ask the poeple that have the datasheets. But unfortunately these datasheets seem to be under NDA anyway.

You could test and set all register value for all states to 1.55V (0x58h = 88). Then in idle you should theoretically get the higher voltage.

Besides if and how OCP gets implemented is a question that the developers need to answer. It's not always the same and you can't say that there will always be a OCP limit that will cause the controller to shut down.
02-01-2009, 07:30 AM
Jamesrt2004

4870 512, works here @ 1.3v going to try 1.35

nice got to 860 on the core :)
going for 870 now :)
02-01-2009, 07:37 AM
fornowagain

Quote:

Originally Posted by celemine1Gig

I think you didn't quite understand what I meant to say.

There are circuits that just keep on supplying higher currents to drive higher voltages, until they reach the OCP and/or OVP trip point and shut down to prevent damage.
And there are circuits that have another way to deal with this problem.
They will only supply a certain amount of current. To get a higher voltage you will need the right amount of current to drive the voltage (higher voltage means higher current). In your case it seems that at about 1.35V you have reached the circuits' current limit. Now you can set any higher voltage via VID. There'll never be enough current to drive the higher voltage, i.e. it'll always remain at ~1.35V. Of course the current draw at 1.26V is lower (see above).
BUT, I'm not quite sure if this is the case with your card, because normally in idle state the higher voltage should show. It would just drop down in load state. Maybe it's because the card switches to a lower VID in idle anyway, or maybe on your card there is a different implementation that doesn't even allow VIDs higher than 1.35V. I have no datasheets here so I can't tell you for sure. You need to ask the poeple that have the datasheets. But unfortunately these datasheets seem to be under NDA anyway.

You could test and set all register value for all states to 1.55V (0x58h = 88). Then in idle you should theoretically get the higher voltage.

Besides if and how OCP gets implemented is a question that the developers need to answer. It's not always the same and you can't say that there will always be a OCP limit that will cause the controller to shut down.

I understand the phased regulation fairly well, I have volt modded many cards and seen OCP in action. Its not changing VIDs and the voltage is the target value which holds solid at 1.346v. Its a regulated value and quite obvious when you observe it in real time via DMM. Voltage increases in discrete 0.0125v steps until 1.346v.
Again, the reason I included idle load current is to show that OCP is not occurring. Also the OCP/OVP register 13 is not changing state. And yes indeed you can activate the 2d VID to target values via register and that is also limited to 1.346v regardless of current (see previous posts). Unwinder (RT designer) has already stated the much higher voltages are supported with the 1165. OVP shouldn't factor here as its nominal VID + 200mV offset.

There is a definite lock here from 0x48h to 0x49h.
02-01-2009, 07:50 AM
celemine1Gig

Set all 4 register entries to 58 and see, as I said.
02-01-2009, 07:57 AM
fornowagain

Quote:

Originally Posted by celemine1Gig

Set all 4 register entries to 58 and see, as I said.

First thing I tried (back in the thread)

/wi1,70,15,58 /wi1,70,16,58 /wi1,70,17,58 /wi1,70,18,58

=1.346v
02-01-2009, 08:01 AM
celemine1Gig

So that's what I wanted to know. If idle voltage is set to 1.55V, you still get 1.35V, right? Your pics didn't show this config, so...
As I said without the exact datasheets and the card itself on hand, this is a hard task.
I hope someone with the right data on hand will be able to assist you further.
02-01-2009, 08:04 AM
fornowagain

Quote:

Originally Posted by celemine1Gig

If idle voltage is set to 1.55V, you still get 1.35V, right?

Yep.

Quote:

Originally Posted by fornowagain

http://img292.imageshack.us/img292/533/87865169yv3.jpg
02-01-2009, 09:51 AM
Demo

Why is the title even saying "ANY" when it only works on a very limited number of cards?
02-01-2009, 09:52 AM
chispy

I have tried this soft mod on both of my cards 4870x2 one its a Sapphire the other one its a Powercolor running on high end H2O EK Blocks both , not one single MHz gain as Jcool stated , it ran the same as before the soft mod , max clocks on the best card (powercolor) on stock voltages its 865/1000Mhz stable on 3D06 chilled water. The Sapphire does 855Max , tried both and again not a single mhz gain , yes i did it correctly ,i have been volt moding and using Riva for a very long time. Not worth the hassle if there is no gain :p:, i dont understand how people are seeing improvements when in my case i did not see any , like many guys already found out it does not help oc further 1Mhz on 4870x2cards , that conclusion its base on my extensive testing already done, so dont know whats up with this soft mod :shrug:.
02-01-2009, 11:23 AM
largon

Nothing wrong with the mod, it's just that RV770 has never been much of a voltage scaler. ~850MHz seems to be the max irregardless of voltage - until sub-zero cooling is involved. But even when supercooled, RV770 doesn't really go crazy on clocks.
02-01-2009, 01:25 PM
celemine1Gig

Quote:

Originally Posted by largon

Nothing wrong with the mod, it's just that RV770 has never been much of a voltage scaler. ~850MHz seems to be the max irregardless of voltage - until sub-zero cooling is involved. But even when supercooled, RV770 doesn't really go crazy on clocks.

That may be right, but the fact that the voltage seems to be somehow limited to 1.35V on the X2s is still strange. It's either a hardware constellation, like a voltage limit through diodes, but then again the current would increase with the higher voltage if it was. Or the registers are used a bit differently on these X2s. :shrug:
02-01-2009, 03:08 PM
Falkentyne

I asked this earlier but no one answered.
Were people doing the previous soldering resistor hardware mod also seeing issues with firefly forest in 3dmark not being happy with higher GPU speed (past what you could do at stock voltage) regardless of the hardware changed volts?
02-01-2009, 03:12 PM
saxile

Quote:

Originally Posted by chispy

I have tried this soft mod on both of my cards 4870x2 one its a Sapphire the other one its a Powercolor running on high end H2O EK Blocks both , not one single MHz gain as Jcool stated , it ran the same as before the soft mod , max clocks on the best card (powercolor) on stock voltages its 865/1000Mhz stable on 3D06 chilled water. The Sapphire does 855Max , tried both and again not a single mhz gain , yes i did it correctly ,i have been volt moding and using Riva for a very long time. Not worth the hassle if there is no gain :p:, i dont understand how people are seeing improvements when in my case i did not see any , like many guys already found out it does not help oc further 1Mhz on 4870x2cards , that conclusion its base on my extensive testing already done, so dont know whats up with this soft mod :shrug:.

I may not aid with 4870x2s, but with straight out 4870s, it helps massively, bumping up from 1.26 to 1.4 (via software, not sure on exact, don't feel like crawling under my desk right now, just had knee surgery) took me from a max stable of 800 to 825, so it helps, just not a massive amount.
02-01-2009, 03:40 PM
Radamanthys

Quote:

Originally Posted by Hardass

Running GTX 260 I can change voltage using this, but for some reason in RT Monitor my voltage regulator output never changes even when going to full load. Current increases but not output.???

I have too Gtx260 216 and in rivatuner show the vtt regulator on 1.06v , never changes with the mod :s :shrug:

any help?
02-01-2009, 05:25 PM
eva2000

Quote:

Originally Posted by chispy

I have tried this soft mod on both of my cards 4870x2 one its a Sapphire the other one its a Powercolor running on high end H2O EK Blocks both , not one single MHz gain as Jcool stated , it ran the same as before the soft mod , max clocks on the best card (powercolor) on stock voltages its 865/1000Mhz stable on 3D06 chilled water. The Sapphire does 855Max , tried both and again not a single mhz gain , yes i did it correctly ,i have been volt moding and using Riva for a very long time. Not worth the hassle if there is no gain :p:, i dont understand how people are seeing improvements when in my case i did not see any , like many guys already found out it does not help oc further 1Mhz on 4870x2cards , that conclusion its base on my extensive testing already done, so dont know whats up with this soft mod :shrug:.

darn and i was hoping for some improvements with my hd4870x2 heh .. so anyone actually report gains with hd4870x2 with this method ??
02-01-2009, 06:01 PM
NBF

I passed the link for this thread to some friends and one is having great results with 4870X2 after the mod, 880/960 on 3dmark06 and Vantage, card is WC.

I'll see if I can get him to post here.
02-01-2009, 10:48 PM
Unwinder

Quote:

Originally Posted by fornowagain

Again, the reason I included idle load current is to show that OCP is not occurring. Also the OCP/OVP register 13 is not changing state.

My bad, I had to add a note and explicitly say that provided VRM status register address was in decimal format. You were checking wrong register, correct command for checking VRM status is /ri<bus>,70,d
02-02-2009, 12:41 AM
scottc19

The card I have is hard modded 4870x2 and its stable at 900, Ive ran through vantage at 940 on water and I think I could make it to 950+. I guess I have a gem :P Though if it could have been done softmodded it would have saved a lot of trouble....
Keep up the good work with the softmod :)

Show 100 post(s) from this thread on one page