ATI 5850/5870 voltmod thread

[T_M]

George has done that type of thing before successfully

@Jimbo - try holding a block of dry ice against your radiator, this should be capable of bringing your core temps down to 10C or so pretty easily to check what the roadblock is

[EvoCarlos]

heres what i had in mind
an old vapo chill heater fitted bettwen the pot and the card

[CL3P20]

I have a Sapphire 5830 that I hardmodded (vGPU only) but the core runs out of steam unusually quickly and more voltage doesn't help.

My fear is that it's simply hitting a wall and there's nothing I can do about it, but I want to explore the other possibilities (e.g. OCP, VR cooling etc.) Here's the details:

On stock voltage (~1.15V) and cooling, the core can do 1030Mhz.

vGPU mod (~1.25V) and watercooling (core only, 38'C full load), it rapidly hits a wall at around 1150Mhz (i.e. system locks up / screen goes blank).

I've taken the voltage to 1.35V with no improvement.

Having checked the 58xx overclocking thread, it does seem that most 5850s and 5870s also crap out at about 1100Mhz until using very high voltages and extreme (sub-ambient) cooling, reinforcing my feeling that this is a chip limitation as opposed to cooling / OCP issue.

Any thoughts?

In addition, has anyone found any improvement from vMem mod and/or memory cooling? (I can't find any info on this yet - still looking).

Thanks.

*sounds like voltage stability..

Use GPU-Z to monitor vDDCI and OCP% when running a graphics stress test like furmark.. chances are with such a fast core speed, you are max on input capacitance and causing voltage to drop too low on load. I noticed the same thing on my 5770, and had to add 16v caps to keep the input from dipping below 11.9v on load. Now I can run 1105mhz stable for daily gaming and bench at 1120mhz.



@ EvoCarlos- vapo chill heater..nice..and yeah, almost the same thing.

[EvoCarlos]

@ EvoCarlos- vapo chill heater..nice..and yeah, almost the same thing.

do you think it will work ?


it gets walm to the touch at 3v and its rated @ 5w 12v

[EvoCarlos]

oh dear!!
i did not even get past the 1st post screen


any option to work around this problem ?

[hiwa]

ouch

[Jimbo Mahoney]

*sounds like voltage stability..

Use GPU-Z to monitor vDDCI and OCP% when running a graphics stress test like furmark.. chances are with such a fast core speed, you are max on input capacitance and causing voltage to drop too low on load. I noticed the same thing on my 5770, and had to add 16v caps to keep the input from dipping below 11.9v on load. Now I can run 1105mhz stable for daily gaming and bench at 1120mhz.

Thanks for the suggestion.

GPU-Z reports rock-solid VDDC (1.165 V) and OCP (0.0 %) both before and during MSI Kombustor in DX11 mode using "Unlock Power Draw" mode.

Other things I've tried are finding any hot components on the board after reading about the other 58xx series getting very hot fets and VRs (apologies if I'm using the incorrect terminology - I'm a novice @ electronics). However, not a single chip on the board is hot to the touch apart from the memory chips. The whole thing runs very cool and I'm only pointing a couple of 5v 120mm fans at it.

Unless anyone has any other ideas, I'm ready to admit defeat on the core (sadly). I'm going to get a mechanical guy at work to chop me up some sinks for the memory and see if that improves the headroom on those, as they do get pretty hot (too hot to hold a finger on). I've not had terribly good experiences in the past with heatsinks on graphics card memory though, so I'm not expecting much.

[CL3P20]

@ EvoCarlos- ouch!! Too much voltage or what? Accidental short to ground?

[EvoCarlos]

@ EvoCarlos- ouch!! Too much voltage or what? Accidental short to ground?

it went in to windows then when it was done loading it give me green and white stirpes then it restarted and got stuck at 2b on the post code display

i had checked all the voltages
vgpu = 1.15v
vmem1 = 1.62v
vmem2 = 1.62v
12v+ = 12.03v
and the 2nd voltera chip was 1.25v

not sure what happend it was all coverd in putty and at 25c

im sure rma will fix it

[Jimbo Mahoney]

Thanks for the suggestion.

GPU-Z reports rock-solid VDDC (1.165 V) and OCP (0.0 %) both before and during MSI Kombustor in DX11 mode using "Unlock Power Draw" mode.

Other things I've tried are finding any hot components on the board after reading about the other 58xx series getting very hot fets and VRs (apologies if I'm using the incorrect terminology - I'm a novice @ electronics). However, not a single chip on the board is hot to the touch apart from the memory chips. The whole thing runs very cool and I'm only pointing a couple of 5v 120mm fans at it.

Unless anyone has any other ideas, I'm ready to admit defeat on the core (sadly). I'm going to get a mechanical guy at work to chop me up some sinks for the memory and see if that improves the headroom on those, as they do get pretty hot (too hot to hold a finger on). I've not had terribly good experiences in the past with heatsinks on graphics card memory though, so I'm not expecting much.

Actually, I think it could be voltage stability.

I found the following:

1.25 V = 1126 Mhz before strange behaviour
1.30 V = 1144 Mhz
1.35 V = 1152 Mhz

It's weird to have such tiny improvements from voltage increase and I think the voltage could be unstable, although it's hard to separate actual core voltage fluctuations from variable contact with the pin on the graphics card (I'm just holding a Multimeter probe against it with some difficulty, as opposed to soldering a nicer readout point etc.).

I also bought some clearance heatsinks for the memory and they make *slight* improvements - I can bench at 1300 Mhz instead of 1250 Mhz Wow, an extra 50 Mhz - pretty much what I expected from previous experience

[db87]

Is there still no vmem mod (hardmod) for the reference HD5850?

[Dahaka]

Guys, i'm running my reference Sapphire 5870 on 1080 - 1350 (or 1400) MHz on GPU and Mem, but i set one mhz more on GPU and 3dmark's crashes instantly.

1.35v (set on MSI AB) max temp 37º on WC Swiftech MCW80.

Any suggestion? i was thinking that maybe it's a problem with OCP or OVP because vGPU goes to 1.16v when 3DMark crashes.

thanks in advance.