Highest voltage to use on register 18 ??

[jcusimano2]

I have msi 4870 reference 1Gb so how high can I go or can I go as high as I want as long as I can keep temp right??

[jcusimano2]

Welll I must be completely dumb I guess because I cannot get voltage to go higher. I have reference msi 4870 & basically I want to have my card at stock voltage clocks except my 3d which I guess will be clock info 3, 4-6, & 9 I do realize that 4-6 is for movies or whatever but I usually make it to whatever my 3d info box 3 & 9 are. So am Idoing this right??. I want 2d 400/1000 & 3D like 850/1000 would be nice but I cannot get any higher than stock volts & therefore cannot go higher than 800/925 in 3D. BTW I keep my mem clock same all the way through right now 925

So for box 1, 2, 7, 8 I would like volts to be 1.143 & clocks to be 400/1000 if possible & box 3, 4, 5, 6, 9 to be like well I don't knoiw yet but I was going to try 1.32 for now unless this is too stupid high on air but anyways someone school me a little bit here.

thanks

joe

[nafets]

I posted this in another thread, but it applies here too. Just substitute your 2D clocks and voltage (400/1000 @ 1.143v) and 3D clocks and voltage (850/1000 @ 1.32v).

Your UVD clocks (Clock Info #s 4,5, and 6) should all be set to your default 3D LOAD clocks and voltage. This is 750/900 @ 1.263v. If you don't do any hardware encoding/decoding, then you don't have to worry about what the UVD settings are.

Boot up clocks (Clock Info #0) is fairly useless, from what I can tell. Just leave it at 750/900 @ 1.263v.

2D IDLE clocks (Clock Info #s 1 and 2) should be set the same, at whatever GPU/MEM @ voltage you decide on.

3D LOAD clocks (Clock Info #3) should be set at 750/900 @ 1.263v.

Clock Info #s 7 and 8 can be set to either your 2D IDLE or 3D LOAD clocks. It really doesn't matter. Whatever you choose, both 7 and 8 must be the same.

Clock Info #9 should be set to 750/900 @ 1.263v. It should always be the same as what is set for Clock Info #3.

Phew. Hopefully the picture below helps;


Software overvolting doesn't usually help with overclocking your memory. If 925MHz is your highest stable, you probably won't get much higher. But it's certainly possible.

With stock cooling, 1.32v (or around that) should be fairly safe, as long as you keep your temperatures in check. You could probably go to 1.35v or a bit higher. More than that and you will be reaching diminishing gains (if any at all), with highly increased power usage and heat output.

850MHz is going to be very hard to reach with air cooling and software voltage adjustments, unless you received a "golden" HD4870. Somewhere around 815 to 820MHz would be more likely. Memory only running at 925MHz isn't a good sign, as the chips that come on HD4870s are spec'ed to run at 1000MHz. So you might just have a dud card. Good luck though...

[jcusimano2]

Ok so how do I get my 3D voltage to 1.32?? register 18?? & do I enter the same amount into the drop box # 3 & 9? or does the registers override the drop boxes?? well actuyally I can run applications/games at memory 1000 but if I try furmark then mem @ 1000 goes black

[nafets]

Whatever voltage you enter for register 18 is activated by selecting the 1.263v choice in the dropbox for each Clock Info. So you would enter 1.32v in the register 18 box, and choose 1.263v in the dropbox, for both Clock Info #s 3 and 9. The dropbox selections are "placeholders" (maybe not the word I'm looking for). If you chose a dropbox value of 1.083v for Clock Info #s 1 and 2, your card would run at whatever voltage is set in register 15; 1.143v would run at the voltage set in register 16, and so on.

Furmark stresses most every video card far beyond what you'd actually encounter during applications or games. It is not indicative of proper stability, but rather is a tool for excessive temperature testing. If you're testing for stability, try the applications or games you play/use for that purpose. Sure you can get your card stable enough to run Furmark, but that's fairly useless as you don't actually use Furmark for anything. If you can clock higher with stability in all of your other programs, then just forget about trying to get Furmark to play along.

[jcusimano2]

So are you positive that changing the voltage only changes the voltage to the core/gpu & not the memory?? If so then how do you up the voltage to your memory?? &why does it seem if I have a super low voltage in 2d mode if I put my memory to 1000 I automatically start artifacting in 2d but if I have a higher voltage in 2d mode then I can have a high memory clock & I don't artifact in 2d mode. This seems to me that changing the voltage also affects memory behavior.

[nafets]

Yes, adjusting the voltage by the method listed above only changes the VDDC (GPU) voltage. You can't change the memory voltage via software; only by hardmods.

Changing the VDDC voltage affects memory behavior because the two work in tandem, with the GPU processing memory operations (there's a better way to say/explain that). Certainly you could gain a bit of memory OC from overvolting your VDDC (GPU), but nothing like the gains from overvolting your MVDDC/Q (memory) directly by hardmods.

The HD4890 has two of the Volterra VRM controllers, one for the GPU and one for the memory. In time it should be possible to control the memory via software. But for the HD4870, it's not possible...

[BAGZZlash]

Originally Posted by nafets

The HD4890 has two of the Volterra VRM controllers, one for the GPU and one for the memory. In time it should be possible to control the memory via software.

Yes, and that's why I'd really like to put my hands on one of those...

[Fernandz]

Originally Posted by BAGZZlash

Yes, and that's why I'd really like to put my hands on one of those...

Hey BAGZZlash!!!

You are the master of universe!!!
The very first thing that i did when my 4890 arrives was editing her BIOS in RBE 1.20.

All is working fine, only not recognized in the model, we can change correctly the frequencies (the original bios has "almost" the same distribution as nafets say), we can edit too the voltage registers, with (in my case) the Register 16 for 2D and Register 18 ofr 3D. I changed the register 18 from 1.3125 (original) to 1.4125 (for 970 stable). i can't go to 1000 for now, even at 1.450.
We can extract and load the hash, i take the ASUS hash and put on Sapphire BIOS, only for testing (not work for SmartDoctor, but the ASUS original BIOS in a NOO too ). This is unnecesary now, but if we take some strange OC BIOS with differrent CCC limits, there will be interesting.
As you can imagine, the fan behavior values are working perfectly.

Then, we dream with a 1.21 version of RBE with 4890 model, and Volterra Registers for memory.... , apart of this, the 1.20 version is doing all GOOOOD.

Thanks for all the time and effort you put in this.

[BAGZZlash]

Originally Posted by Fernandz

Hey BAGZZlash!!!

You are the master of universe!!!
The very first thing that i did when my 4890 arrives was editing her BIOS in RBE 1.20.

All is working fine, only not recognized in the model, we can change correctly the frequencies (the original bios has "almost" the same distribution as nafets say), we can edit too the voltage registers, with (in my case) the Register 16 for 2D and Register 18 ofr 3D. I changed the register 18 from 1.3125 (original) to 1.4125 (for 970 stable). i can't go to 1000 for now, even at 1.450.
We can extract and load the hash, i take the ASUS hash and put on Sapphire BIOS, only for testing (not work for SmartDoctor, but the ASUS original BIOS in a NOO too ). This is unnecesary now, but if we take some strange OC BIOS with differrent CCC limits, there will be interesting.
As you can imagine, the fan behavior values are working perfectly.

Then, we dream with a 1.21 version of RBE with 4890 model, and Volterra Registers for memory.... , apart of this, the 1.20 version is doing all GOOOOD.

Thanks for all the time and effort you put in this.

Hi!

Thank you very much for the kudos. I really appreciate this. In fact, posts like these are what keeps me running.
Additionally, you seem to be an expert RBE user. I think, most people don't know and don't care about all the features I accumulated in RBE over the last months. So it's nice to see people using RBE to its full extent.

Thanks again. If I'd ever find the money for a 4890, I would love to take a look at that second VT1165MF...

Very kind regards

BAGZZlash

[Fernandz]

Originally Posted by BAGZZlash

...
Thanks again. If I'd ever find the money for a 4890, I would love to take a look at that second VT1165MF...

Very kind regards

BAGZZlash

Continuing with this topic...
There are some information around... Second VT1165MF isn't for the memory. The second volterra is also for some part of the core, and if we see the power cosumption in GPU-Z, is very constant, 2D, 3D, with or without load, the aperage remains the same. The voltage is 1.200V and Wizzard (thru GPUTool), is able to change this voltage (surely by registers).

If this voltage is for decoupling capacitors ring, surely is better to mantain the relationship between this voltage and the vcore. (it means, if we have 1.313/1.200 stock, with 1.400V we need 1.275, and for 1.450V:1.325V ond second volterra. This is not tested yet, but if exist ANY posibility to discover and port this register to the BIOS and add this function in RBE registers, it will be AWESOME.
And seeking around, the non-reference PCB's with pre-overclocked cards went with up to 1.500 Vcore. And no second regulator.... surely both domains have 1.5V or have a fixed voltage relationship.

Then if all have sense, will be a forum post of "Voltage Register 18/Voltage Register XX relationship for max OC".

Best regards, and as always, Thanks for your dedication!

[Fernandz]

For the topic: 1.5V seems to be perfectly safe for 1GHz core in HD 4890 (as new factory overclocked cars) no critic temperature rising in stock cooling with improved fan control in BIOS.

[largon]

The secondary VT1165MF serves "something" in the GPU die whose power consumption scales hand in hand with GDDR5 frequency...