Attention HD5970 owners

[jcool]

Update 2:

Believe it or not, my "problem card" has a cold bug at 35C positive (!!!). Different bios doesn't help either.. means I am pretty much screwed.

Again, it turns out there is NO issue with EK Waterblocks coolers, or any other water block for that matter. Just got a really, REALLY retarded graphics card. I recommend staying the hell away from Diamond brand graphics cards.

Update:

It seems adding the 2nd pad did not fix my freezing issues for the Diamond brand HD5970 card. Even with the fix and perfect contact it continues to freeze. Connected the stock fan with the card watercooled to rule fan speed out as the problem, no effect. Next up: Flashing a different bios to the card.. somebody with a working 5970 + WB PLEASE send me ypur VGA bios!

---------------------------------------------------------------------------------------------------------------------------

Anyone looking to watercool his HD5970, listen up. I have been fighting with a watercooled HD5970 for several days now, and couldn't figure out what was wrong. Problem was that the card would just hardlock the system when in 3D, 3DMark, Furmark etc. would just freeze to a halt and keep on displaying whatever scene there happened to be. For more details see here

I narrowed it down to a problem with the waterblock after endless testing with the card (which was working fine @ stock cooling), and together with Eddy from EK, I may have finally figured it out.

Here's a pic of the card with the block pulled:



Notice the smaller of the two CPL chips circled in red? For the sample design that guys like Eddy got from AMD, this CP-2-50 labeled chip was about 0,5mm higher than the longer CPLA-3-50 chip next to it.

Therefore, the EK block has a 0,5mm step at the location of this chip:



However, this is how the pad I used on those 2 CPLs looked like, upon close inspection:



Notice anything? There is a deep imprint from the longer CPL, but no sign of contact from the shorter one.

So, I went ahead and checked my card.. both CPLs are equally high on this Diamond brand HD5970, there is no 0,5mm step!

And this is not an isolated case: Today, I received a brand new Sapphire OC (10Mhz, lol) HD5970. I stripped it down and both CPLs are level on this card, too!

However, Eddy said that his retail Gigabyte card has the step. And BulldogPO just told me his card has the step as well.

So, what do you do if your card has level CPLs as well? The solution is fairly simple, at least for the EK block. Just add a 2nd thermal pad for the smaller chip to make up for the 0,5mm height difference. The pad that EK supplies for the PLX bridge chip is 0,5mm and big enough so you can "steal" a piece of it. Just add on top of the standard pad for the CP-2-50 chip and you're all set.

I'm finally running the card on water with no freezing issues

Note I will have to do more testing on this matter. I used the brand new Sapphire card with the "fix", so I cannot be 100% sure that this is what caused the problems in the first place. Might be some other problem with the Diamond card. Eddy will send me a sample block for testing with the Diamond card - unfortunately this rig needs to ship out tomorrow to a customer, so I simply don't have the time to bleed the loop again and re-test.

[nusihb]

Interesting finding.. this might have something to do what others (including me) has experienced with 'super hot' VRMs, discussed in this thread:
http://www.xtremesystems.org/forums/...d.php?t=240234

BTW - does anyone know what chip(s) is what in the pic above, as listed in GPU-Z (VDDC phase 1/2/3)?

[Waterlogged]

Anyone looking to watercool his HD5970, listen up. I have been fighting with a watercooled HD5970 for several days now, and couldn't figure out what was wrong. Problem was that the card would just hardlock the system when in 3D, 3DMark, Furmark etc. would just freeze to a halt and keep on displaying whatever scene there happened to be. For more details see here

I narrowed it down to a problem with the waterblock after endless testing with the card (which was working fine @ stock cooling), and together with Eddy from EK, I may have finally figured it out.

Here's a pic of the card with the block pulled:

[img]snip[/img]

Notice the smaller of the two CPL chips circled in red? For the sample design that guys like Eddy got from AMD, this CP-2-50 labeled chip was about 0,5mm higher than the longer CPLA-3-50 chip next to it.

Therefore, the EK block has a 0,5mm step at the location of this chip:

[img]snip[/img]

However, this is how the pad I used on those 2 CPLs looked like, upon close inspection:

[img]snip[/img]

Notice anything? There is a deep imprint from the longer CPL, but no sign of contact from the shorter one.

So, I went ahead and checked my card.. both CPLs are equally high on this Diamond brand HD5970, there is no 0,5mm step!

And this is not an isolated case: Today, I received a brand new Sapphire OC (10Mhz, lol) HD5970. I stripped it down and both CPLs are level on this card, too!

However, Eddy said that his retail Gigabyte card has the step. And BulldogPO just told me his card has the step as well.

So, what do you do if your card has level CPLs as well? The solution is fairly simple, at least for the EK block. Just add a 2nd thermal pad for the smaller chip to make up for the 0,5mm height difference. The pad that EK supplies for the PLX bridge chip is 0,5mm and big enough so you can "steal" a piece of it. Just add on top of the standard pad for the CP-2-50 chip and you're all set.

I'm finally running the card on water with no freezing issues

Note I will have to do more testing on this matter. I used the brand new Sapphire card with the "fix", so I cannot be 100% sure that this is what caused the problems in the first place. Might be some other problem with the Diamond card. Eddy will send me a sample block for testing with the Diamond card - unfortunately this rig needs to ship out tomorrow to a customer, so I simply don't have the time to bleed the loop again and re-test.

This just further reinforces what I posted yesterday.

http://www.xtremesystems.org/forums/...38&postcount=7

[Elpy]

This just further reinforces what I posted yesterday.

http://www.xtremesystems.org/forums/...38&postcount=7

But in this case wasnt it exactly the gigabyte card that has the step like AMD informed Eddy and which the block is designed for. So if you were to buy another card, like Sapphire for example... Then the block wouldnt be 100% compatible. Although of course then there is the question... Which one is the "real" reference model... With or without the step? I'd guess it was the one with the step if that's the information that AMD gave...?

But well personally if I would be buying a HD5970 now... I would probably wait until this is 100% clear.

[sargatanas]

gonna check my powercolors later on.
so far I had no issues at all.

[BlueAqua]

I have an EK block on a Diamond 5970 and haven't had any temperature issues, only driver related. I have it running at 1000Mhz Core and it's solid. I didn't notice if mine were level or offset though, but I don't remember one being taller.

Well, maybe I'll tear it apart and check it out. Thanks for the info Jcool.

[WaterFlex]

That`s very important finding for all 5970@water users. Good job mate!

[Diverge]

I bet is doesn't vary with the retailer who sticks their sticker on the cooler and fan, but more to do with the assembly of the cards that is varying from card to card. Since those components are surface mount, and probably have a pretty high tolerance (kinda needs to be for automated testing), I guess some just aren't soldered down correctly... maybe the solder didn't flow all the way and the part isn't properly seated on the PCB.

Of course I'm just speculating, since I haven't tore my 5970 apart to check mine out yet.

edit: hmm, I just looked up the datasheet for the part to see if they have tolerances for the package, and they only have a MAX for the height (4.80mm).

http://www.cooperbussmann.com/pdf/13...93a49c57f3.pdf

[jcool]

That`s very important finding for all 5970@water users. Good job mate!

Thanks, after pondering this issue for days and testing the card on 3 (!) different platforms to make sure it works properly, I don't want anybody else go through the same ordeal.

The sapphire card is rock stable with the "mod" by the way, highest I saw was ~60C across the VRMs with pretty warm water (~38C). It just completed a 6h bench run of the Unigine "Heaven" DX11 bench and is now running Milkyway @ home for the night.

The VRM temp readings are NOT from the CPLs by the way, but from the digital VRM itself - so you shouldn't see any temp increases there. Also the reason why temps seemed fine all the time

[sargatanas]

ok checked my 2 powercolors: same problem and placed some of the 0,5mm pad

[Diverge]

here's something odd too, here is another picture of the EK 5970 block. It doesn't seem to have that step.

[Crazystang01]

I just looked at the underside of my Nickel/Acrylic EK5970 block and mine did have the step in it. I had to look closely to it to see the step but it was there.

[Eddy_EK]

Hi, yeah we had a more than a hour skype chat and trying to find what was the problem.
We'll definitely do more research if this was the issue the first card did not work properly, as this is the first issue that came to our attention.

Right now we see the solution is easy and safe and it comes with the block.

[jcool]

Right now we see the solution is easy and safe and it comes with the block.

Exactly.

I'd like to thank Eddy for his outstanding support on this matter, as always

[Scaniris]

Well I didn't had issues with my XFX card using the EKWB FC-HD5970. But I had lockups issues because of the drivers... not because of temp issues (as far as I know... temps were pretty good in GPU-Z so I hope nothing was wrong on this side).

But what I can't understand: if all existing HD5970 on the market are based on the reference model, and if oem's only put a sticker on it... how is it possible to have height difference between cards that may be normally the same? High tolerance? But they also build a air cooler... how do they do to fit standard air cooler on cards if they are not exactly the same? I mean if some of you have problems with contact between this VRM and the waterblock, the same issue should normally happen with the stock cooler...yes/no?

[Scaniris]

Hmmmm... question: did you get some high temps into GPUz? Was it possible to see the temp issue? Or did you discover it only when removing the waterblock?

I'm asking this question because now I have a doubt... and I'm not sure that driver issue was the only provoking component... of lockups. I didn't noticed any temp issues, on logs in GPU-Z. But do you think I'll have to check? Because I swap my card with some nVidia and if there is any risk I want to warn the new owner of my card and ask him to check if contact is good or not... if required.

[nusihb]

I just took off my Koolance block to check this 'step' and guess what:
Koolance AR-597 does not even have a step... which doesn't really matter I guess, because my Powercolor 5970 doesn't have it, either.

Now to the fun part:
I swapped to a DangerDen DD-5970 because of VRM cooling problems - the DD block has a stepping in that area, but not just a mm or so, more like a 2-3 mm.
The instructions following this block does not call for any pad at all for the chip in the picture above!

I put in one anyways, but had to triple it before it evened out with the other 'non-stepped' part of the waterblock.

This DD block works fine for me though, with the VRM temps in the upper 70s when overvolting (1.25v/950/1200).
With the Koolance block I had towards 150C when doing that...

[Sadasius]

Good job you guys!

[Diverge]

I just took off my Koolance block to check this 'step' and guess what:
Koolance AR-597 does not even have a step... which doesn't really matter I guess, because my Powercolor 5970 doesn't have it, either.

Now to the fun part:
I swapped to a DangerDen DD-5970 because of VRM cooling problems - the DD block has a stepping in that area, but not just a mm or so, more like a 2-3 mm.
The instructions following this block does not call for any pad at all for the chip in the picture above!

I put in one anyways, but had to triple it before it evened out with the other 'non-stepped' part of the waterblock.

This DD block works fine for me though, with the VRM temps in the upper 70s when overvolting (1.25v/950/1200).
With the Koolance block I had towards 150C when doing that...

Same with the DD block that Evilclocker gave me. I haven't put it on my card yet, or checked to see if my card's coils are flush. Just today, I ordered thermal pads for it, good thing I ordered a lot heh

[Kurz]

Hmmmm... question: did you get some high temps into GPUz? Was it possible to see the temp issue? Or did you discover it only when removing the waterblock?

I'm asking this question because now I have a doubt... and I'm not sure that driver issue was the only provoking component... of lockups. I didn't noticed any temp issues, on logs in GPU-Z. But do you think I'll have to check? Because I swap my card with some nVidia and if there is any risk I want to warn the new owner of my card and ask him to check if contact is good or not... if required.

From How I understand it that specific part doesn't output temps.
So you wouldn't know from GPUZ.

Just pry her open and find if it made and Indention.

[Romir]

I'm not sure that driver issue was the only provoking component... of lockups. I didn't noticed any temp issues, on logs in GPU-Z. But do you think I'll have to check? Because I swap my card with some nVidia and if there is any risk I want to warn the new owner of my card and ask him to check if contact is good or not... if required.

On my XFX 5970 both inductors are the same height meaning the small one wasn't making contact with the block.

Sapphire's parent company PC Partners reportedly makes the vast majority of ATI partner cards so Sapphire's card having similar height inductors means most probably do.

[Waterlogged]

I was given the impression some years ago (by some articles published at the time) when Sapphire showed up that they were to be to ATi like eVGA is to nVidia. . .the unofficially official "retail arm".

But in this case wasnt it exactly the gigabyte card that has the step like AMD informed Eddy and which the block is designed for. So if you were to buy another card, like Sapphire for example... Then the block wouldnt be 100% compatible. Although of course then there is the question... Which one is the "real" reference model... With or without the step? I'd guess it was the one with the step if that's the information that AMD gave...?

But well personally if I would be buying a HD5970 now... I would probably wait until this is 100% clear.

This is the real mystery. I think we'd need to ask Eddy for clarification on whether the block was designed around the GB or actual specs gotten from AMD/ATi. . .Eddy, can you share this info please?

I myself got burned by Gigabyte's lack of respect for ref designs on a 7600 GT. Every other card I looked at had the same GPU hole pattern as the older 6800/7800 cards yet the GB I bought had a smaller hole pattern as well as a lot of the caps and FET's in different places. Since then, I've very carefully looked at GB's cards and compared to others and they very seldom match.

[Scaniris]

OMG! This means that my lockups issues are maybe because of the BAD CONTACT WITH EKWB WATERBLOCK

I'll immediately warn the new owner of my card, and ask him to check this and add another 0,5mm thermal pad on this component if he discovers that the contact is not good.

That may explain why I did not have troubles with the stock cooler Lockups issue happened AFTER block installation And thought that was only a driver issue...

BAd new, really bad new. And hopefully I'm a XS forum member, thank you very much gentlemen for sharing these information with us. Let's imagine what would happen if we are not informed about this king of issue Because of a waterblock not fitting with the card perfectly, we may have broken this expensive card....

[jcool]

Hmmmm... question: did you get some high temps into GPUz? Was it possible to see the temp issue? Or did you discover it only when removing the waterblock?

I'm asking this question because now I have a doubt... and I'm not sure that driver issue was the only provoking component... of lockups. I didn't noticed any temp issues, on logs in GPU-Z. But do you think I'll have to check? Because I swap my card with some nVidia and if there is any risk I want to warn the new owner of my card and ask him to check if contact is good or not... if required.

See here:

The VRM temp readings are NOT from the CPLs by the way, but from the digital VRM itself - so you shouldn't see any temp increases there. Also the reason why temps seemed fine all the time

So the answer is no, you do not see any high temps, it just locks up.

This screenshot was taken from one of the first block installations, where the card would lock up after a few minutes running milkyway - temps are fine, as you can see.

And this is not an isolated issue with EK waterblocks.. Eddy designed the cooler after a sample card he got from AMD, so I'd say this is yet another in the long line of HD5000 series screwups

[Scaniris]

Ok... I had a look in my photo collection, and here is the card I have from XFX:



Seems that the VRM is higher than the other.

And having a look on the stock rad:



At this place, there is a open space, so probably they are cooled with the fan, not with direct contact with radiator, and if there is some height difference.... it doesn't matter with the stock rad.

But I want to make sure everything is fine and ask the new owner of my card to check this if he has some lockups. I'm not sure he'll have because since 9.12 update... I did not experienced lockups anymore but I was testing only 1 day before sending the card...