4890 highest core clock

[largon]

I'm not going to do that.

 

[eidairaman1]

 

[purecain]

@largon... try replacing the caps.... cant believe it just died like that....

i think i'll wait a while on modding mine... not feeling as brave now....

bad luck m8...

 

[muon]

Any of the reference 4890s can get software voltage control

Other than flashing on an Asus bios, how? That voltage software posted earlier using v1.10 and rivatuner v2.24 doesnt work. Followed the direction in the text file accomanied with the software to the letter.

edit: nvm, got it to work. Needed the guru3d .cfg file.

 

[theorw]

I'm not going to do that.

WHYYYYYYYYYYYYYYYYYYYYY????
And another thing?How did it happen ? ESD harms when the card is IN OPERATION RIGHT?

 

[eidairaman1]

WHYYYYYYYYYYYYYYYYYYYYY????
And another thing?How did it happen ? ESD harms when the card is IN OPERATION RIGHT?

ESD can affect anything whether it's operating or not.

 

[SonDa5]

Surprised at how cool this card is. Much more cooler than my Toxic HD4850 and VisionTek HD4850.

A quick view into my system with the Diamond OC edition 4890.

 

[eidairaman1]

well its a reworked card afterall

 

[largon]

WHYYYYYYYYYYYYYYYYYYYYY????

Because I killed it while tampering with it. My mistake. My loss.

How did it happen ?

Anyways, I found the culprit - or rather, the part I destroyed.

Volterra VT242WF. It's a memory vDD or vDDQ converter.
Impossible to replace.

The edge of the part is ~2.75mm so the chipped corner is positively tiny. But apparently enough to kill it.

The brown spongy thing is the head of a match, for scale.

One would think the thermal pads I used would protect the buggers against chipping...

 

[purecain]

omg m8.... that is really unfortunate.... looks like you might have a new keychain....

thanks for the info....

 

[largon]

*mad scientist laughter*

The chipped VT242WF is dead. That's a given, as it's missing a piece. I confirmed this by once again plugging in the card and probing the poor thing with a DMM. The other memory voltage read around 1.52v which is ok for GDDR5, but the other was 0.036v. Now that's a problem, and there's little to be done about it. I certainly can't fix or replace the broken VT242WF as it's a BGA part with solder balls under it. I can't practically rebuild the phase. But the thing is, there are two VT242WFs on a reference HD4890. One produces memory vDD and the other vDDQ. And since these voltages are very close to each... See where this is heading?

I figured I have nothing to lose, right? The card doesn't even boot as it is. I decided to perform a surgical... erm... a soldered "vDD/vDDQ bypass".

Here's what I did:
I desoldered and lifted the anterior leg of the inductor of the broken memory phase, effectively cutting off the broken VT242WF from it's inductor and as such, it's load. Then I simply bridged the two inductors by their anterior legs, meaning both memory vDD and vDDQ are now fed by the same converter. The broken VT242WF has no load anymore. Like it didn't exist anymore.

Installed the HSF.
Plugged it in.
Plugged both 6pin plugs and DVI.
System power on.
Monitor status led was yellow, meaning no signal. But that's normal It will take about 5 seconds for the VGA to give a picture.

Seconds feel unusually long...

Dissapointment creeps below my skin.

But what's this?
Monitor status: BLUE LED!
There's a bloody video signal!
Quarter of a second later: POST screen! Sweet, sweet POST screen!

I have no idea how this hackjob will affect the card in practice. I reckon memory won't OC much after this. It's also quite possible that it wouldn't even be 3D stable... Dunno yet, because just that things wouldn't be too easy the OS on my Vertex SSD got corrupt earlier today by some weird twist of fate for reasons which are beyond me.

So, it'll be some time to run some 3D on the card...

 

[SonDa5]

You got skills largon.



Congrats on bringing back your 4890 from the dead.

 

[a_ump]

LARGON your the fucking MAN!!! lol that's some nice thinkin and modding dude, crazy shits i'd have never thought of. hope it works for ya that'd be tight.

 

[largon]

It was stable for atleast an hour long L4D session at stock 850/975. There were some minor memory related artifacts, though. I guess I'm pretty much SOL for what comes to memory OC'ing...

I guess I need to start planning a way to rebuild the bypassed phase...

 

[_jM]

try and sell it (cheap of course) and save some cash and buy another ?

 

[largon]

Sell? Of course not.
This way it's more fun. I already got core to 1GHz stable before the mishap and more voltage didn't gain any more (1040MHz -> reset) so it was starting to get boring. Now that the card is crippled, suddenly there's something to gain.

A new card would only get me to the point I started.

edit:
I don't have a spare VT242WF but I do have several similar Volterra "digital PWM" parts like VT232TF, VT233TF and four VT1103SFs and their complete accompanied circuitry and parts. I think I can use them to make an offboard phase to replace the broken one...

 

[a_ump]

Sell? Of course not.
This way it's more fun. I already got core to 1GHz stable before the mishap and more voltage didn't gain any more (1040MHz -> reset) so it was starting to get boring. Now that the card is crippled, suddenly there's something to gain.

A new card would only get me to the point I started.

edit:
I don't have a spare VT242WF but I do have several similar Volterra "digital PWM" parts like VT232TF, VT233TF and four VT1103SFs and their complete accompanied circuitry and parts. I think I can use them to make an offboard phase to replace the broken one...

if you start a mod like that, you should def make your own thread. would be a very nice project i'm sure everyone would love to watch and i'd check up on your progress everyday dude, that sounds awesome and if you can get it to work, well that'd just be dam amazing. New thread!!!

 

[Ketxxx]

Poor 4890.. bad bad largon. Offboard doesn't count though, you have to route your way around things on the PCB just because it'll be way cooler

 

[Scrizz]

very bad @$$ largon
nice job m8

 

[purecain]

amazing m8..... get busy....

 

[BooStFeD]

I'm trying to figure out a decent overclock stability testing routine. Just passed a 20 minute loop of 3D06 @ 950/1100 stock vgpu.

Would anyone one agree this a stable speed for the card or would a 20 minute loop of Crysis benchmark be more of a stability tester?

 

[Scrizz]

I'm trying to figure out a decent overclock stability testing routine. Just passed a 20 minute loop of 3D06 @ 950/1100 stock vgpu.

Would anyone one agree this a stable speed for the card or would a 20 minute loop of Crysis benchmark be more of a stability tester?

load up furmark on that sucker

 

[BooStFeD]

load up furmark on that sucker

Let FurMark run for 20 minutes??

So after 20 minutes and FurMark does not crash the card will be stable?

Card might also be on fire too

 

[a_ump]

it's more like run it for 20 min and if you see no artifacts(white dots usually) then it's stable. it doesn't crash usually.

 

[purecain]

DONT run furmark for that amount of time..... your vrm temps will be dangerously high....

i dont know of any other application that heats up your card so badly....

use a game like crysis to test stability....not an app wich will shorten the lifespan of your card....