More results with ASUS GTX580 DirectCU II

[G H Z]

Did you restart the rig after the crack? Or was all this during the same windows session? I thought the pot you guys used was copper and aluminum?

[SoF]

just did a little edit you maybe haven't read yet
we used like 3 or 4 different pot designs on that card - that was when stummerwinter decided to build his own new pot. the "fill in" part is alu but the base is CU. that crack definitily does not come from it due to the MHZ-issues I explained some minutes earlier.
So we can rule out that this crack comes from the container itself. It must be something related to HS / DIE or HS / Pot.
I think marcus has a hint which paste works best but I can't remember atm (sitting in a hotel far away from home after a "monstrous" set of weizen )

[G.Foyle]

must be something related to HS / DIE or HS / Pot

There's an easy solution

[Splave]

No definitly not - we are talking about full CU containers here. but as the hs is ALU (???) and container is CU - maybe there's your problem? do we need to build ALU containers then? Now that would be

After that "crack" the card won't run 1300 anymore (after passing 1600+ 03 GT3 &4 before), maybe 1200 when you are lucky. after warm-up and remount --> 1600+ again...I can tell you that was PITA during cebit ^^

@oanvoanc I don't think that's OT - that's one of the basics you need to know about CUII and it would be my pleasure to speak about that in public and togehter with thoose guys who maybe have the most experience with CU II. If more ppl. think this is not of major interest, ok I'll go pm then

I use Freeze and never get the crack
Although there seems to be different batches of freeze some is as dark as a pencil lead and other is nearly white.

[ME4ME]

Hey SoF,

The issue you discribe isnt really new. Me and elmor experienced the same when pretesting 480 Lightning before MOA 2010 finals. We're using OCZ Freeze thermalpaste. It's conductivity seems great, until around -160 when it suddently freezes and cracks. Here's some data about that: link

one workaround would be to contestly bench and not stay in idle for long, so even with the VGA pot filled the GPU itself will not get any colder than -160. Or when things crash and you may need to reboot and such ... always keep high voltage applied. VID mod with +400 like elmor and kinc posted should work great i would imagine. At the time when running Lightning, we didnt do any such mods, and had do keep stuff warm with heatguns.

So when hearing that crack, you'd definitly have to back off and let the card and thermalpaste heat up a bit. Then things should be just fine

[SoF]

I use Freeze and never get the crack
Although there seems to be different batches of freeze some is as dark as a pencil lead and other is nearly white.

There's an easy solution

That was what I wanted to do on cebit after remounting for the 4th time but christian from asus hold me back

Hey SoF,

The issue you discribe isnt really new. Me and elmor experienced the same when pretesting 480 Lightning before MOA 2010 finals. We're using OCZ Freeze thermalpaste. It's conductivity seems great, until around -160 when it suddently freezes and cracks. Here's some data about that: link

one workaround would be to contestly bench and not stay in idle for long, so even with the VGA pot filled the GPU itself will not get any colder than -160. Or when things crash and you may need to reboot and such ... always keep high voltage applied. VID mod with +400 like elmor and kinc posted should work great i would imagine. At the time when running Lightning, we didnt do any such mods, and had do keep stuff warm with heatguns.

So when hearing that crack, you'd definitly have to back off and let the card and thermalpaste heat up a bit. Then things should be just fine

when I was on the first run on thursday with stuwi we cooled the card down to ~-100 and let it run GT 3 and GT4. then we went down to ~-150°C and tested GT3/GT4 another time. Then we wanted to go full fire - 1625 on the core and full pot...GT1 passed, GT2 passed, GT3...middle.....




....CRACK!!!!



...reboot, 1200 ok, 1300 fail...


...and it happened another 3 or 4 times we got that crack not in idle but under load. I had the feeling that ultra-hot load on the one side and ultra-cold on the other side just couldn't work together well
At some point I gave up and wanted to throw everything in the corner as that problem really cranked me up ^^ I have one 580 CU II at home and one sended to rma today - before making them cold again I need to figure out more about this issue...

[SF3D]

We have solution for thermal issues:



Indium foil does not resist heat at all. It is soft enough and does not care about cold.
We have tested it many times with AMD Phenoms in the past, so it is not a new thing for us.

In this form 1kg of indium would cost 100 000€, but luckily you can use it multiple times.

My card can run Unigine heaven around 1420MHz, so let's see does indium help anything.
I hope to test it on upcoming weekend.

E: forgot. Sick score in 3dm11 Hipro5!

[Splave]

^ancient finnish secret?

[SF3D]

Hehe..don't know about ancient or secret We used it first time in 2009, but because it is very expensive I have not been thinking of it.

I just like to try it and see how it does. At least there is nothing to clean after test session.

[G H Z]

If I remember correctly, macci used that a few years back. Mike had some $150 thermal grease @ GOOC last time, he swore by it.

[SoF]

1st: perfect print of the pot - good mounting!
2nd: good info
3rd: Indium foil? Doesn't sound common to me - any more info about these things? I just know "liquid metal paste" some time ago in germany, but is it the same?

[SF3D]

3rd: Indium foil? Doesn't sound common to me - any more info about these things? I just know "liquid metal paste" some time ago in germany, but is it the same?

Indium: http://en.wikipedia.org/wiki/Indium

Very rare soft metal. The foil is 99,9% pure indium, so there is practically no heat recistance.

Thermal conductivity comparison:
OCZ Freeze = 3.8 W/m-K
Titan Nano = 4.5 W/m-K
Indium metal (+foil) = 81.8 W/m-K

Liquid metal pastes are Gallium + Indium combination, where gallium is the main part. Gallium is slightly toxic and it does expand when it is getting colder. So it can not be used with ln2.

[KanGaXx]

i found great improvments with ICD7....with AF7 i'm stuck 100MHz down respect to ICD7.

where can you buy Indium ?

[aristidis]

We have a lot o issues with our card also. We have to heat the pcb with two heat guns all the time. At the end we have manage to run 1420/1330 for 3DMark11 and take world record with 10280 points.

[KanGaXx]

@aristidis : what thermal paste ?

[aristidis]

Titan nano but my opinion is that we have to use a pot with some distance from PCB...

[Dualist]

where can you buy Indium ?

Here..
http://www.cmr-direct.com/product_in...roducts_id=683

or..
http://cgi.ebay.co.uk/INDIUM-99-99-F...item35af0d34b0


They were the first hits I had just with Google

[SoF]

Indium: http://en.wikipedia.org/wiki/Indium

Very rare soft metal. The foil is 99,9% pure indium, so there is practically no heat recistance.

Thermal conductivity comparison:
OCZ Freeze = 3.8 W/m-K
Titan Nano = 4.5 W/m-K
Indium metal (+foil) = 81.8 W/m-K

Liquid metal pastes are Gallium + Indium combination, where gallium is the main part. Gallium is slightly toxic and it does expand when it is getting colder. So it can not be used with ln2.

Need to tell stuwi - gotta prepare a week-end with him, 580 GTX CUII and some indium stuph thanks man for sharing!

[SF3D]

Just check that you have good contact. It is not as easy with indium as it is with thermal pastes.

[chew*]

The proper way to use indium would be to heat the copper to the melting point of the indium and solder it and the pot to the IHS of the card........

Using in the solid form that you are using it in won't work any mircacles.

Also the conductivity charts you are showing are at like normal temps.

Indium actually sucks conductivity wise at like say -192........

Also last but not least the point of using it is direct to core and not to an IHS which then has paste........the whole point is to eliminate all the thermal barriers.....in a perfect world you would solder the POT direct to the die of the GPU.

First to manage soldering the pot direct to die of GPU without killing it gets a cookie.

[chew*]

Double post but that crack you guys are hearing is the bond snapping from the paste you guys are using.

The heat generated by GPU and the cold generated and the large shift in temp delta is enough to snap the bond.

I know this becasue i bench at -192 90% of the time, Been through various pastes to find pastes that actually don't freeze which in turn can not crack the bond....

Sounds like some serious heat dump.......takes a 6 core thuban in wprime at 6.3+ to create those circumstances.....

after the crack any loading of the gpu sounds like something boiling under pot doesn't it

You can verify this, just unbolt pot cold real quick after the crack, don't let it defrost, it will pop right off and not be "stuck" to gpu anymore.

[rbuass]

Feel sad with mine GTX 580 CUII
I modded one and reached 1400/1210 Mhz 3DM06 with just 10 LN2 liters in the first day...but the card was not mine (i modded to a friend)...
Yesterday i full mod mine...
VID mod 0.4 Volts
VRAM, VGPU and VPLL working fine (each mod with a switcher on/off)...
CBB Mod (joined the 3 points)
OCP Overall and OCP per Phase
Did try:
VID 0,4 + Smart Doctor (1,0 to 1.18V)
VID 0,4 + VMod (same)
But no one settings works fine...
Only way worked was only VID 0,4 + stock Voltage - about 1220 Mhz vantage (sucks...worst than the old GTX 580 reference)
Ram...the worst..
Did try all voltages - 1.74 to 2.1V and can t reach more than 1200 Mhz

PLL did try on (trimpot - 1.2 ... 1.35V)...and off stock Voltage... no change

Lots of issues...the card can run ok and no coldbug...but i can't add GPU voltage (Smartdoctor or trimpot) that the system crashes at the middle of benchmark...
I don t know why the card is not good....but can t understand the videocard can pass 1.4V (down to 1.35 in the middle of benchmark) 1220 / 1200 Mhz...but appear to be the limit of the card...
The Videocard can't go more than 1220 mhz with any voltage...
I think MINE VGA sucks...no lucky with this one

[SoF]

something can't be right there - EVERY CUII can do 1400+ easy, there must be a mod maybe not sitting correctly or whatever but it is impossible that a CUII can't pass 1300...(except me beeing the one benching it with a multi-rail PSU during cebit )
What PSU are you using??

[steponz]

I think there is an issue with some cards... I was using a corsair 1200, single rail... Not the issue... Used multiple tims...
My card would hate voltage, anything over 1.34 volts would fail @ higher then 1200..
Did hit 1297 2 times @ 1.34..

No matter what high voltage bombed on my card...

[Dumo]

No matter what high voltage bombed on my card...

Which s/n batch your cards? B1XXXX or B2XXXX