5870 and 5970 voltmods guide

[chew*]

VGPU 1 solder and read points for 5970 = almost identical to 5870

[BulldogPO]

Looks very easy mod to me, except that cap mod

[Diverge]

The soldering on the 5970 looks HORRIBLE... Did you use the iron from a wood burning kit or something?

[hiwa]

5870 caps mod

471 6.3V

[madmax999]

Nice hiwa
2 bad that memory on your HD5870 sucks

[hiwa]

Nice hiwa
2 bad that memory on your HD5870 sucks

Samsung memories are damn fragile and they die easy
no need OC memories higher than 1250/1300 i think, higher GPU core is better

[madmax999]

Yeah, i agree. We had big problems with memory on our HD5870.
How much MHz more did you get on vcore with CAP mod?

[zalbard]

How much MHz more did you get on vcore with CAP mod?

I'd be curious to know as well. And no point for this mod for air / water, or is there?

[hiwa]

Yeah, i agree. We had big problems with memory on our HD5870.
How much MHz more did you get on vcore with CAP mod?

air 1070 1.32
chiller 0°c 1150 1.45
ln2 :
-40 1200 1.41
-80 1300 1.45
-100/-110 1350 1.46

[kayl]

does anyone know the vmod for memory for the 5870's, does it help much with clocks?

[rym]

I tried rbe with my memory v to 1.175 from 1.15 i cant get past 1300 . will the caps mod help and whats highest memory v thats safe ?

my cooling of 5870 here

[zalbard]

You cannot adjust memory voltage with software.

[BenchZowner]

I tried rbe with my memory v to 1.175 from 1.15 i cant get past 1300 . will the caps mod help and whats highest memory v thats safe ?

The voltage RBE is referencing as Memory Voltage isn't correct, in fact it's another voltage, which AMD calls Vddci.

[rym]

5870 memory volt guide ?

[BenchZowner]

I don't to sound pessimistic, but I don't think we'll see any benefits from altering ( raising or lowering ) the memory's voltage.
I need to try it on a few cards first though, so take this with a pinch of salt.

[Barr3l Rid3r]

Sadly I couldn't realize OCP/OVP and vGPU for 5850, any possible light on it my friends?

[BenchZowner]

I tested a few 5870's up to 1.65V or so, and didn't get any OVP.
OCP was needed for most of the cards to get 1300MHz pass 3D Mark03, Vantage & AquaMark 3.

OCP mod should be the same for the 5850 ( I'll try to check it out later to confirm that )

[sno.lcn]

So any vmem yet?

[BenchZowner]

No, I didn't confirm it yet, neither talked with George ( Hipro5 ).
I had lots of things to catch up with.

From what I gathered from a phone call though isn't good news, upping or lowering the voltage did nothing for most of the reference cards a few peeps tried and the ones that showed some gains were just 20MHz or less.
Don't think it's worth it at all if that's the case with every reference card with the samsung chips out there.

I'll try to confirm the mods asap and post them unless somebody delivers before I do ( it would be greatly appreciated, I'm working 20 hours per day the last 3 days and that seems to be the case for the whole week at least )

[CL3P20]

The voltage RBE is referencing as Memory Voltage isn't correct, in fact it's another voltage, which AMD calls Vddci.

Correct... and vDDCI is what seems to be helping up to 1300-1400mhz for mem clocks though.

vDDCI runs off PCI-E slot voltage supply... and not PCI-E power connection... so volts are limited unless you hardmod vDDCI with another power source.. setting 1.4v + LN could bring much faster mem clocks.

*I dont think anyone has tried that mod yet.. might be worth a shot if someone wants to crack 1.5ghz for vmem. Since vDDCI is voltage for I/O on MCH... we could consider this like 'uncore'..?


**vDDCI mod for 5770 could yield WR too... as most are mem clock limited over core anyhow.

[BenchZowner]

Did you try bumping Vddci on LN2 ?
Because most of the people I spoke with experienced a big GPU Core overclock drop with high Vddci under LN2.

[CL3P20]

Did you try bumping Vddci on LN2 ?
Because most of the people I spoke with experienced a big GPU Core overclock drop with high Vddci under LN2.

i have not... but my question is-> Is the experienced drop from the high voltage itself... or is it from the voltage being drawn through the PCI-E slot is now 'dirty' at the higher level, and causing stability issues...?

Since it seems that raising vDDCI is needed for highest possible mem clocks, it just doesnt seem logical to me, that adding a touch more, would send the OC to the crapper...

[BenchZowner]

Well it's a 50-50 scenario.
It could be the case though, just like the Core i7's do with higher UnCore voltage.
We still don't have a clear image of what the Vddci voltage is exactly.

[CL3P20]

From Garret Wu:

VDDCI is voltage of GPU I/O (memory controller & bus voltage between GPU <-> Memory)
VDDCI on R5870 Lightning is using PCI-E bus power, so DO NOT USE TOO HIGH VOLTAGE (1.35V is recommended highest vaule)

*1.6v is doable for vDDCI as well [according to Garret]... unsure of its benefits though

[Stelios]

I have modded vddci with little gains . Optimal value is around 1.20-1.22v . It helps RAM clocks a bit .

The first 5870 i modded died cause of vmem . Vmem mod is very tricky and difficult to stabilize . With a small turn on trimmer , the voltage scales very fast and impossible to control it .
It probably needs a fixed resistor and then a small trimmer .